LIBRARY
Michigan State
University

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DateDuesz-p. 15

 

INTERPRETING THE ASIAN CURRENCY CRISIS:
EMPIRICAL ANALYSIS AND PREDICTION.
By

Hoon Kim

A DISSERTATION

Submitted to
Michigan State University
In partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Economics

2002

 

ABSTRACT
INTERPRETING THE ASIAN CURRENCY CRISIS:
EMPIRICAL ANALYSIS AND PREDICTION
By

Hoon Kim

This dissertation investigates the causes of the Asian crisis and improves
implementation in predicting actual currency crises. Chapter 11 presents an overview of
the inception and development of the Asian crisis with a focus on the movements of the
macroeconomic variables and the structural conditions of the financial systems. Chapter
II shows some evidence of deterioration of fundamentals. Yet the deterioration was not so
severe as to make the outbreak of the Asian currency crisis an inescapable result.

Chapter III’S survey of the currency crisis literature finds that most nonstructural
empirical studies are limited by a lack of robustness to various sensitivity tests and poor
performance in the prediction of actual crises. Therefore, to determine the uniqueness of
the Asian crisis and to improve performance in predicting actual crises, structural model
studies are used to model the currency crisis. Chapter IV then offers an analysis of the
time series properties and forecasts of each variable of the structural currency crisis
models introduced in Chapter III for the derivation of shadow exchange rates and
probabilities of collapse. As a result of the addition of the ARFIMA(p,d,q)-
FIGARCH(P,5,Q) model to the analysis, it is found that some processes exhibit long

memory in both their conditional mean and variances. In Chapter V, long and short-run

 

 

 

 

real money demand functions are estimated for the derivation of shadow exchange rates
and probabilities of collapse. The empirical results of this chapter suggest that both long
and short-run models can be specified in South Korea and in Malaysia. This justifies the
monetary approach using the structural currency crisis model. Chapter VI estimates
shadow exchange rates and probabilities of an exchange rate regime change for South
Korea and Malaysia. Two countries experienced severe currency devaluation. This
employs forecasts for the analyzed variables in Chapter IV and the estimates of real
money demand function found in Chapter V. Both shadow exchange rates and
probabilities of collapse reflecting the presence of weak fundamentals show that there
were reasons to anticipate the 1997 Asian currency crisis.

In Chapter VII, a more extensive analysis of currency crisis with respect to the
number of countries and the currency crisis episodes is performed using panel data. Here,
the focus is on the role of contagion effects on the spread of currency crisis. The
empirical results show that lending booms impact the currency crisis index much more
among developing countries than industrial countries. In addition, contagion effects,
represented by trade linkage and market sentiment, significantly improve the ability to
predict the eruption of a currency crisis after controlling for other macroeconomic
variables.

Based on the preceding empirical results, it appears that weak fundamentals and
contagion effects can be indicators of upcoming currency crisis implying cumulative
depreciation pressure. Nevertheless, a currency crisis cannot erupt without triggering
events such as bank failure. corporate failure or political uncertainty that induce an

equilibrium, currency crisis, to be an inescapable result among the multiple equilibria.

 

 

 

 

To my parents,
Yeon Tae Kim
And
Sung Ja Jung,
And to my wife,

Won Young

 

 

ACKNOWLEDGEMENTS

The completion of my dissertation would not have been possible without the
assiStance of my dissertation committee. The Chair of my committee, Dr. Richard T.
Baillie, provided me with invaluable advice, encouragement, and patience from the
beginning of the process to end. His insight and attention to detail greatly improved the
quality of my dissertation. Truly, he is a great mentor as well as a respectable scholar. I
really appreciate his sincere dedications and inspirations. Dr. Jeffrey Wooldridge’s
insights into the empirical analysis were extremely helpful. Dr. Susan Chun Zhu offered
very helpful comments on my dissertation. The insightful and intuitive comments of Dr.
G. Geoffrey Booth added depth to the final narration of my dissertation. There are other
faculty members, though not on my committee, that have also contributed to my success.
Dr. Christine E. Amsler and Dr. Ana Maria Herrera provided many helpful comments and
suggestions throughout many stages of the dissertation. I would also like to acknowledge
Dr. Young Wook Han and Dr. Chi Rok Han for their help and valuable comments. My
fellow graduate students deserve my thanks for having commented on many parts of my
dissertation. Especially, I would like to thank Sung Kwan Kim, Jong Byung Jun, Yong
Su Cho, Jae Boong Hwang, Jeong Seok Song, and Hyuk Jae Lee.

Finally, I would like to thank my family for their support. My father and mother
always supported me and encouraged me to do my best. I would also thank my father-in-
law and mother-in-law for their support. My deepest gratitude goes to my wife, Won
Young Park, for her faith, encouragement, and patience. This dissertation would not have

been possible without her.

 

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................................. ix
LIST OF FITURES ........................................................................................................... xii
CHAPTER I
INTRODUCTION .............................................................................................................. 1
CHAPTER II
THE CAUSE OF THE ASIAN CURRENCY CRISIS ...................................................... 7
1.The inception and development of the Asian currency crisis ...................................... 8
1.1 The period leading to the crisis: 1995-96 ............................................................. 8
1.2 The unfolding of the crisis in 1997 ....................................................................... 9
2. Movements of macroeconomic variables ................................................................. 1 1
2.1 Current account imbalances ................................................................................ 11
2.2 Output growth ..................................................................................................... 11
2.3 Inflation ............................................................................................................... 12
2.4 lnvestrnent ........................................................................................................... 13
2.5 Savings ................................................................................................................ 13
2.6 Real exchange rate appreciation ......................................................................... 14
3. Structural conditions in financial system .................................................................. 15
3.1 Weak banking system ......................................................................................... 15
3.2 Imbalances in foreign debt accumulation and management ............................... 17
4. A debate between "weak fundamentals and financial panic" analysts ..................... 18
4.1 Evidences presented by financial panic analysts ................................................ 18
4.2 Evidences presented by weak fundamental analysts ........................................... 20
5. Summary ................................................................................................................... 20
CHAPTER III
SURVEY OF LITERATURE AND EXTENDED MODEL ........................................... 32
1. Theoretical literature ................................................................................................. 32
1.1 First generation models ....................................................................................... 32
1.2 Second generation models .................................................................................. 42
2. Empirical literature ................................................................................................... 50
2.1 Nonstructural empirical analyses ........................................................................ 50
2.2 Structural empirical analyses .............................................................................. 54
3. Extended currency crisis model ................................................................................ 64
3.1 Basic model ........................................................................................................ 64
3.2 Extended model .................................................................................................. 68
vi

4. The probability of currency crisis ............................................................................. 70

4.1 Explicit form of probability in the basic model .................................................. 70
4.2 Explicit form of probability in the extended model ............................................ 73
CHAPTER IV

TIME SERIES PROPERTIES OF VARIABLES IN CURRENCY CRISIS MODEL... 78

1. Introduction ............................................................................................................... 78
2. Data set ..................................................................................................................... 79
3. Analysis of time series property and forecast of variables ....................................... 80
3.1 ARFIMA-FIGARCH model ............................................................................... 80
3.2 Empirical results for US inflation ....................................................................... 85
3.3 Empirical results for deviations from PPP .......................................................... 86
3.4 Empirical results for domestic credit .................................................................. 88
3.5 Empirical results for interest rate ........................................................................ 88
3.6 Empirical results for real GDP ............................................................................ 89
4. Conclusion ................................................................................................................ 90
CHAPTER V
ESTIMATES OF REAL MONEY DEMAND FUNCTIONS ....................................... 124
1 . Introduction ............................................................................................................. 124
2. Theoretical framework ............................................................................................ 125
3. Empirical model ...................................................................................................... 126
3.1 Error-correction models .................................................................................... 126
4. Application of ECM to the estimation of real money demand ............................... 127
4.1 Data set .............................................................................................................. 127
4.2 Unit-root tests .................................................................................................... 129
4.3 Residual based cointegration tests .................................................................... 134
4.4 Johansen's full information maximum likelihood estimation ........................... 137
5. Conclusion .............................................................................................................. 144
CHAPTER VI
FORECAST OF SHADOW EXCHANGE RATE AND PROBABILITY OF COLLAPSE
......................................................................................................................................... 173
1 . Introduction ............................................................................................................. 173
2. Estimation procedure .............................................................................................. 174
3. Empirical results ..................................................................................................... 176
3.1 Behavior of variables in the structural model ................................................... 177
3.2 Estimated shadow exchange rate ...................................................................... 178
3.3 Estimated probability of collapse ...................................................................... 181
4. Conclusion .............................................................................................................. 182

vii

 

CHAPTER VII
COMMON FUNDAMENTALS AND CONTAGION EFFECT IN CURRENCY CRISIS

......................................................................................................................................... 197

1 . Introduction ............................................................................................................. 1 97

2. Theoretical framework ............................................................................................ 199

2.1 A simple model ................................................................................................. 200

3. Empirical analysis ................................................................................................... 206

3.1 Defining variables in the empirical model ........................................................ 206

3.2 Data set .............................................................................................................. 213

3.3 Regression analysis ........................................................................................... 214

3.4 Robustness ........................................................................................................ 223

3.5 Predicting the Asian currency crisis .................................................................. 224
CHAPTER VIII

CONCLUSION .............................................................................................................. 244

BIBLIOGRAPHY ........................................................................................................... 250

viii

LIST OF TABLES

Table 1. Current account ................................................................................................... 22
Table 2. GDP growth rate ................................................................................................. 22
Table 3. Inflation rate ........................................................................................................ 23
Table 4. Investment rate .................................................................................................... 23
Table 5. Incremental capital output ratio .......................................................................... 24
Table 6. Saving rate .......................................................................................................... 24
Table 7. Government budget balance ............................................................................... 25
Table 8. Real exchange rate .............................................................................................. 25
Table 9. Bank lending to private sector ............................................................................ 26
Table 10. Non-performing loans .................................................................... _ ................... 26
Table 11. Foreign liabilities of the banking system .......................................................... 27
Table 12. Short-term debt ................................................................................................. 27
Table 13. Ratio of M2 to foreign reserves, Asian countries ............................................. 28
Table 14. Ratio of M2 to foreign reserves, G7 countries .................................................. 28
Table 15. Stock market prices indeces .............................................................................. 29

Table 16. The change of macroeconomic conditions in the Asian countries before
currency crisis ........................................................................................................... 30

Table 17. Data sources and definitions ............................................................................. 92

Table 18.

Estimated ARFIMA(p.d.q)-FIGARCH(P.dQ) models for US monthly inflation

rate ............................................................................................................................. 93
Table 19. Estimated ARFIMA(p,d,q)-FIGARCH(P,dQ) models for deviations from PPP
................................................................................................................................... 94
Table 20. Estimated ARFIMA(p,d,q)-FIGARCH(P,dQ) models for domestic credit ..... 95
Table 21. Estimated ARFIMA(p,d,q)—FIGARCH(P,(2Q) models for interest rate ........... 96
Table 22. Estimated ARFIMA(p,d,q)-FIGARCH(P,5,Q) mOdels for real GDP ............... 97
Table 23. Dickey-Fuller tests for unit roots: Model I ..................................................... 147
Table 24. Dickey-Fuller tests for unit roots: Model II .................................................... 148
Table 25. Phillips-Perron tests for unit roots .................................................................. 149
Table 26. KPSS tests for stationarity: Model I ............................................................... 150
Table 27. KPSS tests for stationarity: Model 11 .............................................................. 151
Table 28. Testing for no cointegration in demand for real M2 ....................................... 152
Table 29. Residual misspecification tests (South Korea) ............................................... 153
Table 30. Residual misspecification tests (Malaysia) ..................................................... 154
Table 31. Test of the cointegration rank (South Korea) ................................................. 155
Table 32. Test of the cointegration rank (Malaysia) ....................................................... 156
Table 33. Normalized cointegrating vectors (,8) and error correction coefficient (6:)
(South Korea) .......................................................................................................... 157
Table 34. Normalized cointegrating vectors ( ,8) and error correction coefficient (oi)
(Malaysia) ............................................................................................................... 158
Table 35. Weak exogeneity test (South Korea) .............................................................. 159

Table 36. Weak exogeneity test (Malaysia) .................................................................... 159
Table 37. Estimated coefficients of short-run model (South Korea) .............................. 160
Table 38. Estimated coeffcients of short-run model (Malaysia) ..................................... 162
Table 39. Summary of studies of the demand for real money balances involving
Cointegration/Error-Correction modeling in South Korea and Malaysia ............... 164
Table 40. Actual and shadow exchange rates ................................................................. 185
Table 41 Probability of collapse .................................................................................... 186
Table 42. Currency crisis index ...................................................................................... 229
Table 43. Lending boom ................................................................................................. 230
Table 44. Real depreciation ............................................................................................ 231
Table 45. Reserve adequacy (Industrial countries) ......................................................... 232
Table 46. Reserve adequacy (Developing countries) ...................................................... 233
Table 47. Trade linkage .................................................................................................. 234
Table 48. Country effects ................................................................................................ 235
Table 49. Benchmark regression ..................................................................................... 236
Table 50. Contagion effects ............................................................................................ 237
Table 51. Additional determinants .................................................................................. 238
Table 52. Robustness for the crisis index ....................................................................... 239
Table 53. Robustness for the dummies ........................................................................... 240
Table 54. Actual and predicted currency crisis index ..................................................... 241
Table 55. Previous empirical studies .............................................................................. 243

xi

 

LIST OF FIGURES

Figure 1. Attack time in a certainty model ....................................................................... 75
Figure 2. Attack times with attack-conditional policy shift .............................................. 76
Figure 3. Devaluation cost and policy loss ....................................................................... 77

Figure 4. Correlograms of standardized residuals from ARFIMA(0,d_,I) model for

U.S.CPI inflation ....................................................................................................... 98

Figure 5. Correlograms of standardized residuals from ARFIMA(0,d,1)-FIGARCH(1,6,1)

model for U.S.CPI inflation ...................................................................................... 99
Figure 6. Actual and fitted values of US. CPI inflation rate .......................................... 100
Figure 7. Forecasted values of US. CPI inflation rate ................................................... 101

Figure 8. Correlograms of standardized residuals for deviations from PPP, Indonesia. 102
Figure 9. Correlograms of standardized residuals for deviations from PPP, South Korea
................................................................................................................................. 103
Figure 10. Correlograms of standardized residuals for deviations from PPP, Malaysia 104
Figure 11. Correlograms of standardized residuals for deviations from PPP, Philippines
................................................................................................................................. 105
Figure 12. Correlograms of standardized residuals for deviations from PPP, Thailand. 106
Figure 13. Correlograms of standardized residuals for domestic credit, Indonesia ........ 107
Figure 14. Correlograms of standardized residuals for domestic credit, South Korea 108
Figure 15. Correlograms of standardized residuals for domestic credit, Malaysia ......... 109

Figure 16. Correlograms of standardized residuals for domestic credit, Philippines ..... 110

xii

 

mun

Figure 17.
Figure 18.
Figure 19.
Figure 20.
Figure 21.
Figure 22.
Figure 23.
Figure 24.
Figure 25.
Figure 26.
Figure 27.
Figure 28.
Figure 29.
Figure 30.
Figure 31.
Figure 32.
Figure 33.
Figure 34.
Figure 35.
Figure 36.
Figure 37.

Figure 38.

Correlograms of standardized residuals for domestic credit, Thailand ......... 1 1 1

Correlograms of standardized residuals for interest rate, Indonesia .............. 112
Correlograms of standardized residuals for interest rate, South Korea ......... 113
Correlograms of standardized residuals for interest rate, Malaysia .............. 1 14
Correlograms of standardized residuals for interest rate, Philippines ........... 115
Correlograms of standardized residuals for interest rate, Thailand ............... 116
Correlograms of standardized residuals for interest rate, U.S ....................... 117
Correlograms of standardized residuals for real GDP, Indonesia ................. 118
Correlograms of standardized residuals for real GDP, South Korea ............. 119
Correlograms of standardized residuals for real GDP, Malaysia .................. 120
Correlograms of standardized residuals for real GDP, Philippines ............... 121
Correlograms of standardized residuals for real GDP, Thailand ................... 122
The log of real M2 (South Korea) ................................................................. 166
The log of real GDP (South Korea) ............................................................... 166
Interest rate (South Korea) ............................................................................ 167
Difference between domestic and foreign interest rate (South Korea) .......... 167
The log of real M2 (Malaysia) ....................................................................... 168
The log of real GDP(Malaysia) ..................................................................... 168
Interest rate (Malaysia) .................................................................................. 169
Difference between domestic and foreign interest rate (Malaysia) ............... 169
Recursive estimates of the long-run parameter of real GDP (South Korea) . 170

Recursive estimates of the long-run parameter of interest rate (South Korea)

........................................................................................................................ 170

xiii

Figure 39.

Recursive estimates of the long-run parameter of the difference of domestic

and foreign interest rate (South Korea) ................................................................... 171
Figure 40. Recursive estimates of the long-run parameter of real GDP (Malaysia) ....... 171
Figure 41. Recursive estimates of the long-run parameter of interest rate (Malaysia) 172
Figure 42. Recursive estimates of the long-run parameter of the difference of domestic

and foreign interest rate (Malaysia) ........................................................................ 172
Figure 43. The log of real M2 (South Korea) ................................................................. 187
Figure 44. The log of domestic credit (South Korea) ..................................................... 187
Figure 45. The log of real GDP(South Korea) ................................................................ 188
Figure 46. Interest rate (South Korea) ............................................................................ 188
Figure 47. Deviation from PPP (South Korea) ............................................................... 189
Figure 48. The log of real M2(Malaysia) ........................................................................ 189
Figure 49. The log of domestic credit (Malaysia) ........................................................... 190
Figure 50. The log of real GDP (Malaysia) ................................................................... 190
Figure 51. Interest rate (Malaysia) .................................................................................. 191
Figure 52. Deviation from PPP (Malaysia) ..................................................................... 191
Figure 53. US interest rate .............................................................................................. 192
Figure 54. US CPI ........................................................................................................... 192
Figure 55. The actual and shadow exchange rates (South Korea) .................................. 193
Figure 56. The actual and shadow exchange rates (Malaysia) ........................................ 193
Figure 57. The probabilities of collapse I (South Korea) ............................................... 194
Figure 58. The probabilities of collapse II (South Korea) .............................................. 194
Figure 59. The probabilities of collapse 111 (South Korea) ............................................. 195

xiv

Figure 60. The probabilities of collapse I (Malaysia) ..................................................... 195

Figure 61. The probabilities of collapse II(Malaysia) ..................................................... 196

Figure 62. The probabilities of collapse 111 (Malaysia) .................................................. 196

Figure 63. Actual and predicted currency crisis index .................................................... 242
XV

 

CHAPTER I

INTRODUCTION

Episodes of speculative attacks on currencies in the 1990s (such as the 1992-93
crises in the European Monetary System and the 1994 Mexico peso collapse) have
generated considerable debate on whether currency and financial instability should be
attributed to arbitrary shifts in market expectations and confidence instead of weak
economic fundamentals. These viewpoints about the underlying causes of a currency
crisis are summarized by two main views. According to one view, advocated by
‘fundarnentalists’, crises reflect a sustained deterioration in macroeconomic fundamentals
and defective economic policies. Although market overreaction can exacerbate currency
crises, fundamentalists stress that the cause of crises are due to structural factors. Another
view, favored by ‘non-fundamentalists’, is that sudden shifts in market expectations and
confidence are the crucial sources of initial financial turmoil, its propagation over time,
and regional contagion. While the macroeconomic performance of some countries with
currency crises was somewhat weak, the extent and depth of the crises should not be
attributed to the sharp deterioration in fundamentals but rather to the panic of domestic
and international investors. Yet, advocates of both the ‘fundamentalist’ and the ‘non-
fundamentalist’ view, agree in principle that a deteriorating macroeconomic outlook is a
necessary condition for an economy to be vulnerable to a crisis. In fact, it is well understood

that multiple instantaneous equilibria, which provide the theoretical preconditions for self-

 

 

fulfilling crises to occur as rational events, are only possible in a region in which the current
or anticipated economic performance is sufficiently weak.

Identifying the source of currency crises has important implications for economic
policy: if currency crises are indeed caused by fundamentals, the most effective way to
prevent them is to support fiscal and monetary policies that stabilize exchange rates. If, on
the other hand, self-fulfilling speculation can trigger crises regardless of fundamentals, there
might be a case for specific measures to deter such speculation. One of the measures is a
capital control that might help governments defend their currencies.

Whereas the speculative attacks on currencies in the early 19905 have been
sufficiently analyzed to reach an agreement about the main source, the cause of the Asian
currency crisis in 1997-98 is still under debate.

The main objective of this dissertation is to further study the issue of the causes of
the Asian crisis and improve implementation in predicting actual currency crises. The
analysis presented here examines the crisis using higher frequency data and more refined
models than previous studies. The subsequent chapters are organized as follows:

Chapter II presents an overview of the Asian crisis with an emphasis on the
movement of macroeconomic variables and the structural conditions of financial systems.
The analysis concentrates on South Korea, Indonesia, Malaysia, the Philippines and
Thailand. These countries experienced more severe currency depreciation than other
Asian countries. Causal analysis of the currency crisis does not allow one to draw
conclusions on their causes. That is, the fundamentalist view is not a more appealing

explanation of the crises than the non-fundamentalist view, and vice versa.

In Chapter III, a survey of literature on currency crises is offered in addition to an
extended model. The theoretical literature consists of two generations of models. First
generation models, as in Krugman (1979), show how speculative attacks occur when the
fundamentals are weak. Second generation models study the following two questions:
“What happens when government policy reacts to changes in private behavior?” or “What
happens when the government faces an explicit trade-off between a fixed exchange rate
policy and other objectives such as economic growth, low unemployment or low
inflation?”

The nonstructural studies such as the classic study by Frankel and Rose (1996)
have attempted to exploit the high variability associated with multi-country information.
Estimation results from their probit regression are largely consistent with the theoretical
literature, however, the results are not robust and do not forecast crises well. Structural
studies, beginning with Blanco and Garber’s (1986), have presented strong evidence
suggesting that domestic macroeconomic indicators play a key role in determining a
currency crisis. Otker and Pazarbasioglu (1996, 1997b) also computed the probability of
an exchange rate regime change from the European financial crises in 1992 and 1993 and
the Mexican financial crisis in 1994. These studies focus on a particular country in a
specific time period illustrating the uniqueness of each country’s currency crisis.

An extension of the speculative attack model, suggested by Krugman (1979) and
formalized by Flood and Garber (1984a), is derived to capture the uniqueness of the
Asian crisis and to improve the performance in predicting actual crises in Chapter III. The

model is a stochastic version of the monetary approach to exchange rate determination, in

which the government and monetary authority of a small open economy are committed to
maintaining the exchange rate by employing some form of a fixed exchange rate system.
Chapter IV provides an analysis of time series property and forecast of all of the
variables introduced in the models of Chapter III. The analysis is used to derive shadow
exchange rates and probabilities of an exchange rate regime change. Most of the previous
studies on the structural analysis of currency crises, Blanco and Garber (1986), Cumby
and Van Wijnbergen (1989) and Otker and Pazarbasioglu(1996, 1997b), do not estimate
the properties of variables in the model but assume an AR(p) model. Unlike those studies,
Goldberg (1994) estimates variables’ time series properties and forecasts values one step
ahead, using ARIMA models and Akaike tests. Whereas ARIMA models are able to
capture autocorrelations that decay at an exponential rate, they cannot be applied to long

memory processes where autocorrelations decay slowly. Therefore, the ARFIMA(p,d,q)-

FIGARCH(P,5,Q) model is used to capture the part of economic and financial time series
that exhibit long memory in both their conditional mean and variances. Once the specific
form of the model is determined and the parameters are estimated, the model is fit over
each of the sample time periods to calculate forecasts. The forecasts are used to derive the
shadow exchange rate and the probability of collapse.

In Chapter V, long and short-run real money demand functions of the structural
currency crisis models introduced in Chapter III are estimated. The structural analyses of
currency crises, Blanco and Garber (1986), Cumby and Van Wijnbergen (1989) and
Goldberg (1994), estimated a real money demand function without consideration of the

non-stationarities of the variables. Therefore, their results suffer from a spurious

regression problem and the conventional t-ratio and F significance tests cannot be
applied. Unlike previous studies, cointegration and error correction techniques are applied
for the modeling of real money demand to remove the spurious regression problem and to
use the r-ratio and F significance tests. First, a theoretical framework and an empirical
model are presented. Next, unit-root tests are presented to detect the non-stationarity of
variables. Lastly, a residual based tests, testing for the number of cointegration relations
and estimating the cointegrating vectors, were performed.

Based upon forecasts of each variable and estimates of the real money demand
function, shadow exchange rates and probabilities of an exchange rate regime change are
derived for South Korea and Malaysia in Chapter VI. The derived shadow exchange rates
and probabilities show that fundamentals were weak prior to the Asian crisis.

Chapter VII introduces an empirical model that performs an extensive analysis on
currency crisis episodes using panel data. The model investigates a currency crisis
focusing on the various variables or other external effects, e. g. contagion effects, whereas
the traditional approach emphasizes the role played by declining international reserves in
triggering the collapse of a fixed exchange rate. Sachs, Tomell and Velasco (1996) and
Tomell (1999) seek to identify macroeconomic variables that can help explain which
countries were vulnerable to “contagion effects”, but only in emerging markets. Glick and
Rose (1998) find that countries with important trade links to the country that initially
experienced a crisis are more likely to experience a crisis themselves. Masson (1998)
suggests that the contagion effect unexplained by the common external effects and trade
links played a major role in the Mexican and Asian crises. I check if the macroeconomic

variables that explain the cross-country variation in the severity of crises in emerging

 

markets also have explanatory power in non-emerging markets. In addition, the extent of
the contagion effect in all aspects of common external effects, trade linkage and market
sentiment is examined. Finally. a currency crisis is predicted using the contagion effect as
well as the weak fundamentals to make the predictions more precise than the previous
studies’.

Chapter VIII summarizes all the results derived in this dissertation and suggests

policies for the prevention of currency crises.

CHAPTER II

THE CAUSE OF THE ASIAN CURRENCY CRISIS

There have been two main alternative views about the causes of the Asian
economic, currency and financial crisis that started in 1997.l One of the views focuses on
the role of weak fundamentals such as growing current account deficits, real currency
appreciation, bad loans, overinvestment, and foreign debt accumulation, as being
contributors to the crisis.

In contrast, the other view stresses sudden arbitrary shifts in market expectations
and confidence, i.e. financial panic, as the key cause of the crisis. Radelet and Sachs
(1998b) admit that there were significant underlying fundamental problems in the Asian
economies, but assert that these problems were less severe than financial panic in
triggering a crisis of such magnitude.

This chapter presents an overview of the Asian currency crisis. For the purpose of
finding evidence for either weak fundamentals or financial panics, the movement of
macroeconomic variables and the structural conditions of the financial system are
discussed. The analysis focuses on South Korea, Indonesia, Malaysia, the Philippines and
Thailand. These countries experienced more severe currency depreciation than other

Asian countries.

 

A list of recent studies are available at http://www.stem.nyu.edu/~nroubmI/asra/ASIaHomepagehtml

___— ~fl
—,

 

1. The inception and development of the Asian currency crisis
1.1 The period leading to the crisis: 1995-96

In Thailand, the macroeconomic and structural weakness that was growing
throughout the 19905 became more serious in 1995-96. The real GDP growth rate slowed
down to 5.5 percent in 1996 from 8.9 percent in 1994. In addition, the current account
deficit worsened from 5.6 percent of GDP in 1994 to 8.1 percent of GDP in 1996. These
deficits had been financed by short-term capital inflows that led to a sharp accumulation
of short-term debt which increased from 95.98 percent of foreign reserves in 1993-94 to
106.95 percent in 1995-96. By the end of 1996, the macroeconomic indicators of
Thailand already showed very unstable conditions: large current account deficits,
accumulation of short-term foreign debt, and low profitability of real investment projects.

In Indonesia, a sharp increase in the GDP growth rate to 15.9 percent in 1994 and
8.2 percent in 1995 brought along worrisome signs of overheating. Inflation remained
high, while the country’s trade surplus suffered a steep drop. The govemment’s response
of a slightly deflationary budget and a modest tightening of monetary policy was initially
cautious. The government did not want higher interest rates to fuel further capital inflows
and appreciate the currency. The Bank of Indonesia also widened the rupiah’s trading
band from 2 percent to 3 percent around the daily mid-rate, hoping that the additional
trading risk of holding the rupiah would balance the incentive to invest in domestic assets
provided by the higher interest rates. The band was further widened from 3 percent to 5
percent in June 1996, and again from 5 percent to 8 percent in September 1996.

The current account deficit had widened between 1994 and 1995 in Malaysia, as

well, reaching 8.4 percent of GDP in 1995. Notably, in 1994 and 1995 foreign direct

 

investment failed to cover the full amount of the deficit. During the effort to restrain
domestic demand, the Malaysian interest rate had become too attractive to be ignored by
foreign fund managers. In 1996, short-term debt sharply increased to 40.9 percent of
foreign reserve compared to that of 30.6 percent in 1995.

A serious worsening of macroeconomic conditions already had occurred in South
Korea between 1995 and 1996. The current account deficit rapidly widened from 1.5
percent of GDP in 1994 to 4.8 percent in 1996, leading to a record-breaking accumulation
of short-term foreign debt. The 1996 growth rate of GDP decreased to 7.1 percent from
the previous year’s 8.9 percent. Reflecting weak financial conditions of the
conglomerates, the stock market fell sharply in the two-year period 1995-96, down by 36
percent relative to the 1994 peak. The won also weakened during 1996.

Relative to the other countries in the region, economic conditions were more
stable in the Philippines. Under IMF supervision, the Philippines experienced a
sustainable GDP growth rate in the 1990’s although lower than some of the southeastern
Asian countries. The government’s budget was in surplus. However, the current account

deficit was large, and the currency had severely appreciated in real terms.

1.2 The unfolding of the crisis in 1997

By early 1997, macroeconomic conditions had deteriorated in most of the region.
In the government’s effort to defend collapsing financial institutions, strong speculative
attacks on the baht forced Thailand to let the currency float on July 2, a crucial date in the
chronology of the Asian crisis. Before the change of the exchange rate regime, Thailand

used a highly managed exchange rate system which allowed a narrow band for the float

 

of the exchange rate. Following Thailand. the Philippine central bank allowed the peso to
move in a wider range against the dollar. Subsequently, the peso started to depreciate
sharply.

As with Thailand, Malaysia had over a decade of extremely large current account
deficits. Bank Negara announced ceilings on lending to the property sector and for
purposes of stocks and shares in order to regulate a booming speculative bubble in real
estate and equity lending. This caused foreign investors, led by US fund managers, to
start selling their Stocks. Under depreciation pressure, the Malaysian central bank
abandoned its defense of the ringitt on July 14.

The Indonesian rupiah began to come under severe depreciation pressure with
heavily increasing external debt. Failing in its defense, Indonesia abolished its system of
managing the exchange rate through the use of a band and allowed it to float on August
14.

In early 1997, South Korea was shaken by a series of bankruptcies by large
conglomerates that had heavily borrowed in previous years to finance their investment
projects. The bankrupt conglomerates included Hanbo steel, Sarnmi steel and Kia. The
macroeconomic indicators in early 1997 fully reflected the extent of this crisis; the
current account deficit was increasing, export growth was falling, and industrial
production growth rates were below previous levels. The speculative attack started in
early November and South Korea requested IMF assistance on November 21. Finally the

government announced it would allow the Won to float on December 16.

 

2. Movement of macroeconomic variables
2.1 Current account imbalances

As shown in Table 1, several Asian countries whose currencies sharply
depreciated in 1997 had experienced somewhat sizable current account deficits in the
19905. Thailand and Malaysia, both of which experienced deficits for over a decade,
exhibit the largest and most persistent current account imbalances in our sample. The
current account deficits in the two countries were over 6 percent of GDP on average
between 1995 and 1996. The Philippines also experienced long-term imbalances. The
deficit problem worsened in 1996. Starting the decade with a large imbalance, the current
account imbalance of Indonesia increased to 3 percent of GDP between 1995 and 1996
although it shrank in 1992-93. In South Korea, the current account deficit was low in the
early 19905 (1-3 percent of GDP) and virtually negligible in 1993. However, since 1993
the imbalance grew very fast, approaching almost 5 percent of GDP in 1996.

Fast-growing current account deficits likely increase currency depreciation
pressure. The expanding current account deficits in these five countries could be

considered as one of the factors forewarning a coming currency crisis.

2.2 Output growth

Table 2 presents the growth data in our sample of Asian countries in the 19905.
As shown in Table 2, GDP growth rates were remarkably high in the 19905. Growth rates
averaging more than 7 percent were the norm. But the growth rate slowed down in 1996,
a year before the crisis. Only the Philippines, where growth rates were low in the early

19905, geared up its growth rate to 6 percent in 1996 from 5 percent in 1995.

ll

Accepting the traditional view that a large current account deficit is likely to be
sustainable when growth is high, the Asian countries did not appear to have a
sustainability problem until 1995. But the consumption and investment boom, as well as
large capital inflows driven by overly optimistic beliefs that the economic expansion
would persist, added instability to the value of currencies with a slowdown in the growth
rate. In such conditions, an external shock that leads to a sudden change in expectations

can cause a rapid reversal of capital flows and trigger a currency collapse.

2.3 Inflation

Table 3 presents inflation rates in our sample of Asian countries. In all countries,
inflation rates were relatively low in the 19905. The only exception was the Philippines
where inflation was close to 20 percent in 1990-91(but falling to 8 percent by 1995). It is
believed that high inflation rates leave fixed or semi-fixed exchange rate regimes
potentially exposed to speculative attacks. The low inflation rates observed signal sound
macroeconomic policy and sustainability of the regime. However the banking and
financial sector problems experienced by several Asian countries over the 19905 raised
considerable doubt about their ability to keep inflation low in the near future. These
doubts were related to the possibility that the cost of the banking sector bail-outs might
induce increased usage of seigniorage, and would require infusions of liquidity to prevent

systemic runs.

2.4 Investment

Evidence on investment rates in Asian countries is shown in Table 4. Unlike the
Latin American countries that experienced currency and financial crises in the recent past.
the Asian countries were characterized by very high rates of investment throughout the
19905. These rates were well above 30 percent of GDP in most countries, with the
exception of the Philippines that had rates in the 20-25 percent range.

Despite the high investment rate of the Asian countries, the profitability of
investment -the ratio between the investment rate and the rate of output growth- given by
Table 5 suggests the efficiency of investment was falling in the three years, 1994-1996,
prior to the 1997 crisis with the exception of the Philippines. Also the investment boom
was confined to the non-traded sector (commercial and residential construction, as well as
inward-oriental services) adding an unsustainable factor to the sharply growing current

account deficit.

2.5 Savings

Data on saving rates in Asia are reported in Table 6, and to some extent stand for
the mirror of the investment rates in Table 4. Asian countries were characterized by very
high savings rates throughout the 19905- in many cases above 30 percent of GDP and in
some cases above 40 percent. Looking at the data in Table 7 before the crisis, there is
little evidence of public dissaving so that the current account imbalances do not appear to
be the result of increased public sector deficits. The absence of fiscal imbalances in the
years preceding the crisis, however, should not be regarded as pervasive evidence against

the fiscal roots of the Asian crisis. The pre—crisis years were a period of excessive credit

13

growth in the banking system, leading to a large stock of non-performing loans and the
eventual collapse of several financial institutions. The cost of restructuring the financial
sector could have been an implicit fiscal liability for the Asian countries. Such a liability
was not reflected by data on public deficits until the outbreak of the crisis, but affected
the sustainability of the pre-crisis current account imbalances since it generated

expectations of radical policy changes or currency devaluations.

2.6 Real exchange rate appreciation

A significant real exchange rate appreciation may be associated with a loss of
competitiveness and a structural worsening of the trade balance, thus weakening the
sustainability of the current account. Data on the real exchange rate of the Asian countries
in Table 82 shows that the real exchange rate had appreciated by 15.7 percent in Malaysia,
26.1 percent in the Philippines, 8.0 percent in Indonesia, and 0.5 percent in Thailand by
the end of 1996, taking 1990 as the base year. In South Korea, the currency depreciated in
real terms by 9.2 percent. This suggests that, with the significant exception of South
Korea, all the currencies that crashed in 1997 had experienced real appreciation. It should
be stressed that in a number of countries, a large part of the real appreciation occurred
after 1995, in parallel with the strengthening of the US dollar. The sharp increase of the
US dollar relative to the Japanese yen and the European currencies since the second half
of 1995 led to deteriorating competitiveness in most Asian countries whose currencies

were effectively pegged to the dollar.

3. Structural conditions in the financial system
3.1 Weak banking system

In the 19905 the countries of East Asia performed a financial deregulation and
capital liberalization. The financial liberalization involved loosening restrictions on both
interest rate ceilings and the type of lending allowed. Bank lending increased sharply
prior to the crisis, with much of it financed by inflows of international capital.

Of course, the problem was not that lending expanded, but rather that it expanded
so rapidly that excessive risk-taking occurred. In fact, the increasing proportion of non-
performing loans indicates that many of the loans made by banks were invested in risky
and low profitable projects or used for real estate, non-traded area. Therefore, they

weakened the banking system.

3.1.1 Lending boom
As shown in Table 9, domestic bank lending to the private sector shows a steep
upward trend in the five countries prior to the crisis. The most extreme case was the
Philippines, where banking claims on the private sector, as a percent of GDP, increased
by more than 60 percent between 1994 and 1996. It was also large in Malaysia (25
percent) and Thailand (12 percent). Though more modest in Indonesia and South Korea,
the magnitude of credit growth between 1994 and 1996 was much higher than in the early

19905

 

2 . . .
The source of these data IS the JP Morgan RER series that go back to 1970; the base year for the trade weight rs

l5

3.1.2 Accumulation of bad loans

One of the main problems faced by the Asian countries was that many of the loans
made by banks were invested in risky and unprofitable projects or used for real estate,
property and the purchase of equity funds. A possible indicator of investment in risky
and low profitable projects is the proportion of non-performing loans (NPLs) in total
loans (Table 10). Since the 1997 crisis may have crippled otherwise healthy loans, it is
appropriate to refer to data at the onset of the crisis. In 1996, the NPLS were estimated at
8—14 percent for the five afflicted countries. For the purpose of comparison, the estimated

NPLs were 3-4 percent in Hong Kong, Singapore and Taiwan.

3.1.3 Loans financed by foreign liabilities

Large increases in the foreign liabilities of banks point out that much of the bank
lending was mostly financed by borrowing from abroad (Table 11). In the Philippines,
foreign liabilities soared from 5.5 percent of GDP at the end of 1993 to 17.4 percent of
GDP three years later. In South Korea, the corresponding liabilities of the banking system
more than doubled from 4.5 percent of GDP in December 1993 to 9.4 percent of GDP in
December 1996. In Thailand, foreign liabilities jumped more sharply from 5.9 percent of
GDP in 1992 to 26.6 percent in 1996. In Indonesia, though the liabilities remained at a
more modest level, much of the offshore borrowing was undertaken directly by private
firms. The only exception was Malaysia where foreign liabilities fell off sharply in 1994

and slightly increased to 11.4 percent in December 1996.

 

I990.

3.2 Imbalances in foreign debt accumulation and management
3.2.1 Rising share of short-term debt

If a large fraction of a country’s external liabilities are short-term, a crisis may
take the form of a pure liquidity shortfall — the inability by a country to roll over its short-
term liabilities. The experiences of Mexico with its short-term public debt in 1994-95 and
of several Asian countries with private external liabilities in 1997 provide examples of
liquidity problems. The figures corresponding to the ratio of short-term debt to foreign
reserves are presented in Table 12. All the countries except the Philippines have
somewhat increasing ratio after 1993. In South Korea, the ratio sharply grew from 54.1

percent to 171.5 percent in 1995.

3.2.2 Foreign Exchange Reserves

Large foreign exchange reserves facilitate the financing of a current account
deficit and enhance the credibility of a fixed exchange rate policy. Foreign exchange
reserves and a small external debt burden reduce the risk of external crises, and enable a
country to finance a current account deficit at lower costs. The real rate paid on the debt
indicates the market’s evaluation of the country’s ability to sustain a current account
deficit.

To measure the sufficiency of foreign exchange reserves, the ratio of money
assets to foreign reserves is considered since in the event of an exchange rate crisis, all

liquid money assets can potentially be converted into foreign exchange. Calvo(l998)

l7

suggests using the ratio of a broad measure of liquid monetary assets to foreign reserves,
for instance, the ratio of M2 to foreign reserves.

Table 13 reports the ratio of M2 to foreign reserves. In most Asian countries the
ratio was unusually high in 1996-97. In Indonesia, the ratio constantly rose throughout
the 19905 and reached a peak as high as 7.1 in 1995. In South Korea, M2/F X was equal to
6.5 beforel997, and rose to 10.5 by the end of 1997. In Malaysia, the ratio increased from
2.9 in 1990 to 3.7 at the end of 1996. In the Philippines, the ratio declined marginally
from 4.8 in 1991 to 4.5 in 1996. In Thailand, the ratio went from 4.5 in 1990 to 3.9 in
1996.

Table 14 indicates that the ratios of most G7 countries are very high compared to
those Asian countries. In addition, the ratios of all G7 countries except Japan also rose
throughout the 19905 even though their currencies were not attacked by speculative
agents in 1997. Therefore, the M2/F X needs further empirical study to verify whether it

is an appropriate indicator of currency crisis.

4. A Debate between “weak fundamentals and financial panic” analysts
4.1 Evidence presented by financial panic analysts

Financial panic supporters accept that warning signs such as current account
deficits, bad loans and overinvestment were reasons for financial weakness in the five
countries. But they maintain that those Signs were not enough to warrant the magnitude of
the Asian crisis.

It is generally believed that some aspects of the real economy in at least some

crisis economies were solid. Government budgets, which were at the center of economic

l8

crises in Latin America in the 19805, indicated regular surpluses in each Asian country
during the 19905, as shown in Table 7. GDP growth rates were very high in the 19905 as
well.

Although some analysts expected the possibility of a crisis’, such warnings were
unusual. Inflow of capital remained strong through 1996 and, in most cases, until mid
1997. The only exception is found in the stock markets in Thailand and South Korea,
where foreign investors became uneasy in 1996, as shown in Table 15. In Malaysia,
though stock markets began a rather steep decline in March 1997, bank lending continued
to be very strong at least until mid-year. In Indonesia, both the stock market and bank
lending remained strong until mid-1997.

Credit rating agencies, such as Standard & Poor’s and Moody, provide an ongoing
assessment of credit risk in emerging markets. If the market had expected a financial
crisis and public sector bailouts, the ratings of sovereign bonds should have fallen in the
pre-crisis period. However, the rating agencies did not signal any risk until after the onset
of the Asian crisis. Long-term debt ratings remained unchanged throughout 1996 and the
first half of 1997 for each of the Asian countries except the Philippines, where the debt
rating was actually upgraded in early 1997. In each country. the outlook was described as
“positive” or “stable” through June 1997. Only until weeks after the crisis started, did
agencies downgrade the region’s debt.

Another measure of expectations for the region may be found in IMF reports. The

IMF gave very little indication that there was any macroeconomic risk to the Asian

 

3 See. for example, Park (1996)

I9

region. For example, World Economic Orr/look (IMF, December 1997) predicted 6
percent growth for South Korea in 1998, 7 percent for developing Asia (or 5 percent for

developing Asia excluding China and India).

4.2 Evidence presented by weak fundamental analysts

Advocates of the weak fundamentals hypothesis challenge the financial panic
View. First, they assert that credit ratings have had no informational value in forecasting
currency crises for the last 20 years. As credit ratings failed to predict the crises of the
early 19805, in which fundamentals were obviously at work, one cannot suppose that
their failure to predict the Asian crisis is evidence that the crisis was due to financial
panic.

Second, IMF reports are not generally informative in predicting a crisis. Given the
ability of IMF reports to sharply affect markets, such reports are always written in terms
that express concern in very cautious terms.

Another piece of evidence is that countries with more sound fundamentals were
spared the most serious collapses. In fact, Taiwan, Singapore, Hong Kong, and China

were less affected by the regional turmoil.

5. Summary

Two main causes of the Asian currency crisis, financial panic and weak
fundamentals, have emerged in recent debate.

Financial panic supporters admit that there were crucial underlying problems in

the Asian economies, but assert that these problems were not severe enough to warrant a

20

 

financial crisis of such large magnitude. In contrast, weak fundamental analysts stress the
significant role of weak fundamentals in causing the crisis. Given the weakness of the
macroeconomic stances in the afflicted countries, the crisis was an inescapable outcome
rather than just a financial panic.

While the evidence given in this chapter reflects that deterioration in
macroeconomic fundamentals and poor economic policies could be a root of the crisis, it
does not strongly convince us that fundamentals had deteriorated severely enough to
make the crisis inevitable.

In summation, the empirical evidence on the causes of the Asian crisis is not

conclusive, demanding more formal studies on the issue.

21

 

.EonU :0 03356 .metmcea. NSueeSK NatcheEmEN .5): Hoe—Dom

 

8.4 m3 2:- 93. Now a; 43 m3 8.» w; 8.: @352.
m3 N2 a3- 8a a? Re ”2. N2 43 who- am 853:5
EL 3m Rs- :2 ”3 3a Ea m3 2: we.” Ea has“:
5m 2a 3:. 8+ was NS 3.2 One 3.0 was 42 28:85
3a 8.2 Be- 5% o: as.” mm.” m; 8a m; Ga «28. .m
co. as. as. as. ea. mm. «a. mm. as. _a. ca.

Age as aaew So :0. 2%...

.EoMA—U so 03285 antenna. Nemonufik #3823523 .m—z— “ooh—Dom

 

E 3.2 ES 82. 2a. 8.x- 8a- mow- BS. EN. o2- 22:5
was :2: mmm mmm. Re- SN- 8.4- mom. a. _- ”3. Se. 853.5
mg 3.8 3.2 m3- at 9;- Es. 8.4- ER. 3”. SN. 5332
3.4 8+ mi. 4.2- Rm. ”3.. mm. _- 92. SN- 3.“. m2- gaseous
8N 5a 92 3.7 m3- of- N3- one ”2. SN- $5- sax .m

 

oo. a? mm. 3. co. w? v? mo. mm. 3. oo.

:oéccow ~58me me 3:23 .EQG mo c\ov €388 €050 A 2an

22

 

.Eomdo co via—RE .mocutecm. Newoeefik 326.2658“: .LS: noocsom

 

v.2 v0.9 mmdm 3.3 m5; 6.; 86¢. 36m 3.3 vav no.3V taxman...
5.: mo“: vmdm 33cm moém NNNN gem mm. mm em. a mmdm 0 fl .vm magnum—Em
gem Emma aoem 5.? v0; OW? mice 5.5“ mimm mmfim «mgm «@332
gm: 2:2 21.8 8.; owdm 3.; 8.; wvdm swam Om.mm 2.0m Ewes—ow:—
om.wm mwém SAN 34% vam 3.5m 3.0m mem wmdm oodm 36m 35% .m

oo. a? mo. 3. 00. mo. v? 8. mm. a. o?

Ego a as as “88:35 .4 03$

23

.Eomdo so 2%sz £65238 Stanza 32626585 ma): Hoohzom

 

hm; mmd was $6 mw.m $6 36 om.m hoe Eh 86 23:22:.
omen :6 :6 56 Saw 2.x cod wms 8.x 2.5 2.3 maxim—Em
mm; Rum BN6 cod omd wmw :um nmd $6 owe SN «@9232
Emm Smom mnfim No.3 v0.0 3.x 00$ cod as ovd 2k. £8289:
RN mad w: med cod 3% vmé amen Nae mod wmfi «80M .m
co. 3. mo. 3. co. 3. v9 ma. mm. a. cm.

88 e289: .m 2an

.EoMAHU co 03283 apamaSm. 382623 3233585 Ht): ”venom

 

 

 

24

 

 

:2 8.8 8.8 $8 8.8 8.8 8.8 8.8 8.8 8.8 8.8 “.588
8.8 8.2 $8: 8.2 8.8 2.: 8.8 8.: 2.8 8.: 8.: 85822.:
8.2. 8:. 8.8 8.8 8.8 8.8 8.8 8.8 8.8 8.8 8.8 58%:
8.8 8.8 8.8 8.8 8.8 8.8 8.8 8.8 :..8 2.8 8.8 gaseous:
8.8 8.8 9.8 8.8 8.8 3.8 8.8 3.8 8.8 8.8 8.8 «23: .m

co. as. ma. 80. ea. ma. ea. 89. 88. _a. co.

Bee .8 8: as 858 a 838
owes 08 83:5: 035 $.88 538w AGO :25 2 0:0 I. .0383: 8 BE 538w n50 ...
.EOMIQU =0 DEN—33w GUZMCBNW NamokBtwK NGtctczkfitN ”as: ”oohzom

.2 88 .- .- 8.8 E. 8:. 8:. 8:. 88 8.8 23:28
91. $8 .- 8.8 h I. 8:. a8 8.: :88 .- 8.8 35823.:
2.8 8.8 .- 8.8 3:. 8:. 8:. 8:. 8.». 8:. 8.8 88%:
2.8 :88 .- 8.0 8.... 8.8 a: 8:. 8.8 :8 8:. 28:85
8.8 8.8 .- 8.8 :8 4:. 8:. Be 88.8 8:. 8.8 «as: .m

cc. am. we. be. ea. ma. 4a. ma. Na. _a. as.

 

038 Base .858 1358805 .m 2an

68% 83 m 8 ooi mat—S 86.. 098.35 :3. 05 .«o 82 5:88an 2:1.

 

Ecw.:aw..cE&..E\\:—:a .SwwhoE .n: ”venom

 

$66 36.6

6.8 8.8 8.8 :8 8.8 8.86: 6.8 _.8 52 8.8 8.8 8.86: 68:2:
2.86 :5

8.8 8.8 8.8 8.8 8.8 3: 6.2: 3: «.8. :6: 86: v.8 85862.:
8.8: 8.6.:

6.8 6.8 6.8 8.8 6.8 8.8: $6: :6: 86: 8.2: $8 :8 28%:
866 8.8

8.8 $6 8.8 v.8 v.8 8.9: 866: 862 8.8: 866: 8.8 88 23.565
8.8.6 .83

N48 5mm 0mm m. K N.mm flaw :5 Bow 8% Now 0N0 NS «802 .m
5. oo. a? mo. 3. 09 mm. v9 3. No. 3. co.
8% Bu» «0 gm .88 owcmaoxo Rom .w 03m..-

.EoM-QU :0 033828 .35.:Sm Numuzufik 32882.82: 53: ”350m

 

62- 8.8- 8.8- 86- 86 8.8 a: 2.8 8.8 8+ 86. 65:9:
8.8- 8.8- 8. _- 666 86 86 2: 9.. _- 62- 6:8- :8- 358:5

86.- 8 _ .8- 8. 7 8.8 86 86 3.8 86 86- 6:8- 68- 28%:

62- m:- 88- $6- 8.: 3.8 86.: $6 3.6- $6 86 «.8865
6: 8...- 8.8- 86 9.6 86 86 $6 86- 8.7 86- 3.3: .m
66. 88. 86. 88. 68. ma. 68. 88. 88. _8. 68.

EDD we .XL 8523 Howvsn EoEEo>o© .N- 28¢

25

flung: Efizom can acumen .5850 ”oosom

 

 

few Swish XE ~ 282$?
fem meoM 95: $3 29332
c\°v anmwfim °\om _ 33:85

£1 again .wa 85x .m
o? co,

 

3:622 .39 mo .5 £82 mummy—£5982 .3 2an

902 can woo. 5953 88.3585:— *... v.02 98 32 5953 BE @3352?

.Eomdu :o Ens—8% £95,235. Ncwuzczi 3.823525 5.): ”oohsom

 

 

5m: A38

as $.02 .3: m3: 3:: 8.8 8.5 5.3 EQ ES 3% 225$
6.3 6.5

8.3 2.; 3.2 $8 3? mm: 8.8 2.3 3.3 3.: 2.2 SEEEE
8va :3

No.5 3.3 mode 3.2: 3.8 33 GE 8: Si 3.: 9m: .2332
9.8 6.3

38 3.8 a? 8% Q? 3% was 8.: 3.3 3.8 $3 38:85
:5: has

8.2 2.2 E: 3% $6 3.3 $8 3% 3.? 5% Ed meoxw

co. oa. ”a, no. ca, mg. «a. mm. mm. _a. ca.

 

EGO 90 fit “Seem 033E 3 $552 gm 6 2an

26

.mocmufim Boa EEme .QUmO ”350m

 

 

 

 

 

Nm.mm 000m 880 0.0: 00.00 Nd: wv00 0v.m0 3mg Tm. Z. mm.m0 052mg“.
3 .0m mmwv 00.00 v.30 3&05 mwdw oo.m0 502 V0: 0N2 0.02V 8:30:30
05.3 m0._m Nv.0m 3.2. w0.ov o0dm vMém _m.mm SAN 3.0— Vm.0_ £93.32
thmw $.00 36w 9%— 0.0: #02 v.00. 5.02 ”N: 002 m0: £82005
mmcnm 506m :wm 0.00m NMON mg: 000m 3:00 N000 m5; mfimn «Box .m
00. 00. w0. N0. 00. m0. #0., m0. N0. 0. 00.
Gateway :wmefi 00 £0 500 8.60-225 .2 203.
.Eomdo co 20263 .85.:80. NSQEVSK $233585 .03: ”venom
00.: 0N: oo.mm mndv V0.0N mnfim Elm wo.: N06 V06 36 0:385.
092 :.0~ 00¢— :xmm 0m.: mw.w 00.0 bvw 00.0 ové w _ .0 8:30:20
00.\. 0N0 Eng 3:.— 0m.: 0N8 m0.w 0o.0~ mnmfi ww.w No.5 «30232
2.0 v00 0mg: NN.: 0m.m m0.m $0 0N0 Sum mm.m 0V0 Emma—005
5.0 wv.0 w0<l 0m.: mm.0 M00 0v.m wvé $6 0w.v 006 «ECM .m
oo. 00. . w0. 00. 00. m0. v0. m0. N0. _0. o0.

 

Eco 0o $0 83% 32:3 230 335m: $55 ._ _ 038

27

.EomA—U :0 Ensign ,BthEm. 38:32.: 3228585 .5): “oohzom

 

 

3.2. :2 835 BE:
2.8 8.3 can.
2:: E: 825
3.2 8.: has
3.x $3 503:0. BE:
23 5% 85E
02: 2.2 5288
8. o9

 

8:258 50 “8282 2928 8 NE mo 2me .3 2an

.EonU :0 0328; .GENEG Namutati 32023225 £3: ”ooSom

 

2; a: am can 03. 3». «ma 2; oz. o; a; 2%.;
mom o; m? 8.0 O? £6 33 o? m? a? 2.2 85%:5
o3 ME. OR a? 3m m2 2% 8a «3 am SN «5332
SM Sm owm Rs 86 2: m3 9; 3m Sm 03 38:85
2:“ Ed m3. one :3 :6 3% So OS 2.» 30 88x .m
8. 8. w? 3. 3. 3. a. a, N? a. 8,

35580 cflm< autumn: eyes.“ 9 NE mo 28% .2 2an

28

8:3 E .5 ..
3285. go owcmzoxm :85 of. new owcmzoxm xoogm 2:92:5—

dwcmsoxm xooum 59:21— m_m:v_ .ewcwzoxm xoowm «:32 .owcmzoxm onm «23. A330: E533. can acumen .5880 ”oohsom

2; SN as ea NR xx 82 82 $3 8” _ K Na Eafifi

S: 3: $8 $2 $2 2:. an 35 8: 82 R: 5 353:5
0% 23 N; 3% am 22 30 § m5 3 0mm 8m 2332
8m 0; 5 gm , 5v 5 2m 8v £3 E E :v 3883
ac 8m 3: New at. Go g 5: 8» E as 0% 85m .m
S. 8, 3. Ma. 3, 8, a. v9 8. No, a, 8.

“.3065 335.81%: xooum .2 2an

29

Table 16. The change of macroeconomic conditions in the Asian countries before the
currency crisis

 

 

 

 

 

Country 1995-96 1997

Thailand - Real GDP growth rate slowed - The government increased its
down to 5.5 % in 1996 from 8.9% in effort to defend collapsing financial
1994. institutions in early 1997.
- Current account deficit worsened - Strong speculative attack on the
from 5.6% of GDP in 1994 to 8.1% bath forced Thailand to let the
of GDP in 1996. currency float on July 2.
- Short-term debt (% of foreign
reserves) increased from 95.98% in
1993-94 to 106.95% in 1995-96.

Indonesia - Real GDP growth rate decreased - The Indonesian rupiah began to
from 15.9% to 8.0% in 1996. come under severe depreciation

pressure with heavily increasing
- Inflation rate remained high at short-term debt in early 1997.
8.0% in 1996.
- Indonesia allows the managed

- Surplus of government budget exchange rate to float on August 14.
balance (% of GDP) decreased from
2.4% in 1995 to 1.3% in 1996.
- The rupiah’s trading band was
widened from 5% to 8% in
September 1996

Malaysia - Current account deficit widened in - Under the severe depreciation

 

1994-95, reaching 8.4% of GDP in
1995.

- Short-term debt (% of foreign
reserves) increased to 40.9%
compared to that of 30.6 % in 1995.

 

pressure caused by lending boom
and current account deficit, the
Malaysian central bank abandoned
the defense of the ringitt on July 14.

 

30

 

 

 

 

 

Country 1995-96 1997
South - Current account deficit widened - Current account deficit was
Korea from 1.5% of GDP in 1994 to 4.8% increasing, export growth was
in 1996. falling, and industrial production
growth rates were below previous
- Short-term debt (% of foreign levels in early 1997.
reserves) increased from 54.1% in
1994 to 203.2% in 1996. - The government announced it
allowed Korean Won to float on
- Growth rate of real GDP decreased December 16.
from 8.9% in 1995 to 7.1% in 1996
- The stock market fall sharply in
1995-96, down by 36% relative to
the 1994 peak
Philippines - Current account deficit continued - Following Thailand, the

 

to be high, 4.8% in 1996.

- Real exchange rate severely
appreciated by 26.2 % relative to
that in 1990.

 

Philippine central bank allowed the
peso to move in a wider range
against the dollar.

 

31

 

CHAPTER III

SURVEY OF LITERATURE AND EXTENDED MODEL

Currency crises in Europe, Mexico and Asia in the 1990’s have drawn global
attention to speculative attacks on government-controlled exchange rates. Research has
proceeded on both theoretical and empirical fronts to find the roots and results of the

crises. This chapter will review the research and extend the currency crisis model for

further research.

1. Theoretical literature
1.1 First generation models

Initial models, now called first generation research, were developed in response to
currency crises in developing countries such as Mexico (1973-82) and Argentina (1978-
81).

The first model came from Krugman(1979). Krugman showed that, under a fixed
exchange rate regime, domestic credit creation caused by the fiscal deficit of government
in excess of money demand growth leads to a gradual loss of reserves. Therefore, this
loss leads to a speculative attack against the currency that forces the abandonment of the
fixed exchange rate and the adoption of a flexible rate regime- a phenomenon known as
the “peso problem”. Because of the nonlinearities involved in his model, however,
Krugman was unable to explicitly derive a solution for the time of collapse in a fixed

exchange rate regime. Later work done by Flood and Garber (1984a) provided an

32

example of how such a solution can be derived in a linear model, with or without
arbitrary speculative behavior.

The basic framework in the following subsection used a simple continuous time.
perfect foresight model. This framework explains the Krugman-Flood-Garber insight
clearly. The model in the framework is a log-linear formulation that allows us to solve
explicitly for the time of occurrence of the crisis, by assuming initially that the exchange

rate is allowed to float pemianently in the postcollapse regime.

1.1.1 Basic framework

In this subsection, Flood and Garber’s model (1984a) is introduced as the basic
framework to analyze the first generation approach to a balance of payments crisis.

Consider a small open economy where residents consume a single, tradable good
whose domestic supply is exogenously fixed at y. The good is perishable and its foreign
currency price is fixed (at unity). Purchasing power parity holds, so that the domestic
price level is equal to the nominal exchange rate. Three assets are available: domestic
money (held by domestic residents only), domestic bonds and foreign bonds. Domestic
and foreign bonds are perfect substitutes. There are no private banks, so that the money
supply is equal to the sum of domestic credit issued by the central bank and the domestic
currency value of foreign reserves held by the central bank, which earn no interest.
Domestic credit is assumed to expand at a constant growth rate. Finally, agents have
perfect foresight.

Formally, the model is defined as follows.

mI—pt =a0—ali,, al>0 (I)

m, =yd, +(1—y)r,, 0<y<1 (2)

d, = ,u, p > 0, (3)
pt : p: ’31 (4)
i, = z" + Es,” . (5)

All variables, except interest rates, are measured in logarithms. m, , p, and i, are

the domestic money stock, price level and interest rate, respectively. d, and r, denote
domestic credit and the domestic government book value of foreign money holdings,
respectively. s, is the spot exchange rate, i.e. the domestic money price of foreign

money. An asterisk (*) indicates “foreign variables”, assumed to be constant. A dot over
a variable denotes a time derivative.

Equation (1) defines real money demand as a negative function of the domestic
interest rate. Equation (2) is a log-linear approximation of the identity linking the money

stock to reserves and domestic credit. Equation (3) assumes that domestic credit grows at
the rate ,u. Purchasing power parity and uncovered interest rate parity are defined in
equations (4) and (5), respectively.
Under perfect foresight, E,s,+1 = s', +,. Setting a0 = it = 0 and substituting (2),
(4), and (5) into (1) yields the following money market equilibrimn condition:
m, =yd,+(1-y)r, =s,—a,.¢,,, . (6)
Equation (6) states that the demand for money can be satisfied either from reserves or

domestic credit.

34

If the exchange rate is fixed at E , then 5', = 0. When domestic credit grows,
reserves, at any time I, adjust to maintain money market equilibrium, according to the
following rule:

r, =[§-7d1]/(1-7)- (7)
The rate of change in reserves is obtained from (3) and (7):

rm =-W9. 9 =(1—y)/y - (8)
Equation (8) shows that if domestic credit grows while money demand remains
unchanged, reserves are run down at a rate proportional to the rate of credit expansion.
Clearly the country will run out of reserves eventually and a fixed exchange rate regime
cannot survive forever.

To find the time of the attack, Flood and Garber introduce the idea of the shadow
exchange rate, which is defined as the floating exchange rate that would prevail if
speculators purchased the remaining government reserves committed to the fixed rate.
After reserves reach the lower bound, the government refrains from foreign exchange
market intervention and allows the exchange rate to float freely and permanently
thereafter. The shadow exchange rate 3 , therefore, is the exchange rate that balances the
money market following an attack in which foreign exchange reserves are exhausted.

Substituting the trial solution '5, = .0 +llm, into equation (6), they find that
[lo =a1yu and 21:1. Thus,

3", = any: + m,. (9)
Equation (9) implies that the shadow exchange rate depreciates steadily and

proportionally to the rate of growth of domestic credit. For simplicity, they assume that

35

 

r,=0 when the fixed exchange rate regime is abandoned. Noting that
d, = do + ,u1 = m, /y , they obtain
3} = 7(do +alfl)+)’/1’- (10)

Figure 1 plots if in equation (9) and the pre-attack fixed exchange rate, E . Let d5
denote the domestic credit level at point A where E“ is equal to 5" , i.e. the two lines
intersect when d = dli.

Suppose that d is smaller than d5. If speculators attack at level d, then the currency
will appreciate and the speculators will experience a capital loss on the reserves they
purchase from the government. Thus, there will be no attack when d < (15. Suppose
instead that d > d5, so 3* >§ . Now there is a capital gain to speculators for every unit of
reserves purchased from the government. Speculators can forecast when that capital gain
will be acquired and compete against each other for the profit. The way they compete in
this framework is to get a jump on each other and attack earlier. Such competition
continues until the attack is driven back in time to the point where d = d5. As a
consequence, arbitrage in the foreign exchange market fixes the exchange rate
immediately after the attack to equal the fixed rate prevailing at the time of the attack.
Exchange rate jumps are ruled out by speculative competition.

The condition that §=§ is used to find both the timing of the attack and the
extent of government reserve holdings at the time of the attack. Substituting 3" for 3*
into equation (10) and 5' = 0 produces the timing of the attack T:

_ E-Ydo
V.“

6’
a1=i—al. (ll)
[1

 

T

36

Equation (11) shows that the higher the initial stock of reserves, R0, or the lower the rate
of credit expansion, the longer it takes before the fixed exchange rate regime collapses.
The (semi-) interest rate elasticity of money demand determines the size of the downward
shift in money demand and reserves when the fixed exchange rate regime collapses and

the nominal interest rate jumps to reflect an expected depreciation of the domestic
currency. The larger a is, the earlier the crisis. Finally, the larger the initial proportion of

domestic credit in the money stock (the higher 7), the sooner the collapse. This result is
due to the fact that Flood and Garber’s model is based on the monetary approach.
According to the approach, a rise in domestic money supply, with demand for money
remaining unchanged would ultimately be offset by an equal and opposite change in the
international reserves through the balance of payments. When reserves run out, the fixed

exchange rate is abandoned.

1.1.2 Extensions to the basic framework
The basic theory of balance of payment crises presented above has been extended
in various directions. I emphasize two models that are strongly related to the empirical
work in the following chapters: One introduces uncertainty into the above context and the
other examines the real effects of a currency crisis in a model with endogenous output,

sticky forward-looking wage contracts, and external trade.‘1

 

4 For details of other major extensions, refer to the survey of first generation models by Agenor, Bhandari
and Flood (1992).

37

Uncertainty and the probability of attack

 

In the basic model developed above, it has been assumed that there is some
binding threshold level, known by all agents, below which foreign reserves are not
allowed to be depleted. The attainment of this critical level implies a regime shift from a
fixed exchange rate regime to a floating rate regime. In practice, however, agents are only
imperfectly informed of central bank policies. They may not exactly know the threshold
level of reserves that triggers the regime shift. If uncertainty about current and future
government policy is prevalent, the assumption of perfect foresight may be improper.

An implication of the perfect foresight model developed above, which is
contradicted empirically, concerns the behavior of the domestic nominal interest rate. In
the model the nominal interest rate stays constant until the moment the attack occurs-at
which point it jumps to a new level consistent with the postcollapse regime. Uncertainty
over the depreciation rate, as modeled below, may help to account for a rising interest rate
in the transition period. Indeed, while specific results are sensitive to arbitrary
specifications regarding distributional assumptions of random terms, only stochastic
models are consistent with the large interest rate fluctuations observed in actual cases.

Uncertainty about domestic credit growth was first introduced by Flood and
Garber (1984a) in a discrete time stochastic model. In their framework, domestic credit is
assumed to depend on a random component. In the basic model, equations (3) and (5) are

modified as follows:

d! =dl—l+#+£! (3),

38

 

. .T - I
1, =1, + E,s,+, —s,. (5)
Variables common to equations (1)-(5) are defined as before, except that I is now

an integer. In equation (3)’. 8, represents a random disturbance in domestic credit

growth. Equation (5)' introduces notation E ,(-), the mathematical expectation operator
conditional on the information set available at time I.
Let E and 3', denote the fixed exchange rate and the shadow exchange rate as

before. In each period, the probability of collapse in the next period is found by
evaluating the probability that domestic credit in the next period will be sufficiently large
to result in a discrete depreciation, should a speculative attack occur. In the Flood-Garber
framework a fixed rate regime will collapse whenever it is profitable to attack it. The

condition for profitable attack is, as in the model developed above, that the postcollapse
exchange rate, 5, , be larger than the prevailing fixed rate, 3" . Profits of speculators are

equal to the exchange rate differential multiplied by the reserve stock used to defend the
fixed rate regime. Since these are risk-free profits earned at an infinite rate (speculators

could always sell foreign exchange back to the central bank at the fixed rate if the attack
is unsuccessful), the system will be attacked if and only if 3', +1 > E . Therefore, the
probability at time t of an attack at time t+1, mm 1, is given by

,7r,+, =pr0b(§,+, >§). (12)
The unconditional expected future exchange rate is a probability-weighted average of
3' and F, +1 .

EtsH-l = [1_tnt+1]§+r7rt+lEt(§r+li§t+1 > §)- (13)

39

:L.“.“A.’n ‘

 

 

Rearranging (13) yields:
Etsl+l -§=I7rt+l[Et(;t+li§(+l > E)" E] (14)
This equation provides the economic intuition of rising interest rates prior to a

crisis.S According to the uncovered interest rate parity condition, the left side in equation

(14) equals the interest rate differential between the domestic and foreign interest rate.
Since the foreign interest rate is assumed to be constant, increases in [E ,s, H — E] would
correspond to higher domestic interest rates.

The expected rate of exchange rate depreciation, [E , s, H —§] , increases prior to
the collapse because both ,71, +1 and [E ,(5, “[3, +1 > E) —§] rise with the approach of the

crisis. The probability of an attack next period ,ir, +1 rises because the increasing value
of the state variable (domestic credit) makes it increasingly likely that an attack will take
place at 1+1. The quantity [E,(§,+ll§,+, > E) — 5] gives the gain that agents may expect
given that there will be a speculative attack at (+1. In turn, that gain depends on the value
agents expect for the state variable next period, given that an attack will occur at (+1. As
the value of the state variable rises from period to period, its conditional expectation also
rises.

The introduction of uncertainty has important implications. First, the transition to
a floating regime is stochastic, rather than certain. The collapse time becomes a random
variable and cannot be determined explicitly, since the timing of a possible future

speculative attack is unknown. Second, there is always a nonzero probability of a

 

5 For an explicit solution, refer to Flood and Garber (l984a).

4O

speculative attack in the next period, which. in turn, produces a forward premium in
foreign exchange markets. Third, the degree of uncertainty about the central banks credit
policy plays an important role in the speed at which reserves of the central bank are
depleted. In the stochastic setting, reserve losses exceed increases in domestic credit
because of a rising probability of regime shift, so that reserve depletion accelerates on the

way to a regime change- a pattern that has often been observed in actual crises.

Real effects of crises

 

The early literature on currency crises emphasized the financial aspects of crises
and overlooked real events that were occurring at the same time. Evidence suggests.
however, that currency crises have been often preceded by large current account deficits
or economic depression.

The real effects of a potential exchange rate crisis have been investigated by Flood
and Hodrick (1986) in economies with sticky prices and contractually predetermined
wages, and by Willman (1988) in the context of a model with endogenous output and
foreign trade. Willman shows that crises are preceded by weak fundamentals such as
economic recession and a current account deficit. A crucial feature of Willman’s model is
the existence of forward-looking wage contracts. Under perfect foresight, an anticipated
future collapse will affect wages, which, in turn, will influence prices, the real exchange
rate, and therefore, output and the trade balance. At the moment the collapse occurs, the
real interest rate falls because of the jump in the rate of depreciation of the exchange rate.
Output therefore increases, while the trade balance deteriorates. But since wage contracts

are forward looking, anticipated future increases in prices are discounted back to the

41

 

present and affect current wages. As a result, prices start adjusting before the collapse
occurs. The steady rise in domestic prices is associated with appreciation in the real
exchange rate and a negative impact on real output.

The continuous loss of competitiveness caused by the real appreciation, unless it
is outweighed by effects from a fall in output, implies that the trade balance deteriorates

in the period before the collapse of the fixed exchange rate regime.

1.2 Second generation models

Newer models, second generation research, are designed to capture features of the
speculative attacks in Europe and in Mexico in the 19905. Second generation models
focus on potentially important nonlinearities in government behavior. They study what
happens when government policy reacts to changes in private behavior or when the
government faces an explicit trade-off between the fixed exchange rate policy and other
objectives such as economic growth, low unemployment, or low inflation rate. Two

examples of second generation research are introduced in this section.

1.2.1 Attack-conditional policy changes

Assume a conditional shift occurs in the growth rate of domestic credit from ,uo
to #1. If there is no attack on the fixed exchange rate, domestic credit grows at the rate
,uO; if there is an attack, domestic credit grows at the faster rate ,u,.

Figure 2 has the same shadow line as Figure 1, but it has an additional line

representing the rate of credit expansion #1. The shadow rate line for y: yo intersects the

42

 

3' line at point E and the shadow rate line for ,u= lit], at point A. d5 and a” indicate the
domestic credit level at points E and A. respectively.
Suppose now that domestic credit lies in the range to the left of d5. If there is no

attack, the shadow rate is on the 3",,“ line. If speculators attack, the shadow rate moves to

the Em line, which is still below the fixed exchange rate. Since any attack leads to

capital losses, there is no incentive for the speculators to attack the fixed exchange rate if
domestic credit is less than d5.

If domestic credit is in the range between d5 and dA, then multiple equilibria could
be possible assuming speculators are small and uncoordinated as a group or face costs in
confronting the government. The economy could reside on the lower shadow rate line
indefinitely if agents believe it is impossible that the market will be attacked. On the
other hand, the economy could jump to the higher shadow rate line if agents are confident
there will be a run. Convinced of a run, no individual agent will find it profitable to hold
domestic currency since this would result in a sure capital loss when the run occurs.
Consequently, all agents will participate in an attack, leading to a collapse of the fixed
rate and a more expansionary credit policy.

Suppose there is a large trader who can take a massive position against the fixed
exchange rate, as George Soros supposedly did against the sterling in 1992. Then, there is
a unique equilibrium. The economy faces only the attack equilibrium since a well-
financed speculator always moves to exploit available profit opportunities. But suppose
there is no large trader in the foreign exchange market, only many small credit-

constrained traders. Without anything to coordinate their expectations and actions, they

43

 

cannot mount an attack of sufficient size to move the economy from the no-attack
equilibrium to the attack equilibrium. Then as suggested in Obstfeld (1986), there are
multiple equilibria. The economy can maintain the fixed exchange rate indefinitely unless
something coordinates expectations and actions to cause an attack.

Morris and Shin (1995) show how some types of uncertainty can eliminate
multiple equilibria and make the attack outcome the unique equilibrium. They describe a
speculative game in which each economic agent obtains information about the state of the
economy, but with a small amount of error. Specifically, if the true state of the economy
is 67 , the agent observes a message that lies in the interval [(7 — 5, L? + .9] , where 5 is a
small positive number. Messages are independent across agents. With noisy differential
information, it is never common knowledge that the fixed exchange rate is sustainable.
Accordingly, each investor should consider the full range of possible beliefs held by
others and should think of what to do if the rate is unsustainable. If there is a good chance
other speculators believe the fixed exchange rate is unsustainable, and if it is not too
costly to take a position against the currency, then it makes sense for the individual
investor to speculate, even knowing the peg is otherwise viable. Holding onto the
currency may yield a bigger gain if everyone else holds on as well, but it is a riskier
course of action because it relies on everyone else behaving similarly. Consequently, the

only equilibrium in the region bounded between a?A and d3 is the attack equilibrium.

44

1.2.2 Escape clause

The second example comes from Obstfeld (1997). The model‘s basic framework
is drawn from Kydland and Prescott (1977) and Barro and Gordon (1983). Here, a
policymaker desires to raise employment above its natural rate through surprise currency
depreciation. The model assumes an open economy and identifies the (log) nominal
exchange rate, 9, (the price of foreign money in terms of domestic money) with the
domestic price level. In this model, devaluations are triggered by the government’s
desire to offset negative output shocks, but a sudden shift in market expectations on the
change in the exchange rate can trigger a devaluation that would not have occurred under
different private expectations.

The government minimizes the loss function
L, = (n, -n$)2 +H(e, —e,_l), (15)
where n, is employment, n" isthe government’s employment target, e, —e,_, is home
inflation, and 9>0. Employment is determined by

n, = ni +x/—(T[(e, —E{e,

 

It—l})"ut—k]' (16)
where l,_, is the information set, u, is an i.i.d mean zero employment shock, and k>0 is

a fixed distortion in the economy that causes employment systematically to fall short of

n*, the target employment level. While labor markets pre-set wages in ignorance of the
realized value of u,, the policymaker is assumed to set the exchange rate after having

observed the shock. In general, the policymaker will want to use the exchange rate to

45

 

offset some of the effect of u, on employment by unexpectedly depreciating the
currency.
There are at least two distinct policymaking processes that might govern

management of the exchange rate. Under discretion authorities choose e, to minimize L,

given E{e,|I,_,}and 21,. The exchange rate change a policymaker chooses under
discretion is

a

+6.

 

_ f I
e, —e,_, — ).,E, e,

 

I,_,,’-e,_,)+}.flc +u,), ,1 s
In addition, rational expectations in the labor market imply an expected loss of

ELD=yE(i-I£k7+k+u)2, yen-Am.

L

Under the other rule, a fixed exchange rate: e, = e,_]. expected loss is

ELF = aE(k + u)2 .

Assume the policymaker faces a personal cost g of revaluing the currency and a
cost 5 of devaluing. Under discretion the policymaker takes the market’s expected
devaluation or revaluation rate, 6(g,17) 6, as given. Then if the choice is to realign, the ex
post social loss is

LD{(5(2.17),u}= rz’0'(y.27)+k +u}2.

and if the fixed exchange rate is maintained, the loss is

LF,’6(g,17),u} = a{(5(g.17) + k + u}2.

 

6 . . — — .
The expected exchange devaluation or revaluation rate when u,> u or u,< y where u and 11 are optimal
policy switch points.

46

Without substantive loss of generality, assume that revaluation is ruled out from
the start: only a large positive realization of u induces discretion, in which case
devaluation occurs. The case of a single equilibrium boundary, 17, accurately depicts
devaluation-prone countries while simplifying the algebra. In addition, to make matters

simple the specific distribution assumed for u is the tent-shaped density function

_ 2 _
g(u)={ (71 MW for uE[ 1L #7 .
0 for u E [ "—71. #7

Since the policymaker’s sole concern is the social-cost differential, LF — LD , her

optimal decision rule is to devalue the currency for u 2 L7 . where 17 is the solution to
117.5 (a). 27} — LDfi)‘ mm} = (a — mm) + k + 17,12 = a.
More simply, interior equilibria correspond to values of ii that solve
6(17)+k+175¢(17)= E/(a—y) E K.

Alternative equilibria are most easily found by changing the shape of the function
(13(5) that emerges as the parameters k and A - which respectively measure the severity of
the time—inconsistency problem and the willingness to accommodate - are varied.

Consider an economy with a relatively large time-inconsistency problem k=0.015
and a non-extreme 11:075. The graph in Figure 3 shows the expected policy loss implied
by different possible switch points u e {—0.03, 0.03] (right-hand vertical axis). The bold

graph shows the (13(17 ) function that arises in this case (left-hand vertical axis). The best
equilibrium is at 17* = 0.0145 (Figure 3), with loss L(z7i) = 0.867 , expected depreciation

6‘ = 0.39% , and a 0.133 chance of devaluation. This equilibrium dominates a fixed rate

47

 

because L(0.03) = l. Imposing the fixed devaluation cost K= <l>(u*) might not suffice to

produce this relatively attractive equilibrium. There are two additional interior equilibria,
associated with the boundaries £7. = —0.0123 and 27" = —0.0256, and with the expected
depreciation rates 6' = 3.0% and 6" = 4.4%. respectively. The implied losses,

14(17): 2.402 and L(z'i") =3.29l. are much higher than that under a pure fixed-rate
regime.

If there is a substantial risk of ending up at a bad equilibrium, then it might be
best to go for an irrevocably fixed exchange rate. This could be achieved by confronting
the policymaker with a prohibitively high devaluation cost on entering a common
currency area. Uncertainty about the <D(u) function would reinforce this conclusion,
since a very small mistake in setting E could open the door to a catastrophe in the form
of an additional, undesirable equilibrium. Unfortunately, imposing the appropriate cost on
the policymaker is necessary, but not sufficient, for reaching a socially preferred
equilibrium. Market expectations can be self-fulfilling, leading in plausible cases to any

number of equilibria, most of which are dominated by the original simple rule.

1.2.3 Other objectives and early evidence

The two examples given above provide the essence of the second generation
research, but cover only a part of the many mechanisms that have been discussed. Any
economic objective that is reasonably part of the government’s social welfare function
and whose attainment involves a trade-off with a fixed exchange rate is a potential

fundamentals for predicting crises. Alternative mechanisms introduced in this section

48

mainly depend on the higher nominal interest rates associated with market expectation of
depreciation.

This mechanism was especially evident in Italy’s 1992 dilemma. Highly indebted
governments with mostly short-term or floating-rate nominal debts will see their fiscal
burden increase sharply if market expectations of depreciation drive up domestic interest
rates. A government chooses depreciation of the domestic currency and/or an increase in
the tax rate to repay its sharply increased debt. The government that wants to minimize
the social loss function under this fiscal constraint will find that a certain degree of
currency depreciation is optimal (Obstfeld, 1994). Therefore, public debt could be one of

the objectives whose achievement includes a trade-off with the fixed exchange rate.

Banks

 

Many financial intermediaries come under pressures when market interest rates
rise unexpectedly. Non-performing loans rise quickly, and depositors withdraw their
funds out of concern over the safety of the banking system. The government’s desire to

sidestep a costly bailout at public expense results in a currency devaluation. (Tomell,

1999)

Real interest rates

 

The analysis in section 1.2.2 allowed for an effect of reduced competitiveness on

output and employment7, but not for real interest rate effects. With sticky domestic prices

 

7 An i.i.d. mean-zero shock, u, in equation (16) can be interpreted as changes in competitiveness.

49

and hikes in the nominal interest rate, the real interest rate rises and adversely affects
employment. A slowdown of the economy may generate self-fulfilling devaluation

pressures.

Pure speculation

 

Crises may simply occur as a consequence of pure speculation against the
currency. For example, models of herding behavior emphasize that information costs
may lead foreign investors to make decisions based on limited information and be more
sensitive to rumors (Calvo and Mendoza, 1997). Contagion effects suggest that a crisis in
one country may raise the odds of a crisis elsewhere by signaling that devaluation is more
likely as a result of the initial crisis. This signal may lead to a self-fulfilling speculative

attack (Masson, 1998).

2. Empirical literature
2.1 Nonstructural empirical analyses

Nonstructural approaches for analyzing crisis episodes in a set of countries before
the 19905 confirm the role of traditional fundamentals in predicting crises. For example,
Klein and Marion (1997) used panel data for 80 devaluation episodes in Latin American
countries during 1957-91 and found the monthly probability of abandoning a pegged
exchange rate increased with real overvaluation and declined with the level of foreign
assets. Structural factors, such as the openness of the economy and its geographical trade
concentration, political variables, such as changes in the executive, and time already

spent on the peg also influenced the monthly probability of ending a fixed exchange rate.

50

The speculative attacks of the 19905, particularly those in Europe, challenged the
view that currency crises were due mainly to the government’s inconsistent policy to
achieve fiscal and monetary discipline. Numerous researchers, motivated by the crises in
the 19905 and the theoretical achievements by the second generation models, have turned
to empirical studies that use an extensive variety of informative variables to distinguish
between periods leading up to currency crises and tranquil periods.

These studies differ from one another in the definition of a crisis, estimation
methodology, explanatory variables included, and data coverage. Three examples are

provided below.

2.1.1 Examples

Example 1

Frankel and Rose (1996) apply a probit model to analyze the determinants of
currency crises using a panel of annual data for 105 developing countries, from 1971
through 1992. Their work is non—structural, and takes the form of univariate graphical
analysis and multivariate statistical analysis. They define a currency crisis as a nominal
depreciation of at least 25 percent in the bilateral exchange rate against the US dollar with
respect to the previous year.

They analyze variables relevant to the theoretical literature. Informed by the first
generation models, they choose the rate of growth of domestic credit, the government
budget as a fraction of GDP, the ratio of reserves to imports, the degree of overvaluation
of the real exchange rate, differentials between domestic and foreign interest rates, and

the current account as a percentage of GDP. In the spirit of the second generation

51

models, they also examine the effects of the growth rate of real output, the ratio of short-
term debt to total external debt and the ratio of external debt to GDP.

Most estimation results from their probit regression are consistent with the
theoretical literature. They find that a high short-term debt ratio (as a share of total
external debt), low international reserves, high domestic credit growth and overvaluation
of the real exchange rate increase the probability of currency crises.

The significance of their results, however, is limited by the lack of robustness to

various sensitivity tests and poor performance in predicting actual crises.8

Example 2

Sachs, Tomell and Velasco (1996) analyze the Mexican crisis and its
reverberations in the financial markets of 20 developing countries in 1995. They attempt
to determine whether there exists some set of fundamentals that help explain the variation
in financial crises across countries or whether the variation just reflects contagion.

They measure the extent of financial crisis in 1995 with a crisis index (denoted
IND) that measures pressures on the foreign exchange market. IND is a weighted
average of the devaluation rate with respect to the US. dollar and the percentage change
in foreign exchange reserves between the end of November 1994 and the end of the first
six months of 1995. The weights given to the loss in reserves and the devaluation are

designed to equalize volatilities of the components of the index.

 

8 Only five of sixty nine actual crises during the sample period are predicted by their model.

52

The rationale for this index is as follows. If capital inflows reverse, the
government can let the exchange rate depreciate. Alternatively, it can defend the
currency by running down reserves or by increasing interest rates. Since there are no
reliable and comparable cross-country interest rate data, they construct the index using
levels of reserves and exchange rates.

Using the crisis index, IND, as a dependent variable, they use an ordinary least
square regression and find that low international reserves relative to broad money, real
exchange rate appreciation, and a weak banking system explain about 70 percent of the
variation of their IND. Cooper (1996), however, has warned against overemphasizing

these results, since they are drawn from a single crisis episode.

Example 3

Kaminsky and Reinhart (1999) examine the links between banking and currency
crises and the stylized facts of the period leading up to and immediately following crises.
Initially they provide a definition and chronology of the crises and their links. Then, they
review the stylized facts around the periods surrounding the crises, while the next step
addresses the issues of the vulnerability of economics around the time of the crisis and
the issue of predictability.

The definition of a crisis is a mix of those in Frankel and Rose (1996) and Sachs,

Tomell and Velasco (1996). They construct an index of currency crisis

53

where Ae/e is the rate of change in the exchange rate, AR/R is that of reserves. 0,, is the
standard deviation of A,,/e , and 0,, is the standard deviation of AR/R . Since changes in

the exchange rate enter with a positive weight and changes in reserves have a negative
weight attached, readings of this index that were three standard deviations or more above
the mean were cataloged as crises.

Using monthly data of 16 macroeconomic and financial variables around the time
of crises, they find evidence suggesting that several economic variables behave quite
differently in tranquil periods as compared to crises period through graphical analysis:
The real exchange rate is overvalued; exports and the terms of trade are deteriorated; and
economic grth is slowed down. The periods just before a crisis are also characterized
by a highly expansionary monetary policy, low reserves and increases in interest rate
differentials. Based on the results, they propose an early warning system to prevent future
currency crises. However, a graphical approach has disadvantages. The graphs are

informal. More importantly, they are intrinsically univariate.

2.2 Structural empirical analyses

Since Krugman (1979), currency crises have been thought to have a significant
predictable component, with first generation models identifying fundamentals useful for
prediction. A fiscal deficit financed by domestic credit creation is considered to be the
root cause of a speculative attack. As the monetary authority monetizes the budget deficit,

oversupply of money causes a gradual decline in international reserves. Accordingly,

54

investors attack the fixed exchange rate. depleting the government’s reserve holdings
used for a defense.

Following empirical studies along this line have focused on Mexico between 1973
and 1982 (Blanco and Garber, 1986), Argentina (Cumby and Van Wijnbergen, 1989), and
Mexico in the l9808 (Goldberg, 1994).

The currency crises in Europe and Mexico in the early 19903, however, rejected
the major role of traditional factors in crises. Otker and Pazarbasioglu (1996) computed
the probability of an exchange rate regime change using Blanco and Garber’s model.
They found the Mexican financial crisis in 1994 was not the result of fiscal imbalances,
which had played a major role in Mexico's previous balance of payments crises; rather, it
was the rise in private sector indebtedness and the corresponding increase in credit to the
banking system that augmented the pressures building up in the exchange market in mid
1994. Moreover, the experiences of several European countries in the context of the
European Monetary System (Otker and Pazarbasioglu, 1997b) show that there are other
triggering determinants of crises - e.g. pure speculation - unexplained by the
fundamentals. As these studies focus on a particular country in a specific time period,

they show the uniqueness of each country’s currency crisis.

2.2.1 Examples

Example 1

Blanco and Garber (1986), in their study of repeated devaluations of the Mexican

peso, present the first empirical study of the speculative attack model.

55

A money market equation provides the essential part of their model:
m, — p, = [3+Qy, —ai, + a), (1)
where m, , p, and y, are the logarithms of the money stock, the domestic price level and
the aggregate output level, respectively; i, is the domestic interest rate; and a), is a

stochastic disturbance to the money demand. Equation (1) represents the demand for real
money balances. They further assume that the price level and the interest rate are

determined by
g=fi+mm—g a)

t
[71: Pr +31 +11: (3)
where an asterisk signifies an exogenous foreign variable, and s, and u, are the

logarithms of the nominal and the real exchange rate, respectively. The operator E
represents expectations conditional on information through time t. Equation (2) reflects
the assumption of uncovered interest rate parity. Equation (3) comes from the definition

of the real exchange rate.
Using the money market clearing condition, the flexible exchange rate can be

determined. Substituting (2) and (3) into (1) yields

in = «255,... +(1+a)33 (4)
where h, E log[D, + Rj—fl—Qy, +ai: —p: —u, —a), , D, is the domestic credit
component of the monetary base at time t and 3', represents the permanently floating

exchange rate. R- represents a lower bound on net reserves where the central bank, having

fixed the exchange rate, stops intervening in the foreign exchange market. Equation (4)

56

says that the floating exchange rates 3*, and E5, +1 are determined by the economic
fundamentals h,.
h, is assumed to be a first order autoregressive process exogenous to the
exchange rate. Specifically, the process h, is
h, = 8, +62h,_, + v, (5)
where v, is a white noise process with a normal density function g(v), with zero mean

and standard deviation 0'.
The flexible exchange rate is obtained by solving the difference equations in (4)

and (5). The solution is
37: : [Hag] + rah! (6)
where ,u =1/[(l+a)—a62].
They assume that the new fixed rate is a simple linear function
.t, = s", + (5v, (7)
where 5 is a nonnegative parameter, and 3‘, is the new fixed exchange rate that will be
established if the level of reserves attains the value R at time I.

3', ’s exceeding the current fixed rate is equivalent to a devaluation at time 1.
Therefore, the probability of devaluation at time 1+] based on information available at t is
pram 1 + #17:“ + 5v,“ > 5),
where 3’ is the time I value of the fixed rate. Alternatively, the devaluation probability is

1-F(kz) EP’(Vt+l>kz) (8)

57

where k, E [1/(,u +6)][§—,ua(7’, —,u(fl, +62h, U, and F(k,) is the cumulative distribution
function associated with g(v).

Knowing this density function, agents can form expectations of future exchange
rates from the average of the current fixed exchange rate and the expected rate conditional
on a devaluation, both weighted by the respective probabilities of occurrence:

Es,+l = m, )r + [1 — F(k, )]E(§,.. |v,,, > k, ). (9)
Using (7), the conditional expectation can be expressed as

Ef§z+1lvz+i > k!) = #61“ + a) + #9217: +01 + (5)501“ l":+l > kt) (10)

where E(v, +1|v, +1 >k,)= 5%“ Since g(v) is a normal density function, the
I

unconditional forecast of the exchange rate for 1+] is

a(,u + (5) exp [ - .5(k, /(702 ]

11

The one-step-ahead devaluation probability (8) and the conditional and unconditional

Es... = mar +[1— F(k,)][#]1(1+ a) + #6th +

 

exchange rate forecasts (10) and (11) are the main products of this model. It is expected
that [1- F (kt)] should reach a peak immediately before a devaluation and Es, ,1 should
be closely correlated with the appropriate forward rates. Finally, the conditional forecast
should approximate the exchange rate when devaluation occurs.

The Mexican crisis over the 1973-1982 is analyzed in this study. Their estimated
probabilities of devaluation in the next quarter, which range from highs of more than 20
percent in late 1976 and late 1981, to lows of less than 5 percent in early 1974 and late

1977, reach local peaks in the period of devaluation and reach local minima in the periods

58

following devaluation as predicted by the theory. Furthermore, the expected exchange
rates conditional on devaluation are close to the values that actually materialized in the

major episodes.

Example 2

Goldberg (1994) applied a discrete time model of a collapsing exchange rate
regime to the experience of Mexico between 1980 and 1986. The model was used to
predict the probability that the existing fixed exchange rate regime would collapse due to
a speculative attack on central bank foreign exchange reserves.

She applied almost the same framework as Blanco and Garber (1986). The model

is provided by equations ( 1 )-( 7) below:

 

M" E. —.v

’ =a0-a.1,+a2Y,—a3[—M] (1)
Q: S!
Q, = aP, +(1—a)s,P." (2)
P .
—,’- = P: + p, + m (3)

I

t ES ’3
1, = 1, +—i—’ (4)
St

Mf = D, +R, (5)
D1 : DI—l +l‘z +8: (6)
8, = y: _¢z (7)

While the variables in the equations do not take logarithms, equation (1) reflects

real money demand and equation (2) defines the aggregate price index as the weighted

59

sum of domestic goods prices P, and traded goods prices 3,1): . The weight
(1 corresponds to the share of domestic goods in consumer expenditure.

By equation (3) the deviation of domestic good’s price from the foreign goods
due to medium-term systematic deviations from PPP, denoted by ,0, , and due to
stochastic shocks to relative prices, denoted by 17, , could be modeled and equation (4)
shows the uncovered interest rate parity.

Equation (5) is the money supply equation. R, represents foreign exchange
reserve and D, is total domestic credit.

The domestic credit component of the money supply is modeled in equation (6) as
evolving according to a trend that reflects the mean basic government budget deficit, 11,,

with some period-by-period stochastic component 8, _

She added a currency substitution impetus.(Es, ,1 —s,)/ 5,, associated with an

expected devaluation of the nominal exchange rate to the demand for real balances. Also,

in equation (7), she decomposed the shock to domestic credit expansion by source: (i)

random revenue or expenditure affecting the need to monetize government deficits, y, ;

and (ii) random and constrained access to external credit. (p, , that makes uncertain the

share of government deficits to be financed by external borrowing instead of inflationary
finance.
Using the money market clearing condition, the period t+1 shadow exchange rate

is derived as

60

 

a,” +a, +613

" _ 7
[ jflm + Dr + R + 71+: “PM

 

at+l

CI:

 

(6)

I+l = a
I I
at+llPt+l + at+l(’7t+l + pr+1)l

 

 

t

where am 2 a0 +021?“ —a,1,+l.

Compared to Blanco and Garber’s model. her framework explicitly shows the
effects of the fundamental variables and parameters on the shadow exchange rate.

She used an ARIMA process to describe the evolution of each variable for which
forecasts are required and instrument variables to avoid simultaneity problems for the
estimation of real money demand. Using an iterative estimation procedure, she re-
estimated estimated parameters to yield new parameter values for estimation of the next
pass estimates of collapse probabilities and expected shadow exchange rates. The
iterative estimation procedure is completed until parameter convergence occurs.

By applying her model to the Mexican currency crisis, she found that Mexico’s

monetary and fiscal policies, rather than anticipated external credit shocks, were the

driving forces in triggering speculative attacks on the Mexican peso in the 19803.

Example 3

Otker and Pazarbasioglu (1997b) evaluated the role of macroeconomic
fundamentals in generating episodes of speculative pressures on six currencies of the
European Exchange Rate Mechanism (ERM) in 1992 and 1993.

The study proceeds in two steps. First, it identifies whether the observed regime

changes can be predicted by the presence of speculative pressures. Second, in order to

61

identity the contribution of deterioration in economic fundamentals to such pressures. it
estimates the probability of a regime change as a function of such fundamentals by using
a monetary model of speculative attacks. The latter outlines a process in which fiscal or
financial imbalances may lead to an eventual collapse of the exchange rate peg by
generating domestic credit expansions that initially cause a gradual erosion of the foreign
exchange reserves. The erosion of reserves is followed by generally self-fulfilling
currency attacks as forward-looking investors engage in one-sided bets, anticipating that
the central bank will exhaust its reserves in defending its currency. Eventually, the peg
can no longer be sustained and the prevailing exchange rate peg collapses. involving
either a discrete devaluation or a switch to flexible rates.

In order to identify the episodes of speculative pressures and the associated
regime changes, they estimate the one-step-ahead probability of a regime change as a

function of pressure indicators below
7:, = Pr0b(Y—1)= 7r[(i, —i:),(E—C),,logR,.AlogR,] (1)

where 7:, denote the one-step-ahead probability of a regime change, Y is the central bank’s
decision regarding a change in its exchange rate regime as a discrete variable which can
take only two values; one, if there is either a devaluation or a switch to flexible rates, and
zero, when, existing parity is maintained, i and i * are short-term interest rates in the
domestic and anchor country, (E — C) is the deviation of the spot rate from the central
parity, R is official foreign reserves, and A is the first difference operator. For Belgium,

Denmark, and Italy, a loss of foreign reserves and increased depreciation of the currency

within the band appear to indicate a build-up of speculative pressures, while for France.

62

Ireland and Spain the existence of pressures appear to be mainly associated with the
depreciation of the spot rate within the band and hikes in domestic interest rates.

They also studied the speculative pressure by the monetary model. If we define
E”, to be a shadow exchange rate at time (+1, the one-step-ahead probability, 27,, of a
regime change at t+1 based on information available at t can then be written as a function

of the prevailing fixed rate and a set of economic fundamentals. h,. that influence the

shadow exchange rate:
7r! E Pr6§+l>§1)E ”(hi'.§1)' ht = h(D,.y,,u,.i, J71) (2)
where D, is the central bank’s domestic credit, y, is real output, i: and p: are the

short-term interest rate and price level of the anchor country, respectively, and u, is the

real exchange rate. For the French franc, the expansion of central bank credit appears to
have contributed to pressure. In addition, the positive coefficient on the unemployment
rate for France and Italy and of the loss of competitiveness for France and Ireland are
consistent with explanations that adverse economic conditions can make it costly for the
government to defend the fixed exchange rate. This market perception may set off

speculative pressures and result in an adjustment in the exchange rate.

3. Extended currency crisis model

To evaluate the influence of macroeconomic fundamentals on the Asian
currencies without the contagion effects, this section introduces basic and extended
currency crisis models using the implications of speculative attack model, first suggested

by Krugman (1979) and formalized by Flood and Garber (1984a).9

3.1 Basic model

A number of small country assumptions are used in setting up the model. The
country described by the model is a small developing open economy. The foreign price
level is taken as an exogenous contributor to the randomness in the purchasing power
relationship. The country also lacks well-developed financial markets. Therefore, its
government cannot engage in open market operations through bond sales. Throughout
this section, the transition or ‘collapse’ studied is one in which a fixed exchange rate
gives way to a flexible rate or a developed new fixed rate. This model applies M2 instead
of M] as a money supply to support the particular aspects of the Asian currency crisis in
which the domestic credit by private banks reflected in MZ played a crucial role. In
particular, previous works used M1 as a money supply to represent the domestic credit of
central bank. However, considering that M2 is an account in the debit of monetary survey
which shows integrated accounts of the central bank and domestic banks and that

domestic credits of central bank and domestic banks are accounts in the asset of monetary

 

9 The survey article by Agenor, Bhandari, and Flood (1992) provides a detailed description of the currency
crises models. Blanco and Garber (1986), Cumby and van Wijnbergen (1989), Goldberg (1994), Otker and
Pazarbasioglu (1996, 1997a, 1997b) find empirical support for the basic currency crisis model.

64

survey as well, we can regard M2 as an monetary aggregate which explains the domestic
credits of central bank and domestic banks at the same time. Therefore, the use of M2 as a
monetary aggregate should contribute to the explanation of how a “Lending Boom” by
domestic banks became one of the major causes in the Asian currency crisis. In addition,
the model allows currency substitution impetus associated with an expected devaluation
of the nominal exchange rate and a risk premium.

The following equations describe the basic model.

(1 .
mt _pt :00 “all: +a2yt —a3(E,s,+, -3! +pt) (l)

i, = i: +(E,s,+, —s,)+p, (2)
p, =p:+s,+u, (3)
m,S = d, +r, (4)
m,d = m,9 (5)
d, = d,_, + E,_,p,d + a? (6)

where m, d, r, p, p*,and y are the logarithms of the money stock, domestic credit
extended by the domestic banks, central bank foreign reserves, domestic price level,
foreign price level, and real output, respectively, i is the domestic interest rate, i * is the
foreign nominal interest rate, ,0 is the risk premium on domestic assets, 5 and u are the
logarithms of the nominal and real exchange rates, respectively. E, represents the
expectation conditional on information available in the current period.

Equation (1) specifies the transaction and asset motives for real money balances.

In addition to the standard variables, real money balances are reduced by the opportunity

65

for currency substitution. The impact of currency substitution on real money balances is

proportional to the sum of expected rate of domestic currency depreciation. E, 5, +1 —— s, ,

and a risk premium. ,1). An increase in E,s, H —s, +p, reflects a decrease in the
desirability of holding the domestic currency.

Equation (2) is the interest parity condition, which states that the interest rate
differential between the domestic and foreign country reflects the expected rate of
depreciation of the domestic currency and a risk premium on domestic assets. The risk
premium for domestic investments reflects the standard increased compensation for more
risky investments in domestic assets.

Equation (3) allows for deviations from purchasing power parity.

Equation (4) defines the money supply as the sum of logarithm of domestic credit
extended by the domestic banks and logarithm of central bank foreign reserve. The
currency crises in the 19703, 19805, and early 19905 were rooted in the dynamics of the
domestic credit extended by the central bank to the government. When there is an excess
of domestic credit creation, a new money market equilibrium can be achieved by a
reduction in the central bank’s foreign exchange reserves or by an exchange rate
adjustment. However, the crucial role of the domestic credit to the government in the
crisis vanished in the Asian currency crisis in 1997. Instead, the domestic credit to the
private sector enhanced by the domestic banks fueled by the foreign liabilities took its
role.

Equation (5) is the money market equilibrium condition. The money market

equilibrium condition determines the path of foreign reserves of the central bank under a

66

fixed exchange rate system. When the reserves needed to maintain this equilibrium are
exhausted, or when they reach a critical level, RC, the exchange rate must be adjusted.

Equation (6) assumes that d evolves according to a period-by-period systematic
stationary component, E ,_, ,u ,‘1 , and a stochastic element 8,“, . In addition, p*, y, i, i*, and
u are assumed to evolve by following the same process as d.

The probability at the end of period t that the fixed exchange rate regime will be
abandoned at the end of t+1 is denoted as 75. Therefore, the probability that the fixed

exchange rate regime will continue is ( l-m). This implies that the expected exchange rate

at H] is
EISHI = 7511513”! +(1— ”()St (7)
where 3', ,1 is the shadow, or floating, exchange rate that would clear the market when the

central bank stops defending its fixed parity.
By combining equations, (1). (2), (3), (5), and (7) and assuming that the fixed

exchange rate regime collapses or 7r, = l,

m, =a0 —a,i: +a2y,+p: +s,+u,—(a1+a3)(E,§',+l—s, +p,). (8)
By following Flood and Garber (1984)’s method of undetermined coefficients
positing without the possibility of a bubble path, the shadow exchange rate expressed as
3', = 110 + 21m, . (9)
Then, by using equation (6) to depict the growth rate of the domestic money supply, the

derived coefficients are

.t # d
40 = ‘00 +011: ‘azJ’z ’Pt ““1 +(al +03)(E:—i#: 41):), and

67

Ill =1
Therefore, by substituting 2.0 and ,1, into equation (9). the shadow exchange rate is

derived as

~ .

at t d
s, = m, —a0 +a11, —azy, -p, —u, +(a, +a3)(E,_,,u, +p,). (10)
We can use equation (10) to derive the probability of a collapse, 7:, , occurring at the end

of period (t+1).'0

3.2 Extended model

As indicated by recent literature, virtually all of the variables in the monetary
model can be expected to be non-stationary. Hence individual economic variables may
wander extensively when shocked. However, previous studies did not explicitly take into
account the non-stationarity of the variables. For example, in equation (I), the coefficient
of the each variable should be estimated first in order to obtain the probability of collapse.
However, OLS estimates of these coefficients will display the spurious regression
problem and the conventional t-ratio and F significance tests cannot be applied.
Therefore, an extension to the basic model is necessary.

As far as the demand for money is concerned as in equation (1), economic theory
suggests that the non-stationary variables in the function are expected to move so that
they do not drift too far apart in the long run. The long run equilibrium can be interpreted

with the concept of cointegration in econometric literature (Engle and Granger, 1987). If

 

w A process of derivation of 71', is explained in section 4.

68

each element of a vector series X, becomes stationary after first-differencing. but there
exists a linear combination a'X, that already is stationary, then the X, are said to be

cointegrated with a cointegrating vectora. By interpreting a'X, as reflecting the long-

run equilibrium, cointegration implies that deviations from the long run equilibrium are
stationary, with a finite variance. This is so even though the series, themselves, are non-
stationary and have infinite variance.

Although economic theory suggested", there is no prior reason to believe that the

1(1) variables observed in this study necessarily obey the functional form in equation (1),

mid — P: = 00 ‘aiiz + 02)”: “03(E25H1 ’ 51+ Pt)-

Hence to avoid an invalid restriction of the real money demand function, the likelihood
ratio tests of cointegration rank of Johansen ( 1988b, 1991), Johansen and Juselius (1990),
and a common stochastic trends test of Stock and Watson (1988) have been performed in
the previous studies. Since Johansen’s (1988b, 1991) maximum likelihood methods for
the analysis of cointegration can simultaneously detect the number of the cointegration
rank in the system, estimate, and test for linear hypotheses about the cointegrating vectors
and their adjustment coefficients. this is the most favored technique in recent research for
the reliable form of real money demand function.

To test for the number of cointegration relations and estimate the cointegrating
vectors require that one begins with a VAR(p) representation expressed in first order

difference and lagged levels,

 

H Refer to Chapter V for the theoretical background of real money demand function.

69

HIIAX, : rle,_l +HI+Fk—1Axl—k+l +Hx,_l +,U() +1], (I l)
where x, is a p-dimensional vector of 1(1) variables,27,,---,277 are 11Np(0,A)and

x_,, H ---x(, are fixed. The 17 matrix conveys the long-run information in the data. The
hypothesis of r cointegrating vectors is formulated as a reduced rank of the 17 matrix,
H2(r):17=a,8’ (12)
where a and ,6 are p x r matrices of full rank.
Under H 2(r) : 17 = (111' , (11) can be interpreted as an error correction model (see

Engel and Granger 1987, and Johansen 1988a). Therefore, equation (1) in the previous

subsection can be modified as an error correction model (ECM),

A(m, — p,) = Arm, = 110 + a(L)Ai,_, + ,B(L)Ay,_, + y(L)Aif,_, + 17,x,_, + 77, (13)
where if, = i, — i: = E,_,p,d + p, , 17, is the first row ofthe 17 , and where
x;_, = (rm,_l , a0, 1, i,_, , y,_,, if,_, ). Then, the shadow exchange rate,

3’. = m, — #0 — a(L)Ai,_1 — mm.-. — 14mm — 17.x.-. — rmH — pf — u, (14).

can be explained as driven by the long-run disequilibrium shocks, short-run shocks, and

random innovative shocks respectively using the error correction model (ECM).

4. The probability of currency crisis

4.1 Explicit form of probability in the basic model
The probability of a currency crisis, 75, is the probability that the shadow
exchange rate, '5, , will exceed 3", , the time I value of the fixed rate, in period (t+1). It is

therefore defined by:

70

’7! = P113,“ "'3?! )0]

= Pl'lmm ‘00 + “iii“ ‘02)?“ — 177+] ”11+: +011 + ”flax/Iii] + Pm) ‘5; WI (15)
The rationale for this formulation is straightforward. Since a government’s
commitment to a fixed rate gives speculators unrestricted access to central bank foreign
reserves, speculators who perceive that the shadow rate will exceed the fixed rate will
purchase reserves at the fixed exchange rate. With the opportunity to resell the reserves at
the higher market rate (equal to the shadow rate), their speculative purchases would yield
a profit of (3*, +1 —§, ) ) 0 per unit of reserves. Barring interim intervention using capital
controls and trade restrictions, which would alter the access to and the speed of decline of
reserves, the speculation may draw the central bank reserves down to their critical
minimum level. In this way, the prediction of a currency crisis can become self-fulfilling.
A forced collapse would not occur at shadow exchange rates below the fixed rate since
there would be no opportunity for profit on the purchase of foreign exchange reserves.
Since the forced currency crisis can only occur when a speculative attack is
capable of driving reserves at or below their minimum or critically low level, (15) can be

rewritten with m, H replaced by d, +1 + r,..
'* t
”I = Prldt+l +rc _a() +6111,“ ‘azym —pt+l _ut+l

+(Ui +03)(Ez#1+1 “HOMO-5; )0] (16)
The probability of a crisis in period ((+1), viewed from period t given the
available information set, is composed of both random influences and components known
with certainty in period t. The random influences come about through period-by-period

stochastic components of d, p*, y, i, i*. and u.

71

Before continuing with the derivation of the probability of collapse, it is necessary
to attach statistical distributions to the random components of d produced by the money
supply and the random components of p*. y. i, i*, and u caused by fluctuations in money
demand. The assumption that only d and u have random influences, that is other money
demand’s randomness is neglected as Goldberg (1994), makes the model excessively
simplified and unrealistic in the rigorous aspects of time series analysis. In addition,
stochastic properties of variables should not be discarded in making forecasts of variables

needed to derive the probability of collapse. To do this, the stochastic parts,

1- t
8,”, ,8,’,£,’ ,8,’ ,6,” and a," are assumed to be uncorrelated with each other and their linear

combination is assumed to be normally distributed. s,lQ,_,~N(0,a,2) where

t t

_ d _i i y p u
8, —8, +(a,+u3)a, —a38, -028, -8, -8, and

t t
2 d2 212 2.12 211.2 )2 11212
a, =(0,) +(a,+a3v) (0,) +a3 (0,) +02 (0;) +(a,’ ) +(a,) .
The normal distribution is chosen for analytical convenience. The systematic stationary

components of the variables are summed into E,_1,u, where

t t

d i i 1
Et—lflt = Et—ll‘t +(al +613 )Et—lfli ‘a3EI—lk‘i —a2Et—l:ut) _ El-llulp ‘ E,_,,u,u.
Then, the probability of devaluation at time 1+1 based on information available at t is
7r! = Pr[d, +r,. "a0 +(al +a3)ir ‘03’7 'azy: _pt "ut “PEI/0+1 +81+l ”E1 )0] (17)

For ease of notation, define a variable k, as

 

12 . . I i " ° .
The covariance terms are removed smce 8,‘ ,8, ,8,’ ,£,v,8,” and 8," are assumed to be uncorrelated With
each other.

72

— , l 0‘ ‘
k, = s, —d, —r,. +00 -(a, +a3 )1, +a3l, +a2y, +p, +u, — E,,u,+,.

Subsequently. the it, can be rearranged as

2
(I
___Ltl__

7r, =Pr[8,+,)k,]= f 1 e 215’”7+1de. (18)

where E,a,2+, is a forecasted volatility of a, +1 .

The probability of devaluation in equation (17) is expected to be positively
influenced by the previous level of domestic credit and the interest rate. As the foreign
interest rate, GDP, foreign price, and real exchange rate’s past levels rise, the probability
of devaluation is predicted to decline. Alternatively, an increase in the floor level of

reserves. rc, and systematic stationary components of domestic credit and the interest rate

increase the probability of a devaluation.

4.2 Explicit form of probability in the extended model

The probability of a currency crisis, it, , in the extended model has the same
validation as the basic model. Therefore, the probability is
n. = Pri's‘m — s > 01
= Pr[d,+1 +rc —u0 “07041) ‘WLMy; —7(L)Aifz—171xt“rmz ’P:+1-ur+1 ‘5: )0] -(19)
As before, the randomness comes about through period-by-period stochastic components
of d, p*, and u.

Suppose the systematic stationary component of each variable and the stochastic

part of the processes (1, p', and u are justified in the same way as was the case in the basic

73

model. Then, the probability of devaluation at time {+1 based on information available at
t is,
7:, = Pr[d, + r, — u, — (1mm, — mm, — may, — 171x, — rm, — p,‘ — u,
+E,,u,+,+£,+, —§, )0] (20)
Once more, if we define k, as

k, = ', —d, —rc +u,) +a(1.)Ai, +[1(L)Ay,+y(L)Ai/, +17lx, +rm, + p: +u, — E,,u,+,,

c».

then it, can be rearranged as,
.2
__“!_L

2
J 1 25”!“ d8. (21)

7t =Pr 8 k = ————e
1 [(+1) I] EEIUHIJz—fi

74

C01

Figure 1. Attack time in a certainty model

When domestic credit 0' is less than dE , there is no attack; when d> d5 , speculators attack the

currency.

75

 

 

 

Figure 2. Attack times with attack-conditional policy shift

When d <dE , there is no attack; when dE <d<d4. multiple equilibria are possible; when d >d«4

speculators attack the currency.

76

Devaluation cost, K POIICY loss

    

 

 

 

0.046 ., 3.5
0.044 + ,
. j 3
0.042
0.04 2.5
0.038 1
- 2
0.036
; 1.5
0.034 a
0.032 I I I 1
| I H
0.03 | | I
| l | 0.5
0.028 + , , I +
07 07' MT .
0.026 -. _.—-—— ..—...—-—~ .- . .——.~———#—.~-_—-+— —— __-__._-. 0
-0.03 -0.02 -0.01 0 0.01 0.02 0.03

Figure 3. Devaluation cost and policy loss

77

CHAPTER IV
TIME SERIES PROPERTIES OF VARIABLES IN CURRENCY

CRISIS MODEL

1. Introduction
Application of the currency crisis model to the Asian experience in 1997-98
requires the analysis of the time series properties and the making of forecasts for the
variables, p", y, i, 1'“, d, and u to derive a shadow exchange rate and an one-step-ahead
probability of currency crisis.
However, most of the previous studies about the structural analysis of currency
crisis, Blanco and Garber (1986), Cumby and Van Wijnbergen (1989), Otker and
Pazarbasioglu (1996, 1997b) do not estimate the properties of variables in the model.
Instead, they assume an AR(p) model. Unlike those studies, Goldberg (1994) uses
ARIMA models and applies Akaike tests to determine which ARIMA process should be
used to describe the variables for which forecasts are required. However, while ARIMA
models capture autocorrelations that decay at an exponential rate as associated with
Stationary and invertible ARMA(p,q) models of the first differences of stochastic
processes, ARIMA models could not be applied to long memory processes where the
alJtocorrelations decay more slowly than the exponential rate.
Therefore, the first objective of this chapter is to explain the time series properties

Qf macroeconomic variables in our model that exhibit long memory in both their

QQnditional mean and variances. To this end, the ARFIMA(p.d,q)-FIGARCH(P,6,Q)

78

 

models are added to the families of models to be selected from by the Wald tests. By
making this inclusion, we detect that some processes exhibit dual long memory behavior.
The second objective of this chapter is to forecast of the one-step-ahead levels and

volatilities of d, p*, y. i, i*, and u which will be used for the derivation of the probability

of collapse. Since the ARFIMA(p,d,q)-FIGARCH(P,(2Q) model lets the stochastic

part,e,,of each variable be assumed to have a distribution of £,|[2,_,~N(0.0,2), the

expected conditional variance, E ,0 ,2“. as well as the expected conditional mean, E ,,u, H ,
can be obtained to forecast the levels and volatilities of all of the relevant variables13 . To
this end, we estimate the coefficient of each variable in ARFIMA(p,d,q)-
FIGARCH(P,6,Q) model from the analysis of the time series properties of the variables.
Then, the estimated coefficients are used to calculate the expected conditional mean and
variance.

In this chapter, Wald tests are initially applied to select the appropriate model
explaining the behaviors of the variables. Then, once the specific form of the model is
determined, the coefficients are estimated and the model is fit over each of the sample

time periods to calculate one-step- ahead forecasts.

2. Data set
Monthly data are collected from 1970101 through 2000:12 for Indonesia, South

I<0rea, Malaysia, Philippines, and Thailand. All of these countries mainly experienced the

Please see Chapter III for the details.

79

currency crisis from 1997 to 1998. All of the variables are taken from the CD-ROM
version of the International Monetary Funds International Financial Statistics (IFS).l4
Table 17 reports the description and sources of the data. The sample size is dictated by
the availability of data on the variables.

The analysis in this chapter differs from previous empirical studies[5 in the use of
longer sample time period than preceding studies. In addition. monthly data is expected to
capture all the variation in some of the variables for both the months before and after the
collapse. A careful identification of all of the activities is crucial for the analysis of the
Asian crisis since less frequent data could hide the rapid movements in the second half of

1997.

3. Analysis of time series property and forecast of variables
3.1 ARFIMA-FIGARCH model
Several recent articles have discussed the property of long memory in either the
conditional mean or variance of a process.
Granger and Joyeux (1980), Granger (1980, 1981), and Hosking (1981)
introduced discrete time representations of fractional Brownian motion known as
ARFIMA(p,d,q) processes, which combine the stationary and invertible ARMA model

With the fractional difference operator. The model is.

¢(L)(1- Mo», #1) = We. (1)

\
I

Monthly data for GDP are missing in Indonesia, Thailand, and Philippines. For those countries,
9 lJarterly and annual data are used for the analysis.

80

where. d is a fractional differencing parameter; and d)(L) = 1 — (DIL — — (1),, L” ,
6(L) = l + 6,1. + + (9,,L‘I , and all the roots of (ML) and 6(L) lie outside the unit circle for

stationarity and invertibility; and E03,) = 0, E(e,2) = azand E(8,es) = 0 for s at t. The
Wold decomposition, or infinite order moving average representation of this process is

given by y, = Zy/je,_j ; and the infinite order autoregressive representation is given
j=(),oo

by y, = Z7:,y,_j +s,. For high lag j, these coefficients decay at a very slow
j=l,00

hyperbolic rate, i.e. y], zc,_]'d"l and 7,, zczj—d-I

J , where c, and c2 are constants.

For— 0.5 < a’ < 0.5 . the process is stationary and invertible and y, is said to be
fractionally integrated of order d, or l(a’). Therefore, the parameter d represents the degree
of “long memory” behavior for the series. For 0.5 S d < 1.0, the process does not have a
finite variance, but for d < 1.0 the impulse response weights are finite, which implies that
shocks to the level of the series are mean reverting.

Time dependent heteroskedasticity in conditional variance is a well-known feature

of many asset pricing series and also it is considered useful for some macroeconomic

Series. Usually the ARCH model of Engle (1982) is introduced as
8, = 2,0, (2)
Where E,_,z, = 0 and Var,_,z, =1 . Throughout this chapter, E,_, and Var,_, refer to the

Conditional expectation and variance with respect to this same information set. Thus, by

l
5 See appendix 2 for the summary of previous studies.

81

 

definition, the {8, ,’ process is serially uncorrelated with mean zero. However. the

conditional variance of the process, 0,2 is a time-varying. positive and measurable
function of the information set at time t-1. A useful extension to the ARCH model is the
GARCH(P,Q) specification of Bollerslev (1986). This model is defined by

a? = co + a<L>ef.. + anvil (3)

where, L denotes the lag operator; and a(L)Ea,L+a2L+-~+a0LQ and

~

,B(L)zfl,L+fl2L+-~+,BPLP. In this model, one can guarantee the stability and
covariance stationarity of the {5,} process by assuming that all the roots of
{1—a(L) —,B(L)} and {1 — /)’(L)} are constrained to lie outside the unit circle. Otherwise,

the GARCH(P,Q) model in equation (3) may also be expressed as an ARMA(M,P)

. ‘ 2
process in a , ,

{1— arm—11am?- = w+{1-fl(L)}v, (4)
where M E max{P, Q} and v, a 8,2 -a,2 is mean zero serially uncorrelated. This model
has been useful in describing many volatility processes. However, many high frequency
asset pricing series have very persistent volatility. In other words, the autoregressive lag
polynomial, {1- a(L) — ,B(L)}, has a root that is very close or even indistinguishable from
llnity. This led to the integrated GARCH model or IGARCH model of Engle and
Bollerslev (1986). The IGARCH(P,Q) process is defined succinctly by

man — Us? = w+{1—/3(L)}v, (5)

Where, (ML) 2:— {l — a(L) — [1(L)}(1-— L)‘I is of order M—I.

82

 

Baillie, Bollerslev and Mikkelsen (1996) introduced the FIGARCH process to
model very slow hyperbolic decay in terms of how a shock affects the conditional

variance process. The FIGARCH process has the benefit of being a midpoint between the
extremes of stationary GARCH and IGARCH. The F IGARCH(P,5.Q) process is naturally
given by,

{1— MUM? = w +{1— 1M)- wan — 1mg} (6)
where 0 < 0‘ < 1, all the roots of (ML) and {l - B(L)} lie outside the unit circle. The

FIGARCH process can also be rewritten as,

a? = curl —/m)r' +{1—[1—ti(L)]“w(L)a—L)"}ef <7)
The FIGARCH(P,§,Q) model nests the covariance stationary GARCH(P,Q) for 6 = Oand
IGARCH(P,Q) model for 0‘ =1. Allowing for 0<c5 <1 gives far more flexibility in
modeling long term dependence in the conditional variance. This process implies a slow

hyperbolic rate of decay for lagged squared innovations. For instance, the

FIGARCH(1,6,0) process can be expressed as
2 _ -l 2
a, — a)(l — ,8) + 1(L)£, (8)
where 2,, = F(k +6 —1)/{F/{17(k))}.{(1—/i)—(1—a‘)/k}, and for large lags k,
11,, = ,(1— ,8)/F((5),' k‘H , which generates slow hyperbolic rate of decay on the impulse

response weights. The process is strictly stationary and ergodic for 0 S 5 S l , and shocks

Will have no permanent effect. It is also worth noting that the FIGARCH model can be

expressed as an ARFIMA(P,6,Q) model in 8,2.

¢(L)(1- me} = w +{1— mm, (9)

83

 

where, 0 <6 <1. and all the roots of (ML) and {1—/)’(L)} lie outside the unit circle.

v, = (3,2 - 0,2 is mean zero serially uncorrelated.
To capture the property of economic and financial time series that exhibit long

memory in both their conditional mean and variances, the ARFIMA(p.d.q)-
FlGARCH(P,6,Q) model is applied following the result of the Wald tests. The model is
given by

WL)(1—L)d(y,-/U= was, (10)

a, =z,a, (11)

a? = wtl wan“ +{1-[1-13(L)]"<0(L)('1— L)"}e? (12)
where (ML) , 6(1). [1(L), and (ML) are as defined earlier in equation (I), (3), and (5). As
noted above, the pure ARFIMA(p,d,q)-homoskedastic process will have a finite variance
for — 0.5 < d < 0.5. However, ARFIMA-FIGARCH process will have an infinite
unconditional variance for all d given a at 0. This fact is discussed in the context of the

pure FIGARCH model by Baillie, Bollerslev, and Mikkelsen (1996).

Assuming conditional normality, the logarithm of the likelihood can be expressed

in the time domain as
. _ , 2 2 2
L(A’EI’EZ'W’ET)——(T/2)ln(2”)_(1/2)Zr=|,rl ln(a, )+6,a,} (13)

Where 11'=(p,(D,,---,d>p,0,,---,6q,,6,,---,,Bp,(p,,---,(pQ,d,co,(5). The QMLE of the
parameters are estimated by a similar methodology to that described by Baillie,

Bollerslev and Mikkelsen (1996), where the likelihood function is maximized conditional

on initial conditions and the pre-sample values of 8,2,! = 0, —1, —2. are fixed at the

84

sample unconditional variance. The initial observations yo. y_,, y_2. are also
assumed fixed, in which case minimizing the conditional sum of squares function will be
asymptotically equivalent to MLE.

After the specific form of process is selected and the parameters are estimated. the
process is fit over each of the samples contained in the entire interval. Following Baillie

and Bollerslev (1992). the minimum MSE predictors using the expected conditional

means, E, ,u, +, , and conditional variances, E,a,2+, , are applied to obtain one-step-ahead

. . . . . .*
forecasts of levels and volatilities of variables. d, 12*. y. r, I , and u, where

t i

d - ' ' ' )
EtIuH-l : Erflr+l +(a! +a3)E,,u,'+, -a3E,,u,l+1 —a2E,,u,"+, _Ettu1,+| -‘ Et/ltufl and
5103+! = EJ011112 +(ar+a3)251(0i+r)2 +032Ezf<7i+112

+ azzE,(a,-"+, )2 + EMU/f, )2 + E,(a,”,,)2 '6

3.2 Empirical results for US inflation

Table 18 reports the estimation of various univariate models to represent the US
CPI inflation data described earlier. The final column of Table 18 estimates an
ARFIMA(0,d,1) model with d estimated at around 0.42. The autocorrelations of the
standardized residuals are relatively smaller than other models but the squared residuals
exhibit autocorrelation that is consistent with very persistent ARCH effects. Figure 4
represents the long run ARCH effects in the squared residuals. In order to allow for

ARCH effects, a range of volatility models conditional on the selection of specification

85

for the conditional variance process is considered. The conditional variance processes

considered are the FIGARCH(1,§,1). IGARCH(1,1). and GARCH(1,1) models. The
estimates from these models are reported in column 1. 2, and 3 of Table 18. As for the
first column, the estimated long memory conditional mean. d, is around 0.43, and
significantly different from zero or one. In addition, robust Wald tests can reject the

hypothesis that the long memory conditional variance (5 = 0 with destimated around

0.67. Therefore, we concluded that the estimated ARFIMA(0,d,1)-FIGARCH(1,6,1)
model is the most appropriate specification for accounting for the dynamics of the
conditional mean and variance.

Figure 5 provides visual evidence that the persistent ARCH effect of the process

vanished after the estimation by the ARFIMA(0,d,1)-FIGARCH(1,5,1) model. In

addition, the Q2(20) of ARFIMA model, 622.15, significantly decreased to 15.91. This

also could be an another confirmation that ARFIMA(0.d.1)-FIGARCH( 1,6,1) model is

valuable for eliminating the persistent ARCH effect.

Figure 6 illustrates how well the ARFIMA(0,d,1)-F IGARCH(1,6,1) model
explain the US CPI inflation data. Even though some extreme values are not fitted well
by the model, the ARCH effect of the process is effectively demonstrated. Figure 7 shows
a one-step-ahead forecast using the model estimated with sample period ending in 1996.
It is observed that some of the extreme values are not forecasted appropriately, as are
some of the fitted values, but the trend of forecasted values obviously follows that of the

actual values of the US CPI inflation data.

 

16 See Chapter III for the details

86

3.3 Empirical results for deviations from PPP

Table 19 contains the estimation of assorted univariate models to describe the
deviations from PPP of Indonesia, South Korea, Malaysia, Philippines, and Thailand.
Percentage changes for the deviations from PPP are used for the y, in the model. This
value explains the volatility of deviations more successfully than other values when MLE
estimators are chosen to estimate the parameters in the model. In addition, using
percentage change for the y, does not influence significantly any of the forecasted

values.

The first column of Table I9 estimates an ARFIMA(0.d.0)-FIGARCH(1.cXO)
model with d estimated around 0.13 and 6 assessed around 0.30 for Indonesia. Wald tests
significantly reject the hypotheses that (5 = 0 , indicating strong evidence of long memory
in the conditional variance as well as the conditional mean. Figure 8 provides the
autocorrelations of the standardized residuals and squared residuals after the estimation.
The long memory in conditional mean and variance are noticeably captured. This is
shown both in Figure 8 and by how the @720) estimated around 26.30 could not reject
the null hypothesis that there is no autocorrelation in the squared residuals. The rest of
columns in Table 19, however, do not include long memory in the conditional mean or in
the conditional variance. The relatively small sample size, 324, makes it impossible to
estimate long memory for the South Korea, Malaysia, Philippines, and Thailand. Despite
the size of the data sets, the relative absence of a long memory property in the conditional

mean and variance does not significantly influence the ability of estimation to eliminate

87

autocorrelations from the process. The Q(2()) and Q3(2()) of each country‘s estimation
explain that the autocorrelation of the standardized residuals and squared residuals are
sufficiently captured by the estimation of the model. Figures 9 tol2 offer graphical

confirmations of the statistics mentioned above.

3.4 Empirical results for domestic credit

The growth of domestic credit, one of the main causes in the currency crisis,

shows various aspects in the estimation of ARFIMA(p,d,q)-FIGARCH(P,6,Q) models for

the domestic credit of each country: Indonesia, South Korea, Malaysia, Philippines, and

Thailand. Percentage change of domestic credit is chosen for the y, .

Table 20 offers the result of the estimation. ARFIMA(0,d,0)-FIGARCH(1,(21)
and ARFIMA(1,d.0)-FIGARCH( 1.5.1) are chosen for the estimation of Malaysia and
Thailand based on the Wald tests. Malaysian percentage change of domestic credit has an
estimated d of 0.06 and an estimated 6 of 0.52. For Thailand, the estimated long memory
conditional mean parameter, (I, is —0.06 and the estimated long memory volatility

parameter, 6, is 0.71. Indonesia, Philippines, and South Korea, however, do not show any
long memory property in the conditional mean or variance. Failure to find evidence for
long memory property could be because of the small size of the data set. The
autocorrelations of residuals presented in Figures 13-17 indicate that the estimated

models did reduce the autocorrelations of the process.

3.5 Empirical results for interest rate

88

ARFIMA(p.d.q)-FIGARCH(P.5.Q) models are then estimated for the changes of

interest rate in Indonesia. South Korea. Malaysia. Philippines. Thailand. and the US as
shown in Table 21. The change of interest rate is chosen for the y, based on the same

criterion used above.

For Indonesia, South Korea, Malaysia, Philippines and US. Wald tests detect a
long memory conditional mean parameter, d, in the process. However, for South Korea,
Malaysia and Philippines, the estimated (1 lies in the range of —— 0.5 < a' < 0. Therefore,
the processes have ‘intermediate memory’, and all their autocorrelations. excluding lag
zero, are negative and decay hyperbolically to zero. Whereas Wald tests can detect ‘a" in
all country’s processes apart from Thailand’s, they do not indicate any strong evidence of
long memory in the conditional variance for all countries. For Malaysia, Philippines,
Thailand and US, the changes of interest rate show a highly persistent volatility shocks
represented by IGARCH models. For Indonesia and South Korea, ARFIMA with
homoskedasticity models are more representative of the changes of interest rate. As
shown by Figures 18 to 23, the autocorrelations of residuals imply that the estimated

models do eliminate the autocorrelations of the process for all the countries considered.

3.6 Empirical results for real GDP

ARFIMA(p,d,q)-FIGARCH(P,6,0) models are then estimated for real GDP of

Indonesia, South Korea, Malaysia, Philippines, and Thailand as shown in Table 22. The

89

rate of change of real GDP'7 is selected for the y, and monthly. quarterly. or annual data
are used to the extent that the highest frequency data are available.

The third column of Table 22 estimates an ARFIMA(I,d.0)-FIGARCH(I,6.1)
model with d and 6 estimated around -0.32 and 0.53 for Malaysia. Wald tests
significantly reject the hypotheses that (5 = 0, indicating strong evidence of long memory
in the conditional variance. Figure 26 provides the graphical evidence that the
autocorrelations of the standardized residuals and squared residuals are reduced after the
estimation for Malaysia. Similarly, the Q(20) and @720) estimated around 26.94 and
14.02 indicate that the autocorrelation of the standardized residuals and squared residuals
are sufficiently captured by the estimation of the model as well.

The rest of columns in Table 22, however, do not contain any long memory in the
conditional mean or variance. While the processes of Indonesia and Thailand do not show
any long memory for both the conditional mean and variance, Wald tests reveal a long
memory conditional mean parameter, d, for both South Korean and the Philippines. The
autocorrelations of residuals presented in Figures 24-25 and 27-28 indicate that the

estimated models did reduce the autocorrelations of the process.

4. Conclusion
This chapter analyzes the time series properties of the macroeconomic variables in
Indonesia, South Korea, Malaysia, Philippines. and Thailand which mainly experienced

the currency crisis in 1997-98.

 

'7 mo of the change rate is applied in case of Indonesia, Thailand, and Philippines.

90

As given in the above sections. some hybrid ARFIMA-FIGARCH models are
estimated for the macroeconomic variables. Interestingly, the Wald tests support the
conclusion that the US. inflation rate, the percentage changes of deviations from PPP in
Indonesia, the percentage changes of domestic credits in Malaysia and Thailand and the
change rates of real GDP in Malaysia appear to have both estimated long memory
parameters d and (5 which lie in the range of —0.5 < d < 0.5 and 0 < (5 <1.0,
respectively. However, as shown in Tables 15 to 19, Wald tests do not indicate any dual
long memory behavior in other processes.

Following the analysis of time series properties, the expected conditional mean,

E, a , +1 , and conditional variance, E ,0 ,2“ , obtained using the estimated coefficients in the
analysis, are applied for the forecast of levels and volatilities of variables. In particular,

the forecasted levels of d. i, i", y, p‘ and u'8 are substituted for the
(1H, , i,+, , i3, , y,+, , p1,, . and um to obtain the shadow exchange rate,

30.1 = dt+l +rc ’00 +(a, +0310“ —a3i,:., 'azym “Pin ‘um "5(-
In addition, the forecasted volatilities, E,0,2+,, of d, i, i", y. pi and u are used for the

derivation of probability of collapse,

2

_ C
f l 8 215,0
’ Eto't+l \[2—7r

 

2

2r. = Pris... > k. 1 = M de.

 

'8 d, i, 1", y, p7 and u are assumed to be uncorrelated with each other.

91

Table 17. Data sources and definitions

 

 

 

 

 

 

 

 

Variables Definition Sources
d, Log of the M2 minus log of the central bank total lFS(linel l.d,14.a,14.b),ln-
reserves ternational Monetary Fund
r, Log of the central bank total reserves minus gold in lFS(linell.d), International
millions of US dollars Monetary Fund
i, Money market ra? lFS(line60b,600),
lntemational Monetary
Fund
i,* Treasury bill rate for the United StatesW lFS(line60c), lntemational
Monetary Fund
p, Log of the consumer price index IF S(line64), lntemational
Monetary Fund
p,* Log of the US. consumer price index lFS(line64), lntemational
Monetary Fund
y, Log of the nominal gross domestic product deflated lFS(line99b), lntemational

. . . *** . .
usrng consumer price index in domestic currency Monetary Fund

 

5, Log of the end-of-period spot exchange rate lFS(line60c), lntemational

Monetary Fund

 

Notes. * For Indonesia, monthly data from 1983:01 through 2000: 10 are collected. IFS defines the maturity
of the money market rate to be short-term. For South Korea, monthly data from 1976:08 through
2000:11are collected. The maturity of the money market rate is defined to be short-term. For Malaysia,
monthly data from 1970:01 through 2000:11 are collected. The maturity of money market rate is not
defined. For Philippines, monthly data from 1976201 through 2000:12 are collected. The maturity of
treasury bill rate is defined to be 91 days. For Thailand, monthly data from 1976201 through 2000:12 are
collected. The maturity of money market rate is not defined.

** For US, monthly data from 1970:01 through 2000:12 are collected. The maturity of treasury bill rate is
defined to be 91 days.

**"‘ For Indonesia, annual data from 1970 through 1998 are collected. For South Korea, monthly data of
log of the industrial production index from 1970:01 through 2000:] 1 are collected. For Malaysia, monthly
data of log of the industrial production index from 1971:01 through 2000:11 are collected. For Philippines,
quarterly data from 1981: 10 through 2000: BQ are collected. For Thailand, annual data from 1970 through
1998 are collected.

92

Table 18. Estimated ARFIMA(p,d,q)-FIGARCH(P.(1Q) models for US monthly inflation
rate

(1 — L)d(y, — )1) = (1 +(1L)(1+ 620%,,
a,|.Q,__,~N(0,a,2 ),
(1 - pug} = a) + [1 — (1L — 101(1— LX518}

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(p.d.q)- (0.d.1)- (0.11.1)- (0.0.1)- (0,d,l)-
(P,8,Q) (1.5.1) (1.1.0) (1.0.1) (0,0,0)
,1: 0.5477 0.5697 0.5704 0.3848
(0.2925) (0.4864) (0.5727) (0.2666)
d 0.4324 0.4573 0.4612 0.4222
(0.1210) (0.1791) (0.2055) (0.0920)
9 -0.2250 -0.2501 -0.2496 -0.1754
(0.1333) (0.2005) (0.2266) (0.1275)
9 0.1295 0.1451 0.1449 0.0428
(0.0365) (0.0362) (0.0383) (0.0591)

0.6696 1.0000 - -
(0-3390) ( - ) ( - ) ( - )
a) 0.0012 0.0019 0.0038 0.1054
(0.0012) (0.0016) (0.0022) (0.0107)

fl 0.8085 0.8499 0.8343 -
(0.1291) (0.0684) (0.0685) ( - )

‘P 0.4416 - 0.9548 -
(0.1254) ( - ) (0.0224) ( - )
LL -79.942 -86.929 -83.494 -176.l90
Q(20) 34.0424 33.2755 32.7610 30.8235
Q3(20) 15.9059 17.9971 19.5660 622.1546
Skewness 0.105 0.164 0.175 0.137
Kurtosis 4.419 4.611 4.456 7.150
Wd=1 21.989 9.186 6.876 39.397

"25:0 3.902 ’ ‘ '

 

Key: The table reports the Quasi Maximum Likelihood Estimates (QMLE) for various ARFIMA-
FIGARCH models. The QMLE are calculated using the normal likelihood function. Robust standard errors
are reported in parentheses. LL is the value of the maximized Gaussian log likelihood; and Q(20) and
02(20) are the Ljung—Box test statistics with 20 degrees of freedom based on the standardized residuals and

squared standardized residuals respectively. Wd=1 and W5=0 are both Wald test statistics for testing long
memory property in the conditional mean and variance.

93

Table 19. Estimated ARFIMA(p,d,q)-FIGARCH(P,(1Q) models for deviations from PPP
(1— 491.7(1- ado». — u) = (1+ ()2).,,
8.19.-.~N(0.a.2).
(1— (2’00? = «2+ [1— (IL — (ow — 1.7074."

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Country Indonesia South Korea Malaysia Philippines Thailand
(p,d,q)- (0,d,0)- (0,d,1)- (0,d,0)- (1,0,1)- (0,0,1 )-
(P,5,Q) (1,5,0) (1.0.1) (1.0.1) (1.5.0) (1.0.!)
,u 0.1266 0.1231 -0.1409 0.1080 0.0588
(0.1278) (0.2069) (0.2244) (0.1544) (0.0522)

61 0.1305 0.2458 0.1461 - -
(0.0488) (0.0589) (0.0667) ( - ) ( - )

(b - - - 0.7937 -
( - ) ( - ) ( - ) (02515) ( - )
,9 - 0.1798 - -0.6902 0.1215
( - ) (0.0729) ( - ) (0.2088) (0.0580)

5 0.2977 - - 0.7421 -
(0.1025) ( - ) ( - ) (0.1316) ( - )
a) 0.1460 0.0572 0.8711 0.1241 0.0727
(0.1 140) (0.0322) (0.7754) (0.1352) (0.0367)
fl 0.0221 0.8363 0.4375 0.6266 0.8236
(0.1358) (0.0447) (0.4442) ( 0.1265) (0.0542)
(1’ - 0.9354 0.7081 - 0.9015
( - ) (0.0412) (0.2135) ( - ) (0.0539)
LL -468.249 -419.207 -602.231 -700.643 -402.756
Q(20) 45.0589 14.6086 34.6525 44.1258 29.5997
Q2(20) 26.3045 19.8895 14.3141 19.1193 21.9693
Skewness 1.094 0.287 0.355 -2.426 0.326
Kurtosis 5.542 3 .743 8.022 14.604 3 .668

Wd=1 317.057 163.865 164.009 - -

171/5:0 8.439 - - 31.805 -

 

Key: The table reports the Quasi Maximum Likelihood Estimates (QMLE) for various ARFIMA-
FIGARCH models. The QMLE are calculated using the normal likelihood function. Robust standard errors
are reported in parentheses. LL is the value of the maximized Gaussian log likelihood; and Q(20) and
02(20) are the Ljung-Box test statistics with 20 degrees of freedom based on the standardized residuals and

squared standardized residuals respectively. Wd=1 and W5=0 are both Wald test statistics for testing long
memory property in the conditional mean and variance.

94

Table 20. Estimated ARFIMA(p,d,q)-FIGARCH(P.6,Q) models for domestic credit
(1— mm - L)"(y. - 1:) =(1+ 61m + 01* )g,,
.o.,|r2,_,~N(0,a,2 ),
(1 —- [105,2 = a) + [1 — flL — 1,2271 - 1mg}

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Country Indonesia South Korea Malaysia Philippines Thailand
(p,d,q)- (O9d90)- ( 190-10)’ (0,d,0)- (0,091)- (lad90)-
(P,5,Q) (1.1.0) (1.1.0) (1.5.1) (1.0.1) (1,5,1)
)1 2.0421 1.5424 1.3342 1.7938 1.41 17
(0.0681) (0.1684) (0.1469) (0.3218) (0.1092)
(1 -0.2015 - 0.0588 - -0.0580
(0.0629) ( - ) (0.0403) ( - ) (0.0790)
(1) - -0.0550 - - 0.2712
( - ) (0.0730) ( - ) ( - ) (0.1130)

0 - - - 0.0377 -
( - ) ( - ) ( - ) (0-1078) ( - )
9 0.1719“ 0.4153" 0.2200W - 0.3118"
(0.0589) (0.0429) (0.0413) ( - ) (0.0441)
1.0000 - 0.5236 - 0.7105
( - ) ( - ) (0.0923) ( - ) (0.2448)
0) 0.3320 2.0517 0.0416 0.5906 0.0079
(0.5631) (0.5176) (0.0322) (0.5310) (0.0150)
)3 0.8137 0.1549 0.8788 0.8533 0.9119
( 0.0877 ) (0.1072) (0.0384) (0.0589) (0.0463)
(F - - 0.3989 0.9625 0.4632
( - ) ( - ) (0.0660) (0.0447) (0.1937)
LL -917.429 -701 .616 -574.283 -324.310 -591.442
Q(20) 70.4911 34.1811 18.0219 56.4719 54.1633
@720) 8.2300 24.2733 25.6956 13.4509 16.6580
Skewness 0.324 0.390 0.207 -0.01 1 0.284
Kurtosis 6.845 4.570 4.751 3.320 3.675
Wd=1 364.371 - 546.711 - 179.152
W5=0 - — 32.182 - 8.426

 

Key: The table reports the Quasi Maximum Likelihood Estimates (QMLE) for various ARFIMA-
FIGARCH models. The QMLE are calculated using the normal likelihood function. Robust standard errors
are reported in parentheses. LL is the value of the maximized Gaussian log likelihood; and Q(20) and
02(20) are the Ljung-Box test statistics with 20 degrees of freedom based on the standardized residuals and

squared standardized residuals respectively. Wd=1 and W5=0 are both Wald test statistics for testing long
memory property in the conditional mean and variance. *k = 3, " k = 12

95

Table 21. Estimated ARFIMA(p,d,q)-FIGARCH(P,8,Q) models for interest rate
(1 — 90(1- 0%». — 14) = (1 + 92).,.
a,1Q,_,,~N(0.af’- ),
(1 - 800.2 = w + [1- 111 — 711(1— 0’78?

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Country Indonesia South Malaysia Philippines Thailand US
Korea

(Pad,Q)' (lad30)— (0,d,0)‘ (09d90)' (0,d, 1 )' (0109] )' (0,d,0)-
(P,8.Q) (1.0.1) (1.0.1) (1.1.0) (1.1.0) (1.1.0) (1.1.0)
,u -0.0057 -0.0301 0.0193 0.0172 0.1314 0.0380
(0.0150) (0.0392) (0.0118) (0.0157) (0.0669) (0.1461)

d 0.7942 -0.1023 -0.2919 -0.2791 - 0.3397
(0.1778) (0.0575) (0.1374) (0.1078) ( - ) (0.0862)

¢ 0.7870 - - - - -
(0.0865) ( - ) ( - ) ( - ) ( - ) ( - )

6 - - - 0.4394 0.3021 -
( - ) ( - ) ( - ) (00899) (00761) ( - )

5 - - 1 .0000 1.0000 1.0000 1 .0000
( - ) ( - ) ( - ) ( - ) ( - ) ( - )

a) 0.9676 0.0341 0.0298 0.0131 0.0668 0.0031
(0.3360) (0.0228) (0.168) (0.0089) (0.0266) (0.0021)

,8 0.0956 0.9163 0.6295 0.7130 0.6117 0.7453
(0.1276) (0.0339) (0.0698) ( 0.0694) (0.0572) (0.0591)

(P 0.9809 0.9730 - - - -
(02460) (0.0283) ( - ) ( - ) ( - ) ( - )
LL -234.013 -348.631 -396.977 -343.507 -374.630 -l32.241
Q(20) 20.4605 17.2562 42.1987 41.3027 18.6411 30.4143
Q2(20) 9.9756 1 1.2593 22.4051 14.4615 21.7732 19.9586
Skewness 1.135 0.889 1.912 0.783 0.050 0.082
Kurtosis 6.938 6.126 12.036 8.251 3 .625 5.054
Wd=1 101.788 368.118 88.384 140.725 - 58.677

W620 ‘ ' ' ‘ ‘ ‘

 

Key: The table reports the Quasi Maximum Likelihood Estimates (QMLE) for various ARFIMA-

FIGARCH models. The QMLE are calculated using the normal likelihood function. Robust standard errors

are reported in parentheses. LL is the value of the maximized Gaussian log likelihood; and Q(20) and
Q2(20) are the Ljung-Box test statistics with 20 degrees of freedom based on the standardized residuals and

squared standardized residuals respectively. Wd=1 and W5=0 are both Wald test statistics for testing long
memory property in the conditional mean and variance.

96

Table 22. Estimated ARFIMA(p,d,q)-FIGARCH(P,6,Q) models for real GDP
(1 — d)L)(l — L)d(y, — 11) = (1 + 6L)(1+ 622" )e,.
8.152.-1~N(0.0.2).
(1 — {tug} = a) + [1— ,BL — (0L(l — L)6]8,2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Country Indonesia South Korea Malaysia Philippines Thailand
(p,d,q)' (09091)- (Ord90)° (l,d,0)‘ (19d30)‘ (19090)-
(P.6.Q) (0.0.0) (1.0.1) (1,5,1) (0.0.0) (0.0.0)
,u 0.8294 0.9631 0.8122 0.0489 0.6762
(0.1 146) (0.0975) (0.0438) (0.0198) (0.1083)

d - -0.1072 -0.3236 -0.3361 -
( - ) (0.0436) (0.0707) (0.1303) ( - )
(D - - -0.3271 -0.5115 0.4242
( - ) ( - ) (0.0885) (0.1093) (0.1816)

6’ 0.2556 - — - -
(0-1477) ( - ) ( - ) ( - 1 ( - )

9 - - 0.3769'" 0.7941WW -
( - ) ( - ) (0.0477) (0.0584) ( - )

- - 0.5282 - -
( - ) ( - ) (0.2278) ( - ) ( - )
a) 0.2290 0.8691 0.1561 0.2366 0.1134
(0.0687) (1.1564) (0.3233) (0.0468) (0.0321)

fl - 0.7731 0.8480 - -
( - ) (0.2084) (0.0666) ( - ) ( - )

(P - 0.8928 0.5684 - -
( - ) (0.1382) (0.1913) ( - ) ( - )
LL - 1 7.731 -778.994 -922.806 -43.990 -8.592
Q(20) 8.09227 44.6843 26.9356 95.3482 1 1.26877
Q2(20) 1 1.500?” 14.4686 14.0204 29.9340 4.7804
Skewness -0.057 -0.795 -0. 193 0.256 -0.256
Kurtosis 3.336 7.271 4.948 3.464 3.089

Wd=1 - 644.669 350.816 105.088 -

W5=0 - - 5.374 - -

 

Key: The table reports the Quasi Maximum Likelihood Estimates (QMLE) for various ARFIMA-
FIGARCH models. The QMLE are calculated using the normal likelihood function. Robust standard errors
are reported in parentheses. LL is the value of the maximized Gaussian log likelihood; and Q(20) and
02(20) are the Ljung-Box test statistics with 20 degrees of freedom based on the standardized residuals and

squared standardized residuals respectively. Wd=1 and W5=0 are both Wald test statistics for testing long
memory property in the conditional mean and variance. * Q(IO), *"‘ Q2 (10). “'* k = 12, **"‘"‘ k = 4

97

 

we. 35:05 cm 8 av on mmd-

 

 

 

 

9.2..
on ow ov cm and-
we. 35:2: 11111111111111111
l l.l l. llllllllllllllllllllllllllllllllllllllllllllll (l l l l , °°.°
a l 1 .
.mmd

and

 

 

 

0nd-
NNd-
. 86
Q2.
and

 

 

 

£9662 eonEaecfim mo macaw—obooofixx A3

 

Senna Eu
.m.: 8e 688 2.6.8 <s=mm< 58.. 2328. 6865:8686 6:386:60 .... 2:5

98

 

and-

 

 

 

 

we. bags 3 8 2. cm mg-
. Ocd
de
cmd
£22662 HUBER—053m 828% no macaw—0:80:20. 3
$061
we. 35:2: on so ow om mm o-

 

 

 

     

we. fizz—OE

£3602 eofiEwecfim mo macaw—oboguzxx A3

 

Scams Eu .m: E 688 26.:
mom/42426.8 <25? :3 2226.8 8526255 .6 5588680 .m use...

99

BE c2355 EU .m.D mo mo=_a> Bum use 2253. .0 Saw;

._ 8am 5224.11“.

 

5 mm mm mm mm 5 an R mu mm K mm mm mm mm 5 mm mm mm mm

nv

 

. , . , __
.A J: : 2} . ..‘_’_’_‘ __ I, ._’ . .. r. . : 1‘ .7. 2

 

 

 

g ._._.:;_.._..____.:, :L P: P _\
. ‘

 

 

 

 

md

mé

 

 

 

 

 

100

00 mm

88 cozmcfi EU .m.D .«o 833 “088020; .5 oSwE

{222 8.202.286.0591 1.636548669200059.. 8::

mm mm

mm

1 l. _

.3844

mm mm

 

mm vm

Fm

om

 

/

r

L
C, ..
>2“ ”¢>/t\/\§f<k\ /\\/</

2,

i

/

 

g’.’- k’x.,
A

\ifrlt.rll\/1/\/\l\.
\/

mm:_m> 36390“.

 

 

26$... 7. 2:3. ?

md

 

 

me

101

 

we. 35:9: ow cc ov

 

and-
NNd-

. . cod

de
and

 

 

 

 

 

 

w .A :2:
a == 3 8 S. cm NS-
---.....M-...‘.......:.:..----..-..H:.-..-H-..-H.:....1.-.. in--n-w......-..u-...1M4-H- 8°
de
. and
$5668 BEE—353m @0333 we mcotm—otoooS/V. 3v
9 and-
x

we. 35:08 ow ow on «Nd-
V 11111111111111 .11>.1111 (Flynn: 1.1.111 1111..|.r.1111.11..1111 >1111.1V11.1111H .11. 8.:

11...! 111111 rirrutnurru‘ 11111 uiwrrrrwri- 111111 a 11111 ‘1‘111141141W111W1‘4111.11
mad
and

£9662 pod—0.898% .«o macaw—otoooS/w A3

 

«60:85 dam Boa 32333 Sm mfiaflmoe “6368955 .«o mEEwBotoU .w 3:me

102

 

wfl 35:9: on 00 ov om mmd-

 

 

owd-
we. >222: ow ow ow on mmd-
cod
mmd
one

 

 

 

ME k-fiwZOE ow CO CV ON NN.OI

 

22668 vofiwaeesm we mcouflotoooS/x A8

 

«Box 550m #73 Eofi 223320 L8 $9.662 noficaefiam “0 28328.80 .a oSwE

103

 

wfl 35:9: ow ow ow om

 

 

we. 35:9: ow cc 9. om

 

2:523: woNEEugfim 355cm mo m:osm_o:89=< 3v

cod
de
end

 

 

ma. 258:. 8 8 9. . om

 

£9569 5358985 50 m:9§otooo§< Aev

owd-
NNd-
cod
de
and

 

$98—92 Him :55 39:25“: :5 £323: 5358988 50 mEEmoBboU .3 oBmE

104

 

mg 35:9:

 

 

omd-
w:_ 35:9: mmd-

 

m_::2mo: 5:22:55“: 52:53 :0 m:o5:_o::88=< 3V

 

 

   

w:_ 35:9:

2:328: 252:5:Sm we m:o::_o::ooo:=< A8

 

3:25:23 in: 89m $8525: :5 3:328: 822:2:Sm mo mE::wo_o::oU ._ _ oSmE

105

 

36.

w:_ 355:: ow cc ov om
mmd-

 

 

:2 35:2: 8 8 8 cm

 

 

 

w:_ 355:: on ow ov om

 

3:523: 5oN25::5:::m mo m:o_::_o::oQoS.< AS

 

5::=::H min :85 m:o5:_>o5 :8 2:328: 5:22:55; «0 9:83—2:00 .N: oSmE

106

 

and-
mmd-
cod
de
and

w:— 35:9: on 05 ow om

 

 

and-
NNd-
ocd
de
, and

w:_ 35:9: ow on ov om

 

 

 

 

w:_ 35:9:

2:528: 50258555: («0 505228095»: 55

 

28:55 550:0 58:55 :5 2:528: 5358558 no mE::mo_u::oU .2 0:555

107

 

 

 

mg bats—OE CM CO 0? ON NNOI

 

 

 

 

and-
NNd-
cod
de
and

w:— 358E

2:528: 50258588 50 8288:8083»: AS

 

:23— 55om .2508 58:85 :5 2:528: 50258558: 50 2882280 .E 0:52:

108

 

mm. bficoE

 

 

on 00
mm. bficoE ov om NNd-

 

22658 vufifizwwcflm 88:? mo mcocfluhnvoo5< 3V

 

 

owd-
NNd-
cod
de
and

 

23.2mm: uofiwsvcfim mo mco_§otonvo§< A3

 

£3232 .558 05388 Sm m_m:Em2 uuwfiauqfim mo mfifiwofiboo .2 8:me

109

 

ma. 35:2: ow cc ow om

 

and-
NNd-
ocd
de
cnd

 

 

wfl 35:9: ow co ow om

 

 

 

 

29632 @3632me 38:3 mo macaw—obouofixx 3v
mm. 35:9: ow ow ow cm and-
IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII I I I I de-
I '. 1 ' b > I b p . D 5 . r PI ' ' ’ . D I ' . - C°.o
‘ ‘ ‘ ‘ 1 < 4 ‘ 4 4 4 ‘ ‘ ‘ ‘ ‘ . 4 ‘
IIIIIIIIIIIIIIIIIIIIIIIIIIII III IIIIIIIIIIIIIIIIIIIII IIIIIII IIIII I N00
and

$9668 woumtacqfim mo macaw—utooofiax 3

 

 

moEQnEEm £38 03888 8m 3328“ woumuaauafim mo mfifiwofiboo @— 95me

HO

 

M: 25.55 S 8 8 cm

 

 

m2 358:. on 8 8 on

 

 

 

Omd-
NNd-
cod
de
end

 

92 35:9:

25662 noNESccflm mo macaw—obouofiaa A3

 

wan—RFC. .558 osmofion ho.“ mafia»: uoNEEcqfim mo mESwofitoU .: 059.»

Ill

 

wfl bficoE ow ow

 

flasEmB vofivbwvcflm 828% we m:o_~m_otooows< 3v _

 

m2 358:. on 8 S om one-

 

 

 

wa— AZEOE
cod
de
omd

 

£2.28“ UonEmuqfiw mo m:o_um_ut009=< A3

 

«60:85 .88 6885 How £9634: vofivhmuzfim mo assume—2.80 .2 2sz

112

 

m2 bficoE ow ow ow

 

 

we". bficoE ow cc 9.

 

 

 

mm. 35:9: cm co ow om

 

$2632 voflvavqfim g8 mconm_obooo§< A3

 

«Bov— fiaom .28 6885 5m £9668 cuwaSucfim we wEEwo_otoU .3 oSwE

113

 

one-
we. 55.55 8 8 S cm mg.
86
3°
and

 

 

ME 35:05 on ow 3. cm NNd-

 

 

 

m2 35:9: ow mg-

 

£3662 $363283 90 macaw—otgofixx AS

 

«$332 .88 628.: 5% 25:23“ commufivcflm mo msauwfloboo .om oEwE

H4

 

cm
9.
mm. 35:2: ow co

 

 

mmd-
ms 35:2:

 

 

 

. one-
NS-
86
N3
93

um. 35:2: ow

 

22652 @3682me go mcouflobooousxx AS

 

8:59.:3 .88 $885 5m m_m=Em2 uofiwauqflm mo mEEmoEboU AN oSmE

HS

 

w: 258.: 8 8 8 om

 

 

wa. 25:2: ow co ow om

 

 

 

ms bficoE ow ow ow om

 

29662 coinsvcfim mo macaw—otoooasxx A3

 

tawznf. .28 $285 Sm 22662 “9363983 “0 mESmoBtoU .NN uSwE

ll6

 

ms :55: 8 8 8 on

 

 

mg 35:2: ow co ov om

 

 

 

and-
NNd-
cod
de
omd

ms 35:2: ow oo ow om

 

£5668 88:888.: mo msosmw—otooousxx AS

 

.m.D .88 $835 :8 28668 voNfiavqfim mo mE8w2otoU .mN 8:me

117

 

ma. 3:8»

<><>/\/ _

28658 3888236 3203: («o m:o_§_u:._OooH:< CV

and-
NNd-
cod
de
omd

 

w:. 350»

< </\I_

$86.3: 38835; 688.3 .8 m:o=m_obooo§< 3v

and-
NNd-
cod
de
and

 

 

mm. 3:8»

£828: “5888:qu mo m:o_§o:ooo§< as

end-
NNd-
cod
de
ond

 

83:85 .mQO .8: :8 $8.28: 38:88:38 :0 mE8w2oboU .vm uSmE

118

 

owd-

mu.— 3£:oE a» cc 9. cm

 

 

ms 35:9: 8 co ow om

 

 

 

:2 25:2: 8 8 8 . 8

 

28:28: 38:88:53 :0 mcosflutouousx: A3

 

«ROM 550m .mDO .3: :8 £323: 3888::me mo m88wo_o::oo .3 0:39.”:

119

 

ms 55:08 ow co ov om

 

2328“ woNEum—Efim 0::on mo macaw—mbooowsax on _

 

ow co ov o
mg 35:9: N mad-

oo.o
~00
omd

 

 

 

wfl x222: ow ow ov . om mmd-

 

29662 coNESncflm we muons—obooofizxx AS

 

398—32 .AQO RE How mafia»: vofiwumucflm mo mESwo—oboo .om quE

120

 

wfl b.8530 9» ON and-

 

 

 

(xv / \ /\ <5! <

IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII Kilnllullun Inll<unln de
.omd
man—Em»: vow—caucfim 228% mo macaw—obooo3< on _
wfi barns: ow ON 36.
IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII NN.OI
\l 7 ? k I 7 .oo.o

< t <
innlll..uunununnuauuuuuuuununnunuulnuunuunnuunuunullunuulnn< nnnnnnn mmo
and

 

 

 

. >3):

......... w. <</\< <<<<< mm“

 

 

 

$59335 .ADO 32 SM "”1328“ Bfiwavcfim mo mEEwBESU Sm oSwE.

121

 

ms. 35%

 

 

 

 

 

 

end-
ON 2 > mmd-
(\llI/r< <<|l / cod
de
..................................................................... one
mfisEmB voNEHqufim 8203a mo mcouflotoQoS< A3 _
wfl baa» _
IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII Omd-
om 2 ANS-
Il‘l} \rllk/ > , 2;
< /\ / moo
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn omd
25688 cummwavcsm 33:? mo mucus—obooouzax 3v
mm. 38%
uuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuu and-
om S \/ Rd.
.F >11 Alk >< <\/<>/ 8o
mmé
..................................................................... one

£32m»: voficaucfim mo mco_§oboUoS< A8

 

cam—Rf. #50 32 Sm £8662 BNEEESm mo msahmofioboo .wm oEwE

122

Appendix 1. Previous studies of currency crises based on time series data

 

 

 

 

 

 

 

Author and Countries Time Period Data Comments

Year Published Frequency

Balanco and Garber Mexico 1973-1982 quarterly Main focus is on the

(1986) one-step~ahead
probability of
devaluation. The
expected next-period
exchange rate
conditional on the
devaluation also is
constructed

Cumby and Van Argentina 1979-1980 monthly One-month-ahead

Wijnbergen (1939) probability of a collapse
of the crawling peg is
yielded.

Goldberg (1994) Mexico 1981-1986 monthly The emphasis is on
explaining the forces
contributing to
speculative attacks on
the Mexican Peso.

Otker and Mexico 1982-1994 monthly A main role of

Pazarbasioglu deterioration in

(1996) fundamentals in the
collapse is clarified by a
Probit model.

Otker and European 1979-1995 monthly This study shows that the

Pazarbasioglu countries weak fundamentals are

(1997b) not the only cause in the

 

 

 

 

crisis.

 

123

 

CHAPTER V

ESTIMATES OF REAL MONEY DEMAND FUNCTIONS

1. Introduction

Well-known structural analyses of currency crisis, Blanco and Garber (1986),
Cumby and Van Wijnbergen (1989), and Goldberg (1994), estimate a real money demand
function without taking into consideration the non-stationarity of variables. Therefore,
their analyses display potentially a spurious regression problem and the conventional t-
ratio and F significance tests can not be applied.

However, cointegration and error correction techniques in modeling of real money
demand (see Granger, 1986; Engle and Granger, 1987; Hendry 1986; Johansen and
Juselius, 1990) help to avert spurious regression and correct for the non-standard limiting
distribution of coefficient. Furthermore, it provides a more robust means of estimating
long run relationships and short run dynamics among the set of macroeconomic variables
of interest.

This chapter examines the existence of a stable long-run real money demand
function in South Korea and Malaysia for which high frequency data are available using
the cointegration and error correction model. In addition, a short-run relationship among
real money, real income, interest rate and interest rate differential is tested. The results of
the study in this chapter suggest that both long and short-run models can be specified in

South Korea and in Malaysia.

124

The regression specification to estimate real money demand includes a constant

term, and the coefficients of real income, interest rate and the interest rate differential.
The estimated constant term, [10 , and coefficients, [1] , &2 and (33 will be used in Chapter

VI to obtain the shadow exchange rate,

N

S: = m: ‘90 +511": "512)”: ‘P: "uz +(‘3i “AUX/1: +91) '9-

The following sections present a theoretical framework and an empirical model of
real money demand. Then, univariate unit-root tests are presented to detect the
nonstationarity of variables. Finally, an error-correction model is applied to estimate long

and short-run dynamics in the real money demand function.

2.Theoretical framework

There is a diverse spectrum of money demand theories emphasizing the
transactions, speculative, precautionary or utility considerations. These theories implicitly
address a broad range of hypotheses. One significant aspect, however, is that they share
common important elements (variables) among almost all of them. In general, they bring
forth relationship between the quantity of money demanded and a set of few important
economic variables linking money to the real sector of the economy (see Judd and
Scadding, 1982).

The general specification begins with the following functional relationship for the

long-run demand for money:

 

'9 See Chapter 111 for the details.

125

M

d
[7;] =foui), f..>0. .f.-<0 (1)

where the demand for real money balances, M / P , is a function of a scale variable, such
as real income y, and an opportunity cost variable, interest rate i. Except for interest rate,
all variables have been natural log transformed in the estimation process. Using the real
money balance as the dependent variable means that price homogeneity is explicitly
imposed into the model. In addition, there are less severe econometric problems

associated with using real rather than nominal balances as the dependent variable (see

Boughton, 1891, and Johansen, 1992).

3. Empirical model

In general, the empirical work begins with a typical formulation of a simple
theoretical money demand function relating demand for real money balances, m, to a
measure of transactions or scale variable, y, and the opportunity cost of holding money, i.
Empirical formulations also incorporate lagged dependent variable to capture the short-

run dynamics.

3.1 Error-correction models

Error correction models (ECMs) are one of the most successful tools in applied
money demand research. This is a dynamic error-correction representation where the
long-run equilibrium relationship between money and its determinants is embedded in an

equation that captures short-run dynamics. A specific model is

Ax: = rlet—l + "'+ Fk-let—kH + ”xi—l + #0 + '7: (2)

126

where x, is a p-dimensional vector of 1(1) variables.m.---.17T are IINp(O,A) and

x_k +1 ---x0 are fixed. The 17 matrix conveys the long-run information in the data. The

impetus for this type of model came from the findings that appropriate consideration
needs to be given not just in the selection of an appropriate theoretical set up and
empirical make up, but also in the specification of the proper dynamic structure of the
model. Hence, economic theory should be allowed to specify the long-term equilibrium

while the short-term dynamics can be defined from the data.

4. Application of ECM to the estimation of real money demand
4.1 Data set

The interest in the demand for money of developing countries has increased in
recent years. This interest was triggered primarily by the concern among central banks
and researchers about the impact of movement toward a flexible exchange rate regime,
globalization of capital markets, and ongoing domestic financial liberalization and
innovation. South Korea and Malaysia, both upper-middle-income developing countries,
were selected for the estimation of the real money demand function because their foreign
exchange rate policy toward a financial liberalization continued from the 19805 to the
19905. Then, after the abrupt devaluation in 1997, South Korea and Malaysia chose
opposite foreign exchange regimes. While South Korea allowed a free foreign exchange
rate regime, Malaysia returned to the fixed exchange rate system. Hence, the shadow
exchange rates and the probabilities of collapse obtained later with the estimated real

money demand function in this chapter will provide an opportunity to compare the

127

 

economic conditions after the crisis under two very different exchange rate regimes.
Furthermore, the two countries” data, also, have a higher frequency than the other
countries’ that experienced the Asian currency crisis in 1997-98. This will increase the
power of statistical tests and. in turn. raise the confidence one can place in the empirical
results.

Monthly data from 1970201 through 1996:12 for South Korea and Malaysia are

used for the estimation. All of the variables, m2, y, p, i, and I" are taken from the CD-
ROM version of the lntemational Monetary Fund’s International Financial Statistics
(IFS) and defined in Table 17. Real money balance, rm2 (mZ-p), and y are seasonally
adjusted. In particular, data not seasonally adjusted is preferable for the unit roots tests
and cointegration analysis since the seasonal-adjustment filters have adverse effects on
the power of the unit root and cointegration tests according to Ghysels (1990) and
Davidson and MacKinnon (1993). However, seasonally adjusted data are used in this
analysis due to the availability of data”. Despite the loss of power from the seasonal
adjustment filters, the use of seasonally adjusted data removes the need to test for the
order of seasonal integration. In addition, it removes the need to add seasonal dummy
variables in the real money balance equation. Figures 29 to 36 present graphical

descriptions of the variables used in the analysis.

4.2 Unit-root tests

4.2.1 The Dickey-Fuller unit-root tests

128

A univariate test for unit roots was first advocated by Fuller (1976) and Dickey
and Fuller (1981). The Dickey-Fuller based approach basically involves the running of
the following univariate regression

k

x, = ax,_] + 209-21ij + [set of fixed regressors] + 8, (3)

1:1
where the set of fixed regressors include a constant and a linear trend. We consider two
alternatives, with an intercept and with an intercept and a trend. Since the presence of
autocorrelation destroys the properties of the test, it is important to make the correct
augmentation to be able to interpret test results. If the number of lags is above or below
the number necessary to render the error as white noise, it biases the test’s power and
size. Too many lags reduce the power of the test. Too few lags distort the test size. The

number of lags in equation (3) is selected such that the regression yields non-serially

correlated errors.
If x, is non—stationary, a will assume a unit value and x, likely has a unit root.
The null hypothesis that a = 1 can be tested by reference to its usual t-statistic based on

a , the OLS estimator. This statistic is referred to as the Augmented Dickey-Fuller (ADF)
statistic. However, the distribution of ADP does not follow the usual student’s t
distribution. Approximate critical values of this statistic were originally given in Fuller

(1976). In addition, the normalized bias test statistic, Tc((i - 1), where T is the number of

 

20 Only seasonally adjusted real income for South Korea can be collected in IFS.

129

k
the observations and c = ('1 — Zajf' . can be employed to test the non-stationarity of the
j=|

variable.

The results in Table 23 strongly indicate that there is a unit root in rm2, y, i, and if

(i —i*) in South Korea and rm2 and y in Malaysia since the null hypothesis cannot be
rejected at the 5% level. However, the null hypothesis is rejected at the 5% level by the
normalized bias tests T c(& —-1) for i and if in Malaysia and, at the 10% level, for if in
South Korea. In addition, the results, as summarized in Table 24, unambiguously reveal
that rm2, y, i, and if in South Korea and rm2 in Malaysia contain a unit root. This is
because none of the tests can reject the null hypothesis of a unit root in the regression
with an intercept and a linear trend. The null is rejected at the 5% level by the normalized

bias tests for y, i, and if in Malaysia.

4.2.2 The Phillips-Perron unit-root tests

Phillips and Perron (1988) propose a nonparametric method of controlling for
higher-order serial correlation in a series. The test regression for the Phillips-Perron (PP)
test is the AR(l) process:

x, = (UH +[set of fixed regressors]+ a, (4)

While the Dickey-Fuller test corrects for higher order serial correlation by adding lagged
differenced terms on the right-hand side, the Phillips-Perron test makes a correction to the
t-statistic of the coefficient from the AR(l) regression to account for the serial correlation

in a . The PP t-statistic is computed as

130

A V] 1‘2 A _ A
_ ’(1 —(/. —}’())T(Ta

 

 

(5)

2. 22.5
‘7 . , . A . . .
where 2.“ IS a Newey-W est estimator, ta and aa are the t-statistic and estimated standard

error of a , and 71' and .92 are the jth autocovariance and the estimated variance of c.

The correction is nonparametric since we use an estimate of the spectrum of s at
frequency zero that is robust to heteroskedasticity and autocorrelation of unknown form.
The Newey-West heteroskedasticity autocorrelation consistent estimate
A2 k
x. = yo + 2::[1— j/(k + 1)]3'1- (6)
Fl
where k is the truncation lag. The asymptotic distribution of the PP t-statistic is the same

as the Dickey-Fuller t-statistic. With the PP t-statistic, Z , . the Phillips-Perron p statistic,

Z p , also can be used for the unit root test. The Z p is calculated as

{/12 -

 

ZP=T((2—1)— “ (7)

The results of applying the PP test procedure to the variables in the South Korean
and Malaysian demand functions for real balances are presented in Table 25. While the
results of a regression with an intercept indicate that there is a unit root in rm2, i, and if in

South Korea and rm2 and y in Malaysia since the null hypothesis could not be rejected at

the 5% level, the null is rejected at the 10% level by Z , for y in South Korea and at the
5% level by the Z , and Z p for i and if, respectively, in Malaysia. In addition. the Z , and

Z p tests with an intercept and trend show that rm2, y, i, and if in South Korea and rm2

131

and ifin Malaysia contain a unit root. The null hypothesis of a unit root is rejected at the

5% level by the Z, and Z,, tests for y and i in Malaysia.

4.2.3 The KPSS unit-root tests

Due to the well known low power of the standard unit-root tests from the
empirical evidence, Nelson and Plosser (1982), Kwiatkowski, Phillips, Schmidt, and Shin
(1992) proposed a test of the null hypothesis that an observable series is stationary around
a deterministic trend. They assume that a series can be decomposed into the sum of a

deterministic trend. a random walk, and a stationary error:
y, =g"t+r, +s,. (8)
Here r, is a random walk:
rt: t-1+ut9 (9)
where the u, are iid (0, 0'3 ); the initial value r0 is fixed as the intercept. In this setting,

The stationary hypothesis is simply 03 = 0. Then. since 8, is assumed to be stationary,
under the null hypothesis y, is trend-stationary. They also considered the special case of

the model (8) where if = 0, in which case under the null hypothesis y, is stationary

around a level rather than around a trend.

The KPSS LM(and LBI) statistic is defined as:

T
a, = T‘225,2/s2(1), (10)

t=1

t T l T
where S, = 2e}, t = 1,2 ..... T and szfl) = T—IZef‘ + 2T_'Zw(s,l) Ze,e,_s . Here w(s.

i=1 t=l s=l I=s+l
I) is the Bartlett window which is the same as the Newey-West heteroskedasticity

autocorrelation consistent estimate in the PP t-statistic.
In the study we consider two null hypotheses: one is the level stationary

hypothesis, the other is a trend stationarity. The resulting test statistics are denoted 5,, ,

and 1'], respectively. For each test, we consider values of the lag truncation parameter, I,

from 2 to 16. Since the data series are highly dependent over time and the residuals from
the regressions are serially correlated, it is not realistic to assume iid errors under the null
and use I = 0. no correction for autocorrelation, in estimation of the long-run variance.

The choices of 15 and 16 for I follow the values of I as a function of T : l = integer

[12(T/ 100 )1/4] . This is the rule suggested by Schwert (1989) to sufficiently correct for
the autocorrelation problems in the residuals.

The test results are provided in Tables 26 and 27. First we consider the null
hypothesis of stationarity around a level. The null hypothesis of level stationarity is
rejected at a 5% level for all series except if for South Korea. This is the case regardless
of the value of I chosen. This is shown in Table 26 for both South Korea and Malaysia.
The rejection of the null is not surprising for the real M2 and GDP, since obvious
deterministic trends are present (see Figure 29, 30, 33, and 34). For the if in South Korea,
the test result cannot reject the null hypothesis at a 10% level when the value of 15 is
chosen for I. Then, when the trend stationary hypothesis is tested, as reported in Table 27,

for all of the series apart from rm2 in South Korea, the null is rejected at a 5% or a 10%

133

level for all values of l. The outcome for the rm2 in South Korea depends on the lag

truncation parameter, I.

4.3 Residual based cointegration tests

Engle and Granger (1987) suggest that the residuals from an OLS estimation of
the cointegrating regression can be examined for the presence of a unit root in the
autoregressive representation. If there is no cointegration, there should be a unit root in

the residuals.
Let the observed data X , be a px] dimensional time series, partitioned as
X, = (x,,,x§, ) , where x,, is a scalar and x2, is an m-vector, p = m+1, and each element

of X , is known to be [(1). By regressing one of the variables, say x“ , on the others with
ordinary least squares, we obtain the cointegrating regression:

x,, = oi'Xz, +22, (1 l)
where X 2, may also contain a constant or time trend, other than x2,; 1?, are the
residuals. Our null hypothesis of no cointegration then corresponds to the null hypothesis
that ii, is 1(1) where

ti, =j512,_, +5,. (12)

While a simple Dickey-Fuller test can be used in this model, instead we consider

Z p and Z, tests, since these test statistics have the advantage that they correct for both

potential serial correlation and heteroskedasticity in the cointegrating errors.

134

Following Hansen (1992), we consider two procedures. The first is to run an
unrestricted OLS regression with a time trend included to test for the existence of
cointegration in a series with a drift. This procedure is equivalent to detrending the series

first:

x1, = 13+(2'x2, +57“; (13)
The inclusion of time trends in the regression has the advantage of rendering estimates of
the cointegrating vector invariant to the presence of trends in the regressors. This also
simplifies the asymptotic theory as shown in Phillips and Hansen (1990) and Hansen
(1992). The second approach is to estimate an OLS regression without a time trend.

x,, =17+5'x2, +17, (14)
As Hansen (1992) notes, cointegration tests without a time trend are generally more
powerful than cointegration tests with a time trend.

The results of unit root tests identiry rm2, y, i, and if in South Korea and rm2 in

Malaysia as 1(1) processes. However, the unit root tests do not indicate any non-
stationarity of y, i, and if in Malaysia. Therefore, while the results of the Z p and Z, tests

on the residuals of the cointegrating regression in South Korea are valid for detecting
cointegration, their justification in Malaysia is suspect. In spite of this, the estimated
coefficients in the OLS estimation implemented that were used for the residual-based
tests will be applied for obtaining the shadow exchange rates and the probabilities of
collapse using the same framework defined by Blanco and Garber (1986), Cumby and

Van Wijnbergen (1989), and Goldberg (1994).

I35

Since residual-based cointegration tests are developed from single-equation
regression models, they depend on an arbitrary normalization of the cointegrating
regression. As far as the demand for money function is concerned, the long-run money

demand relation with no structural breaks may be written as
rm2f1= a0 +aly, + azi, +a3if, +u, (15)

where rm2? = m,d — p, and if, = i, -i: . We consider both cases (13), which includes
time trend, and (14), which does not. The results are as follows.

In Table 28. there are the results of both OLS estimation and residual based

cointegration tests. The null hypothesis of no cointegration can be rejected by Z ,, and Z,

tests in South Korea regardless of the inclusion of a deterministic trend. In particular, the
tests show that there is a long-run cointegrating relationship between the variables in the
demand for real M2 equation. From the OLS estimates, we notice that the sign of income,
y, is positive as we expected, but the signs of if in South Korea and i in Malaysia are the
opposite of what we anticipated. However, since the money market interest rate is used
for i, we cannot argue that the sign of the domestic interest rate and the gap between the
domestic and foreign interest rate needs to be negative. In particular, the money market
rate is a representative interest rate reflecting various interest rates in the financial market.
Thus, it is probable that the negative effect on real money demand from an alternative
asset interest rate, such as the treasury bill rate, is not strongly reflected on the sign.

Due to the uncertainty about the nonstationarity of the variables, the results of

the Z p and Z, tests in Malaysia are not reported in Table 28. Despite the uncertainty,

136

however, the t-statistics and adjusted R-squares are exceptionally high which are common

symptoms of a spurious regression.

4.4 Johansen’s full information maximum likelihood estimation

Even though the residual-based cointegration tests indicate a cointegration
between the variables, we cannot determine from those tests whether there are other
linearly independent cointegrating vectors in the system. Approaches other than residual-
based tests for cointegration are available such as the likelihood ratio tests of
cointegration rank of Johansen (1988, 1991), Johansen and Juselius (1990), and a
common stochastic trends test proposed by Stock and Watson (1988). These tests are
developed from system methods designed to help researchers avoid invalid restriction
from arbitrary normalization in cointegrating regression of residual-based tests. As an
alternative, Johansen’s (1988, 1991) maximum likelihood methods for the analysis of
cointegration can simultaneously detect the number of the cointegration vectors in the
system, estimate and test for linear hypothesis about the cointegrating vectors and their
adjustment coefficients. Based on these advantages, we will apply this technique to
continue our study.

To begin with, the following model without a time trend is fitted to the demand

for real M2 data.

Hzfdx, = FIAxl—1+'H+Fk-1Axt—k+1+17xt—1 +110 +77, (16)

137

In addition, a subsequent model with a linear trend in the cointegrating relations is
estimated to check the significance of linear trend in the estimation. That is, under the

null we estimate

H3:Ax, = 1‘,Ax,_, + + 1“,._,Ax,_,.,, + amt/3, )(x;_,,z)' + no + 77,. (17)
The lag length k is chosen to be the minimum length for which there is no significant
autocorrelation in the estimated VECM residuals using the Ljung-Box Q statistics (1979).
The misspecification tests for the normal iid assumption for the residuals in the model are

reported. The normality assumption is tested by the Jarque and Bera statistic (Jarque and

Bera, 1980).

4.4.1 Misspecification tests

The misspecification tests for the model are provided in Table 29 and 30. In Table
29, while the p-value of the Q statistics shows no autocorrelations in the residuals, the
excess skewness and kurtosis in the residuals of y, i, and if cause the Jarque-Bera test
statistic to become significant. However, the deviations from normality are not a serious
problem. So long as the cumulative sums of errors converge to a Brownian motion, the
asymptotic analysis is the same as that given under the assumption of normality
(Johansen, 1991, Johansen and Juselius, 1992, and Gonzalo, 1994).

The results of misspecification tests in Table 30 also do not specify any evident
autocorrelation except for y. Nevertheless, the Jarque—Bera test statistics are significant

for y, i, and if

138

4.4.2 Testing for reduced rank and normalized cointegrating vector

The cointegration tests results are provided in Tables 31 and 32. In Table 31, the

trace and Am,“ tests fail to reject r = 0 at either the 1% or 5% level when the model is

regressed with a drift. However, both the trace and Am,“ tests in the model with a

deterministic time trend can reject r = 0 at the 5% level. This indicates that whether there
is a time trend or not is important to establish the conclusion of cointegration with stable
coefficients in South Korea. Alternatively, in Malaysia, the trace test rejects r = 0 at the
5% level both in the models with and without the time trend. Hence, the existence of a
time trend in Malaysia is relatively less important compared to South Korea.

The normalized cointegrating vector and the error correction coefficients reported
in Table 33 and 34 offer a long run relationship between rm2, y. i, and if However, the

parameters a and [1’ are not identified. This is because, given any choice of the matrix

g(’rxr) , ac and fl(g')"' also produces the same matrix 17 . The data only identify the
space spanned by the columns in ,8 , and the space spanned by a.

As shown in Table 33, the signs of coefficients on income and the interest rate
differential in the model without a deterministic trend are not consistent with the
prediction of the theories. They also are not consistent with the result of the residual-
based cointegration test in the model with a drift. However, since no cointegration
relationship is detected in the model for South Korea, no significance can be attached to
the directions of coefficients on both variables. As such, the signs of the coefficients on
income and the interest rate in the model with a deterministic trend do agree with the

expectations from the theories. Nevertheless, the sign of the interest rate differential, if,

139

representing an impetus of currency substitution caused by an expected devaluation and a
risk premium, does against theoretical expectations. However, since the money market
interest rate is used for i, we cannot presume a negative relationship. While the
cointegation tests detect a cointegrating relationship only in the model with time trend for
South Korea, they indicate a cointegrating relationship both in the models with or without
a time trend for Malaysia. Therefore, we need to consider both models with and without
the deterministic time trend in Table 34. As for the Table 34, the signs of coefficients on
income and the interest rate differential coincide with what we expected in both models
with or without a time trend, but the positive signs of domestic interest rates in both
models indicate the traits of money market rate representing various interest rates.

In general, the a matrix should contain the weights used to enter the cointegrating
vectors into the system. Each nonzero column of the a matrix also measures the speed of
the short-run response to disequilibrium in the equations of endogenous variables. For
example, the coefficients of a in Table 33 measure the feedback effects of the lagged
disequilibrium in the cointegrating vector onto the variables in the vector autoregression
(VAR). In particular, the absolute value of the first term in a represents the speed at
which rm2, the dependent variable in the first equation of the VAR, moves toward
restoring the long-run equilibrium. We can see that equilibrium errors cause i and if to
adjust more rapidly than rm2 and y in the Table 33 and Table 34. This suggests that the
adjustments of the domestic interest rate and the difference between the domestic and

foreign interest rate are crucial to the cointegrating relation.

4.4.4 Weak exogeneity tests

140

The weak exogeneity tests permit one to draw inferences from the cointegration

relationship that is to examine whether the short-run demand for money could be

modeled in a simpler setting. Let observed data X, be a px] dimensional time series,
partitioned as X, = (x],,x'2, 1 , Where x], and x2, are m— and n-vectors, respectively; p =
m+n. The variable x2, is said to be weakly exogenous for aand ,8 , if the conditional
distribution of Ax“, given sz, as well as the lagged values of X, and AX, , contains
the parameters aand ,8, whereas the marginal distribution of Ax2,, given the lagged

values of X, and AX, , does not contain the parameters aand ,6. In particular, the

parameters in the conditional and marginal distribution must be variation-free or, in other
words, they cannot have any joint restrictions. These conditions are taken from Johansen
and Juselius (1990, 1992) and Johansen (1991).

Since one cointegrating relationship has been identified in the cointegration test
with the time trend in South Korea, and with or without the time trend in Malaysia, the

weak exogeneity tests are evaluated under the assumption of rank (r) = 1. The test

statistics will be asymptotically distributed as 12(1) if a given variable for the
cointegrating vector is weakly exogeneous. Here, the null hypothesis is the existence of
weak exogeneity. This is usually examined by the restriction of a particular a to zero.
When the null hypothesis is not rejected, disequilibrium in the cointegrating relationship
does not have a feedback on the variable of interest. However, any disequilibrium of a
given variable will still impact the cointegrating relationship.

Table 35 shows that weak exogeneity is rejected for rm2 and y at the 5%

significance level and for rm2 and i at the same level respectively in the models with a

141

drift and with a time trend. Therefore, a short-run model can be designed with a system of
two equations, one with rm? and another with y or i by considering other variables as
weakly exogenous. However, since the cointegration relationship represents the demand
for money, an alternative single equation framework can be estimated for the short-run
model with Arm2 as the endogenous variable in spite of a loss of efficiency. Besides, the
results in Table 36 indicate that the weak exogeneity is rejected for only y at 5%
significance level in the model with drift. Nonetheless, it is rejected for rm2 and y at 10%
and 5% respectively in the model with a time trend. Given the results in Table 36, the
short-run model involving Arm2 as the endogenous variable in the single equation

framework can be used in the model with a time trend as well.

4.4.5 Stability of long-run parameters

So as to ensure the robustness of estimation parameters, they are evaluated for
their stability throughout the sample period. To accomplish this task, a VECM with drift
is estimated using the recursive estimation method beginning in early 1985. From 1990
onward, a series of deregulatory measures was put in place to meet the increasing need
for liberalization to improve the efficiency of domestic financial markets and to respond
effectively to the rapid changes in international financial markets in South Korea.
Similarly, Malaysia continued its policy to liberalize interest rates first set out in 1978
and continued during the years of the sample. Because of the time frames for policy

changes, the initial point of early 1985 in the recursive estimation should still leave

142

enough data points from which to examine whether the demand for the real M2 has
remained stable over time.

Figures 37 to 39 provide evidence about the stability of parameters of real GDP,
domestic interest rate, and the interest rate differential in South Korea. As expected from
the cointegration tests, the graphical evidence indicates instability in the long-run
parameters during this period. While the parameters are particularly unstable in South
Korea, Figures 40 to 42 present weaker evidence of parameter instability in Malaysia.
The elasticity of real GDP is fairly stable and close to unity throughout the period for
Malaysia. Also, other parameters, such as the coefficients on the interest rate and the
interest rate differential, exhibit notable constancy. This is an assuring result and it is
expected given that the cointegration test found a cointegration relationship at the 5%

critical level.

4.4.6 Short-run model

The short-run model provides information about how the adjustments take place
among various variables to restore the long-run equilibrium in response to short-term
disturbances in the demand for money. Essentially it is an ECM with an error-correction
(EC) term to control for the existence of a long-run relationship. In general, short-run
models have the I(O) representation of the variables both on the left-hand side and the
right-hand side of the equation. Since the variables are assumed to be either 1(0) or 1(1),
the right-hand side will consist of the first differences of the relevant variables with the

exception being the inclusion of level variables in the EC term.

143

Based on the weak exogeneity tests, a single equation reduced form model such as

equation (18) is sufficient to analyze the short-run dynamics for Arm2.
Arm2, = no +a.(L)Ai,_, +/)’(L)Ay,_, +y(L)_/lif,_, + a,(/)",/>’l)(x,'_,,l)' + n, (18)
The right-hand side of the equation (18) includes an EC term, which is a,(fl',/3, )(x;_,,t)' ,

calculated as rm2 minus the estimated rm2 in time t-1 . In economics, it represents excess
money in the previous period. Since all variables are 1(0), the above model can be
estimated by OLS. The results of the estimation are shown in Table 37 and 38. The error-
correction term, a], is negative in both South Korea and Malaysia. This validates the
significance of the cointegration relationship. A significant negative EC term conveys
two pieces of information: first, agents have corrected in the current period a proportion
of the previous disequilibrium in money balances. Rose (1985); second. it assures us that
the cointegration relationship established previously is valid by Granger’s Representation
Theorem, Engle and Granger (1987). The negative EC sign implies that a fall in excess
money in the last period will increase the level of desired money holdings in the current
period. In other words, any particular disequilibrium will be reduced over time. The

results of diagnostic tests concerning autocorrelation and normality are already presented

in Table 29 and 30.

5. Conclusion
This chapter has examined the empirical relationships between money and other
macroeconomic variables in South Korea and Malaysia, using the residual-based

cointegration tests based upon the results of the unit-root tests such as ADF-t, ADF

144

normalized bias test, the PP Z, and Z p tests of a unit root against trend stationarity and

the KPSS tests of trend stationarity and applying Johansen’s procedure along with the
VECM approach.

Whereas the residual-based cointegration tests indicate a stable cointegration
relationship among the variables in the model regardless of the inclusion of a
deterministic trend in South Korea, the Johansen’s likelihood ratio tests of cointegration
rank do not show any cointegration relationship in the model without a deterministic time
trend. The result of the Johansen’s likelihood tests imply that whether there is a time
trend or not in South Korea is important for the conclusion of cointegration with stable
coefficients. However, in Malaysia, Johansen’s test detects a cointegration relationship in
the model with or without a deterministic time trend. The graphical evidence of stability
of parameters in the real money demand equation confirms that there is a stable
cointegration relationship among the variables in Malaysia. However, the parameters are
not stable in the model without a deterministic time trend in South Korea.

Based on the findings from the tests for weak exogeneity, a single equation
reduced form model is formulated to analyze the short-run dynamics for Arm2. Although
a considerable number of variables turn out to be insignificant from the t-statistics in the
estimation, the error-correction term’s sign is negative for both South Korea and
Malaysia, validating the significance of the cointegration relationship.

In conclusion, the results suggest that both long and short-run models can be
specified for both South Korea and Malaysia. However, due to the results of the

cointegration tests and the tests for the stability of the parameters in the model, we cannot

145

determine whether the model without deterministic time trend can be used to explain real
money demand in South Korea. The use of monthly data, previously not implemented in
earlier work on South Korea, helps to discover an unstable long-run relationship in the
real money demand (See Arize, 1994). Finally, the estimates of a constant term, and
coefficients of real income, interest rate and interest rate differential were made in the

procedure of testing the stable long and short-run relationships in this chapter. These
estimated constant term, 520 , and coefficients, a, , a, and a, will be used in Chapter VI

to obtain the shadow exchange rate and the probability of collapse.

146

Table 23. Dickey-Fuller tests for unit roots: Model I

Model 1: Regression with an intercept (South Korea)

 

 

 

 

 

 

Series Lugs (3 t-test Tom — 1) Q
(It) p-value
rm2 2 0.999 -0.920 -0.229 0.634
y 6 0.996 -2.445 -0.886 0.117
i 0 0.963 -2.163 -8.973 0.424
if 0 0.950 2594 42.185" 0.242
Critical
values
5% -2.88 -14.0
10% -2.57 -1 1.2

 

Notes. An "‘ indicates significance at 10% level.

(a) rm2 is a real money demand.

(b) rm2 is from 1970201 through 1996212; a total of T=324;y is from 1970201 through 1996212; atotal of
T=3242 i is from 1976208 through 1996212; atotal of T=245; if is from 1976208 through 1996212; atotal

 

 

 

 

 

 

of T=245.
Model 1: Regression with an intercept (Malaysia)
Series Lags d t-test TC((i - 1) Q
(k) p-value
rm2 0 1.000 0.1 10 0.035 0.520
y 4 1.000 0.374 0.200 0.001
i 3 0.924 -2440 -47003“ 0.053
if 3 0.947 -2.l8l -27.545 0.162
Critical
values
5% -2.88 -l4.0
10% -2.57 -l 1.2

 

Notes. An ** indicates significance at 5% level.
(a) rm2 is from 1970201 through 1996212; a total of T=324; y is from 1971201 through 1996212; a total of
T=312;i is from 1970201 through 1996212; a total of T=3242 if is from 1970201 through 1996212; a total
of T=324.

147

Table 24. Dickey-Fuller tests for unit roots: Model 11

Model 11: Regression with an intercept and a linear trend (South Korea)

 

 

 

 

 

 

Series Lugs (2 t-test T C( & - 1) Q
(If) p-value
rmZ 2 0.958 3.1 10 -10.I9l 0.449
y 6 0.985 -I .304 -7.328 0.116
1' 0 0.952 -2.422 -I 1.750 0.439
if 0 0.951 -2.546 -12.039 0.237
Critical
values
5% -3.43 -21.4
10% -3.13 -l8.0

 

Notes. An * indicates significance at 10% level.
(a) rm2 is from 1970201 through 1996212; a total of T=324; y is from 1970201 through 1996212; a total of
T=324; i is from 1976208 through 1996212; a total of T=245; if is from 1976208 through 1996212; a total

 

 

 

 

 

 

of T=245.
Model 112 Regression with an intercept and a linear trend (Malaysia)
Series Lags é t-test Tc(& — 1) Q
(’0 p-value
rm2 0 0.989 -1 .251 -3.558 0.494
y 2 0.902 -2.81 1 52.570" 0.001
i 3 0.896 -2.861 .62618W 0.056
if 3 0.927 -2.626 -37.021 I 0.162
Critical
values
5% -3.43 -21 .4
10% -3.13 -18.0

 

Notes. An ** indicates significance at 5% level.
(a) rm2 is from 1970201 through 19962122 a total of T=324; y is from 1971201 through 1996212; a total of
T=312; i is from 1970201 through 1996212; a total of T=324; if is from 1970201 through 1996212; a total
of T=324.

Table 25. Phillips-Perron tests for unit roots

Model 1: Regression with an intercept

 

 

 

 

 

 

 

a Z Z
Series I p
S. Korea Malaysia S. Korea Malaysia S. Korea Malaysia
rmZ 0.999 1.000 -0.823 0.1 12 -0.316 0.035
y 0.996 0.996 2827? -0.564 -1.052 -0443
i 0.963 0.820 -2.058 5132" -8.077 -47.259' I
if 0.950 0.902 -2.639 -3.526" -l2.607 43.693"—
Critical
values
5% -2.88 -14.0
10% -2.57 -1 1.2

 

Notes. An ** (*) indicates significance at 5%(10%) level.

(a) For South Korea, rm2 is from 1970201 through 1996212; 3 total of T=324; y is from 1970201 through
1996212; a total of T=324; i is from 1976208 through 1996212; a total of T=245; if is from 1976208 through
1996212; a total of T=245.

(b) For Malaysia, rm2 is from 197020] through 1996212; a total of T=324; y is from 1971201 through

1996212; a total of T=312; i is from 1970201 through 1996:12; a total of T=324; if is from 1970201 through
1996212; a total of T=324.

Model 112 Regression with an intercept and a linear trend

 

 

 

 

 

 

 

a? Z Z
Series I p
S. Korea Malaysia S. Korea Malaysia S. Korea Malaysia
rm2 0.967 0.989 -2.891 -1.268 -15.612 -3 .641
y 0.975 0.752 -1775 -6.520" -4537 43.239"—
i 0.952 0.768 -2.330 -6.249" - 10.869 -68.93 1 n
if 0.951 0.973 -2.582 -2.289 -12.408 -10.524
Critical
values
5% -3 .43 -21.4
10% -3.13 - l 8.0

 

Notes. An ** indicates significance at 5% level.
(a) For South Korea and Malaysia, rm2, y, i, and if have the same sample size as the above model I.

149

Table 26. KPSS tests for stationarity 2 Model I

Model 1: Regression with an intercept (South Korea)

 

Lag truncation parameter (I)

 

 

 

 

 

 

2 4 5 6 8 15 16
Series
in, 2 5% critical value is 0.463
10% critical value is 0.347
rm2 10.802“ 6.527" 5.458W 4.695" 3.677“ 2.119" 2000"
y 10.616" 6.416" 5.365" 4.614" 3.613" 2.081W 1.965"
1 3.051" 1.875" 1.580" 1.369" 1.087" 0.650" 0.612"
if 1.351" 0.842" 0.715" 0.625W 0.503" 0.31 1 0.296

 

Notes. An ** (*) indicates significance at 5%(10%) level.

(a) rm2 is from 1970201 through 1996212; a total of T=324; y is from 1970201 through 1996212; a total of
T=324; i is from 1976:08 through 1996212; a total of T=245; if is from 1976208 through 1996212; a total

 

 

 

 

 

 

 

6172245.
Model 1: Regression with an intercept (Malaysia)
Lag truncation parameter (I)
2 4 5 6 8 15 16
Series
73,, 2 5% critical value is 0.463
10% critical value is 0.347
rmz 10.652" 6.440" 5.387" 4.634" 3.631" 2.096" 1.978"
y 10.316" 6.239“ 5.219" 4.490" 3.518" 2.032" 1.919"
1 3.355“ 2.112" 1.790" 1564““ 1.263" 0.795" 0.759"
if 3.202" 1.998" 1.691" 1.473" 1.182" 0.732" 0.698"

 

Notes. An "”" (‘) indicates significance at 5%(10%) level.

(a) rm2 is from 1970201 through 1996212; a total of T=324; y is from 1971201 through 1996212; a total of
T=312;i is from 1970:01 through 1996212; a total of T=324; if is from 1970201 through 1996:12; a total

of T=324.

150

Table 27. KPSS tests for stationarity: Model 11

Model 112 Regression with an intercept and a linear trend (South Korea)

 

Lag truncation parameter (I)

 

 

 

 

 

 

2 4 5 6 8 15 16
Series
I}, 2 5% critical value is 0.146
10% critical value is 0.1 19
77712 0.277" 0.171" 0.146" 0127* 0.103 0.069 0.066
y 2.005" 1.219" 1.022" 0.881 " 0.69?" 0.407" 0.3 86"
i 1.136" 0.704" 0.596" 0.519" 0.415" 0.253" 0.240"
if 1.357" 0.846" 0.718" 0.627" 0.505" 0.312" 0.297"

 

Notes. An ** (*) indicates significance at 5%(10%) level.

(a) rm2 is from 1970201 through 1996:12; a total of T=324; y is from 1970201 through 1996:12; a total of
T=324; i is from 1976:08 through 1996:12; a total of T=245; if is from 1976:08 through 1996:12; a total

 

 

 

 

 

 

 

 

 

 

6172245.
Model 112 Regression with an intercept and a linear trend (Malaysia)
Lag truncation parameter (I)
2 4 5 6 8 15 16
Series
{7,2 5% critical value is 0.146
10% critical value is 0.1 19
rm2 1.507" 0.914" 0.76?” 0.660" 0.519" 0.304" 0.289"
y 1.497" 0.951" 0.810“ 0.707W 0.570" 0.357” 0.341"
i 0.575" 0.367" 0.313" 0.275" 0.225" 0.148" 0142*
if 0.651" 0.410“ 0.348" 0.305" 0.247" 0.157" 0.151"—

 

Notes. An ** (*) indicates significance at 5%(10%) level.

(a) rm2 is from 1970201 through 1996212; a total of T=3242y is from 1971201 through 1996212; a total of
T=312;i is from 1970201 through 1996212; atotal of T=324; if is from 1970201 through 1996:12; atotal

of T=324.

151

Table 28. Testing for no cointegration in demand for real M2

 

Model: rm2;l = a0 + a,y, +a2i, +a3if, +[fltrend]+u,

 

OLS estimates

Cointegration tests

 

 

 

 

 

 

a0 y, i, if, trend 76.3 Z, Z p
South 2.384 1.042 -0007 0.012 0.991 -4523" 87.146"—
Korea (0.047) (0.009) (0.002) (0.002)
3.804 0.368 -0007 0.010 0.006 0.997 -4569" -36.371*_
(0.077) (0.035) (0.001) (0.001) (0.000)
Malaysia 1.924 1.157 0.041 -0023 0.980 _ _
(0.042) (0.011) (0.003) (0.002)
4.601 0.021 0.024 -0017 0.008 0.991 _ _
(0.138) (0.058) (0.002) (0.002) (0.000)
Critical
Values
5% -4.16 -322
(-449) (-377)
10% -3.84 -27.8
(-420) (-332)

 

Notes. An ** indicates significance at 5% level.

( ) means the critical value when the regression includes a deterministic trend.

152

Table 29. Residual misspecification tests (South Korea)

 

Model: Ax, = FIAX,_1 +”.+Fk-1Axl—k+l +Hx,_] +110 +8,

 

 

 

 

 

Eq. 5.5.59 Skb Ek Q ,c
P-value
Arm2 0.014 -0.117 3.563 0.848 3.704
Ay 0.029 -0.588 7.703 0.078 233.979"
Ai 1.068 0.328 4.988 0.756 43.634"
Aif 1.204 0.081 5.129 0.202 45.387"

 

MOdCl: Ax, = rle,_l +...+Fk-1Axl—k+l +a(fl',fll)(x;_l,t)'+u0 +8,

 

 

 

 

 

Eq. s.E.Ea Skb Ek Q 1c
P—value
Arm2 0.014 -0.015 3.069 0.909 0.056
Ay 0.029 -0.481 7.809 0.289 234.487I '
Ai 1.048 0.304 4.193 0.905 17.498' '
A17 1.157 0.265 4.123 0.972 15.034| '

 

Notes. An ** indicates significance at 5% level. k=5 for the first model and k=10 for the second model.
a. S.E.E denotes the standard error of regression estimate.
b. Sk and Ek are the skewness and kurtosis statistics.
c. The Jarque and Bera test for normality (Jarque and Bera, 1980),

151:2

T — m
= Sk 2 + — ~ (2 where m is the number of regressors.
4 Z

6

 

T

153

Table 30. Residual misspecification tests (Malaysia)

 

MOdClI Ax, = Fle,_, +..'+Fk-1Ax!-k+1 +17x,_1 +110 +8,

 

 

 

 

 

 

 

 

 

 

 

Eq. 3.5.5a Skb Ek Q zc
P-value
Aer 0.014 -0.018 2.882 0.899 0.193
Ay 0.042 0.309 3.707 0.003 11.143"
Ai 1.186 0.289 8.416 0.999 374.529"
211/ 1.352 0.342 7.090 0.989 217.114"
MOdCI: Ax, =FIAX,_1 +.H+Fk-1Axt-k+l +a(fl',fll)(x;_l.t)'+u0+£t
5q. 5.5.5al Skb Ek Q zc
P-value
Arm2 0.014 -0014 2.870 0.898 0.222
Ay 0.042 0.204 3.496 0.003 5.196"r
Ai 1.187 0.228 8.255 0.997 351.240"
Alf 1.352 0.354 7.032 0.990 211.566"

 

Notes. An *(”) indicates significance at 10% (5%) level. k=8 for all the models.
a. S.E.E denotes the standard error of regression estimate.
b. Sk and Ek are the skewness and kurtosis statistics.
c. The Jarque and Bera test for normality (Jarque and Bera, 1980),

Ek2

T — m
= Sk2 + — ~ 2 where m is the number of regressors.
4 Z

T

 

154

Table 31. Test of the cointegration rank (South Korea)

 

MOdCII Ax, = rle,_, +H.+Fk—1AxI-k+l +nx,_l +110 +8,

 

 

 

 

 

H2 eigenvalue trace 2mm,
r = 0 0.097 43.641 24.470
r SI 0.060 19.171 14.750
r S 2 0.017 4.421 4.062
r S 3 0.002 0.359 0.359

 

Model: Ax, = FIAX,_1 +--°+Fk_1Ax,_k+1+a(fl',fl,)(x;_l,t)'+u0 +6,

 

 

 

 

 

H2 Eigenvalue trace zlmax

r = 0 0.127 67.09?" 31.775"
r 51 0.084 35.320 20.572
r S 2 0.044 14.748 10.585
r53 0.018 4.163 4.163

 

Notes. An“ (**) indicates significance at 10%(5%) level.
k=5 for the first model and k=10 for the second model.

155

Table 32. Test of the cointegration rank (Malaysia)

 

MOdClI Ax, = rlet—1+.H+Fk—1Axt—k+l +HX,_I +110 +8,

 

 

 

 

 

H2 eigenvalue trace xlmax
r = 0 0.083 49.591" 26.304"
r S I 0.054 23.287 16.931
r S 2 0.018 6.357 5.491
r S 3 0.003 0.866 0.866

 

Model: Ax, = Fle,_, +...

+rk—1Ax1-k+l +64.53.51 )(xi—IJYHIO +81

 

Eigenvalue

trace xl

 

 

 

 

max
r = 0 0.091 64.826" 29.022
r st 0.068 35.803 21.349
r s 2 0.037 14.454 11.553
r s 3 0.010 2.902 2.902

 

Notes. An *(**) indicates significance at 10%(50/6) level.
k=8 for all the models.

156

Table 33. Normalized cointegrating vectors ( ,8) and error correction coefficient (a)
(South Korea)

 

Model: Ax, = F,Ax,_, +---+F,,_,Ax,_k+, +17x,_, +u0 +8,

 

 

 

 

rm2,_, y,_, i,_, ij Constant
[3 1.000 0.447 0.374 —0.297 -I 1.272
(3.863) (0.951) (0.736)
Eq. Arm2 Ay Ai Aif -
. -0.003 -0.008 -0.042 0.145 -
(0.001) (0.003) (0.103) (0.116)

 

MOdCI: Ax, = F,Ax,_l +"'+rk_le,_k+1+a(fl’,fll)(x;_l,t)'+llo +8,

 

 

 

 

rm2,_, y,_l l,_l if,_, Constant Trend
fl 1.000 —0.037 0.003 -0.005 -4.474 -0.008
(0.1 12) (0.004) (0.003) (0.001)
Arm? Ay Ai Aif - -
,. -0.104 -0.067 6.872 3.245 - -
(0.029) (0.061) (2.224) (2.459)

 

Notes. k=5 for the first model and k=10 for the second model.

157

Table 34. Normalized cointegrating vectors ( fl ) and error correction coefficient (a)

 

 

 

 

 

(Malaysia)
Model: Ax, =Fle,_l +”'+Fk-1Axi—k+l +17x,_] +110 +8,
rm2 ,_, y ,_l 1f,_, Constant
,8 1.000 -1 .133 -0.089 0.036 -1 .732
(0.039) (0.013) (0.009)
Eq. Arm2 21y Aif -
d, -0.013 0.068 -0.720 -
(0.008) (0.025) (0.700) (0.798) ( - )

 

MOdClI Ax, = Fle,_, +H.+Fk—1Axt—k+1 +a(fl',fll)(x;_l,t)’+u0 +8,

 

 

 

 

rm2,_, y,_, ' i,_, if,_, Constant Trend
fl“; 1.000 -4.090 -0. 149 0.072 5.396 0.022
(0.889) (0.033) (0.020) (0.007)
Eq. Arm2 A y Ai A if - -
. -0.007 0.043 -0.054 -0.299 - -
(0.004) (0.01 1) (0.300) (0.342)

 

Notes. k=8 for all the models.

158

Table 35. Weak exogeneity test (South Korea)

 

MOdClI AX, = F1Ax1—1 +---+Fk_|Ax,_k+1+17x,_, +110 +8,

 

 

al=0 02:0 a3=0 054:0
LR 4.892" 4.173" 0.147 1.016
(P-value) (0.027) (0.041) (0.701) (0.313)

 

Model: Ax, = Flel-l +---+F,,_,Ax,_k+l +a(fl',fl,)(x;_l,t)'+u0 +8,

 

 

C1120 02:0 03:0 014:0
LR 7.475" 0.795 7.540" 0.122
(P-value) (0.006) (0.373) (0.006) (0.290)

 

Notes. An *(**) indicates significance at 10% (5%) level.

Table 36. Weak exogeneity test (Malaysia)

 

Model: Ax, = F,Ax,_, +---+ Fk_,Ax,_,,+, +17x,__, +u0 +8,

 

 

a|=0 a2=0 a3:0 04:0
LR 2.093 4.494"! 0.141 0.488
(P-value) (0.148) (0.034) (0.707) (0.485)

 

MOdCl: Ax, =P,Ax,_,+~-+I‘,,_,Ax,_,,+,+a(fl',fl,)(x;_1,t)'+u0 +8,

 

 

011:0 a2=0 a3=0 a4=0
LR 2832* 6.577" 0.017 0.542
(P-value) (0.092) (0.010) (0.897) (0.462)

 

Notes. An *(**) indicates significance at 10% (5%) level.

159

Table 37. Estimated coefficients of short-run model (South Korea)

 

Model: Arm2, :21“ + a(L)Ai,_, + B(L)Ay,_1 + 7(L)Aifl_1 +aj(,3'.~,51 )(x,'-,,t)' + 771

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

coefficient Std. Error t-value
A rm H 0.002 0.072 0.025
Arm,_2 0.1 17 0.072 1.626
Arm,_3 0.043 0.072 0.594
Arm,_4 -0.037 0.073 -0.507
Arm,_5 0.033 0.073 0,454
Arm,_6 0.073 0.073 0.996
Arm,_7 0.026 0.073 0.356
Arm,_8 0.002 0.073 0.031
Arm,_9 0.056 0.074 0,754
Armt—IO 0.125 0.074 1.692
Ay,_, 0.000 0.035 0.004
Ay,_2 0.004 0.038 0.1 17
Ay,_3 -0.049 0.039 -1.252
Ay,_4 0.010 0.039 0.249
Ay,_5 0.059 0.039 1.532
Ay,_6 0.067 0.039 1.746
Ay,_7 0.083 0.039 2.137
Ayi—8 0.062 0.038 1.632

 

I60

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

4th 0.053 0.037 1.437
431—10 0.040 0.034 1.171
Ai,_, 0.001 0.002 0.320
211', ._, 0.002 0.002 1.028
Ai,_3 0.001 0.002 0,534
A,“ -0.002 0.002 -1233
Aj,_5 0.003 0.002 1.442
Aj,_b -0.002 0.002 -1.033
A,” 0.000 0.002 0.113
A,” -0004 0.002 -2054
Ai,_9 0.001 0.002 0.583
Ail—10 -0003 0.002 -1419
21 if,_, 0.001 0.002 0.599
Aif,_2 -0.000 0.002 -0.164
Aif,_3 -0001 0.002 -0722
Aif,_4 0.003 0.002 1.560
Aif,_5 -0.003 0.002 -1.790
Agf,_6 0.003 0.002 1.761
4,734 0.000 0.002 0.095
[jg/4L8 0.005 0.002 2.559
Aif,_9 -0.001 0.002 -0544
Aif,_,0 0.005 0.002 3.193
“0 0.002 0.002 0.884
a, -0104 0.029 -3.558

 

I61

Table 38. Estimated coefficients of short-run model (Malaysia)

 

MOdCl: Arm2, = NO +(1(L)Ai,_l + fl(L)Ayt—l + y(L)AIf,_I +al(fl’, fl] )(x;_l,’)' + 77,

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

coefficient Std. Error t-value
ArmH -0.023 0.063 -0.360
Arm,_a -0.061 0.062 -0.977
Arm,_3 0.1 15 0.063 1.824
Arm,_4 0.014 0.063 0.220
Arm,_5 0.046 0.064 0.717
Arm,_6 0.065 0.064 1.012
Arm,_7 -0.063 0.064 -0.988
Arm,_8 -0.022 0.064 -0.343
Ay,_, -0.045 0.022 -1 .993
Ay,_2 -0.021 0.025 -0.825
Ay,_3 -0.022 0.025 -0.857
Ay,_4 -0.040 0.025 -l.577
A y,_ 5 -0.018 0.026 -0.723
100—6 0.013 0.025 0.512
Ay,_7 0.034 0.024 1.410
Ay,_8 0.027 0.019 1.405
Ai,_, 0.001 0.001 0.766
Ai,_2 -0.004 0.002 -1.988
Ai,_3 -0.000 0.002 -0.181

 

I62

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

417—4 -0000 0.002 -0.162
Ai,_5 -0001 0.002 -0293
411-6 -0000 0.002 -0.186
A,” -0.001 0.002 -0450
A 1'74; -0001 0.002 -0499
4,7,4 -0.001 0.002 -0.855
4,7,4 0.003 0.002 1.623
416-3 -0001 0.002 -0322
4177-4 -0000 0.002 -0109
4,1,4 -0.001 0.002 -0317
4,7,6 0.002 0.002 1.032
4,7,, 0.001 0.002 0.551
[lift—3 0.001 0.002 0.480

uo 0.009 0.002 4.629

a, -0.006 0.004 -1.817

 

163

 

.26.: 8:589:83
8:6: 2:: a :23 BooE 65 .8 33:33
6.6% 2: w:oEa 562, wEEEBES 0:0

.EE a

 

 

Goo: mmmv— 35:88
:3 o 2: o :o «:55 _ 88 322:. .. . .
“cm Home “mm—8:.” _.wEoo%.U_mo._ .xucoi Sam: 2 Ammo: A; 83328.86 38 “3.6:: 2.002 ANOONV
:6: wcoEm :30? w::.§w3:_8 oz xooo: 0A. Ema: “RE ”88 8x38 8:82 855 AEQNEVE -72....— E6.
. 2 50%. o
«xi—WWW: up? 33:88 333— EL :2355 .8 28 ©2898 .
omcmsuxo :w_o:8 E: .28 688:. .0985 ”Boo: 82—5 ”33:88.86 2:. 3.23.: 3.8.839
.8. .52 as :2 52: 85.65 .8: :32: Ba $86.3: M82 :6 «A: 25 £32 .8 Eco 2.6565 38. $8:
wcoEm €089; w::§w3:.oo out: 3 03h. ”Chm: >m ”Cum: LD< 2m.— 7682: ”m0 :0 22> EBVE CLUE—>3:— -_umno_ o~t<
. . «Bov—
£30m
635:3 635.;
moo 855250 28m
5:632...
$58? $38. madamewwxx >628: 85.2.
$535 .._ 8583860 flood -2: D 3::5E56D .0825.)— oEEmm :9:on

 

23282 23 «Box

536m 5 "8:20on :ouootoULomeoufion6U wEZo>E 828—3 3:9: :8: :8 8588: 68 .3 863m 80 bmfiasm .0m 03:...

164

8.:00 .0.0.0:.:.00 00.30050: n R... 30.00.. .0.0.0E:.8 n mmu .0008 0.0.8.00 u mu ”.0680
00:00.... 0:0 008000. N 2 0.0.30 0:0 05...... u 0.. 8020.6. n . ”00> 0:0 0_w..m. u >m. ”0.00. :0..0.w0.:.0u
.0..:-..->00.0.D u ”.0 Him 0:0 ..0.E..0m .00........ 00.030.50.32 u mmmv. 62.0.. 0:0 09...... u a... ..0..:-..->00.0.Q 00.:0Ew00 u -..n.< .300. .03.-..:D
8.:00 .06.0E:.00 .0 00.0. .6800 5:05-005... n 295th 0.0. :3 0.0000... H mm... 6.0. .6080 0:... u MGR ”0.0. .m0.0.:.

00.6.00... 6.8.05 .0 0.00:. n n... ”0.0.8.0.. 0.00m
:0..0...0.6-:.->0:0..:0 u 0.). 00.03.30 0.0.083.
.0000 0.0 080030.000 w:.30=8 0.. .- 00.07.

 

.200 80.8.05...

.08.. 0E: 0 5...... 0:0 ....0 0 5.3
0600:. 0... .8 .0..:0.0...0 0.0. .m0.0.:.
0:0 0.0. .0225 0800:. .00. 08:05
.00. wcoE0 .0.00> wc..0.w0.:.8 0:0

.200 000.00%-.03

...:.0:000 :000 0:0

0000.0 .0 600:. 05 .8 85060.80

0.0. 00:08.00 0:0 00005:. .0 0.0.
00.0098 0.0. 7.0.2:. .0580. .00. .x0:0:.
.00. w:0E0 .0.00> wc..0.w0.:.00 0:0

.200 000.00%.03
0.0. 0w:0..0x0 00.00006 0:0 0.0. .862.

0.80:. .00. .33. 0:0 ...>. .02 :0. 00:0...
.00. w:0E0 .0.00> w:..0.w0.:.8 0:0

800.. 2
.800... On.

800......

800::

.000: mm...
.000. z.
.200: .9.

5.3: .9.
88...:

.800: 0.050
0:0 3000...»...

”200.75...

._0..:0.0...0 0.0. .m0.0.:.
0.0. 60.0.: >.0:0.>.

0.0. 00:08.8 .0:.:.0:
”:0..0..:. .0 0.0. 00.00000
0.0. .m0.0.:. :w.0.0. mm.-
:0

0.0. .::00m.0 .3.meth

0.0. 0w:0..0x0 00.0090
Hmm... ”mm... :0 mm...-

.....c_ 2.002....
3...... 0.0.02...

0.0.0.2....
$0.0 0.02.2.5
.80... 0.002....

.2582.
N . H000.
- . H2.0.

35:05:
N . Hm00.
4.5.0.

.00.-.000.

0.00.
-150.

.0000.
:5.

.0000. .
E0...m

2.00 C
:00

0.0.0.0.).

 

I65

Figure 29. The log of real M2 (South Korea)

 

 

 

 

 

707172737475767778798081828384858687888990919293949596

 

Figure 30. The log of real GDP (South Korea)

 

4.5 e

 

3.5 r

2.5

1.5

 

1 . A A; . A
707172737475767778798081828384858687888990919293949596

 

 

r mi“!

 

166

30

25

20

15

wt

5

Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Au
76 77 78 79 80 81

18

16

14

12

10

2

0

Figure 31. Interest rate (South Korea)

 

 

L

.1—

9- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug— Aug-

828384858687888990919293949596

Figure 32. Difference between domestic and foreign interest rate (South Korea)

 

 

 

 

 

 

 

 

Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug- Aug-

76 77 78 79 80 81

82838485868788899091

I67

9293949596

Figure 33. The log of real M2 (Malaysia)

 

7.5

6.5

5.5

4.5

 

 

4 i i oi
707172737475767778798081828384858687888990919293949596

 

“—776?

 

Figure 34. The log of real GDP (Malaysia)

 

4.5 _

3.5 a

2.5

 

 

2 ‘14 A K i i; it
7172737475767778798081828384858687888990919293949596

 

 

1_y

168

Figure 35. Interest rate (Malaysia)

18

 

16

14

12

 

10

 

 

 

0 g 4L g;

 

 

707172737475767778798081828384858687888990919293949596

Figure 36. Difference between domestic and foreign interest rate (Malaysia)

 

 

-11
-13
-15 a L

 

 

 

 

707172737475767778798081828384858687888990919293949596

1

Q; ,.___...

169

Figure 37. Recursive estimates of the long-run parameter of real GDP (South Korea)

_n

O—‘MwbIUIO’VOJCDO

 

u\‘

 

 

 

 

 

l
l
33 l l. l

Jan-85 Jan-86 Jan-87 Jan-88 Jan-89 Jan-90 Jan-91 Jan-92 Jan-93 Jan-94 Jan-95 Jan-96

 

 

:Ttie estirétgcoechient of y _ 95% confidence interval __ 95% confidence interval l

 

Figure 38. Recursive estimates of the long-run parameter of interest rate (South Korea)

 

 

 

 

-1.4 F l

-16 '1 l '

-1.8 l: i
-2 .ll ‘ J

 

Jan-85 Jan-86 Jan-87 Jan-88 Jan-89 Jan-90 Jan-91 Jan-92 Jan-93 Jan-94 Jan-95 Jan—96

 

‘1er estimEo coefficient oTi "4953/0 confidence interval __.__ 95% confidence interval ‘

\—

 

 

170

2
1.8
1.6
1.4
1.2

1
0.8
0.6
0.4
0.2

0 r
-0.2 t

-0.4
-0.6
-0.8

-1
-1.2
-1.4
-1.6
-1.8

1.8

1.6

1.4

1.2

1

0.8

0.6

0.4

0.2

0

Figure 39. Recursive estimates of the long-run parameter of the difference ofdomestic and
foreign interest rate (South Korea)

 

 

 

 

 

l a

 

Jan-85 Jan-86 Jan-87 Jan-88 Jan-89 Jan-90 Jan-91 Jan-92 Jan-93 Jan-94 Jan-95 Jan-96

7'“ the estimated coefficient-of if 495% confidence interval __ 95% confidence interval ’

 

 

 

 

 

Figure 40. Recursive estimates of the long-run parameter of real GDP (Malaysia)

 

 

 

 

 

 

 

ll\~
\\
‘1\
\\ ~ lx/V‘KA
w\,/_\.-' r-‘/ “\»\
\WW
/ v
/.\ 1/ RM
\\ /
\ ../
V’ \V/W

 

Jan-85 Jan-86 Jan-87 Jan-88 Jan-89 Jan-90 Jan-91 Jan-92 Jan-93 Jan-94 Jan-95 Jan-96

 

 

'_ the estimaEd coefficient of y ._ 95% confidence interval ._..__ 95% confidence interval

 

 

17l

Figure 4|. Recursive estimates of the long-run parameter of interest rate (Malaysia)

03

 

0.25

0.2

0.15

 

0.1

 

 

0.05

~0.05

 

-0.1

 

 

 

Jan-85 Jan-86 Jan-87 Jan-88 Jan-89 Jan-90 Jan-91 Jan-92 Jan-93 Jan-94 Jan-95 Jan-96

;_ the estimated coefficient ofi—L:95‘°/;—corifTid_ence interval—:95?) confidence—interval-

 

 

Figure 42. Recursive estimates of the long-run parameter of the difference of domestic and
foreign interest rate (Malaysia)

0.2

 

0.15

0.1

 

 

 

 

 

-0.15

 

-O.2 i .
Jan-85 Jan-86 Jan-87 Jan-88 Jan-89 Jan-90 Jan-91 Jan-92 Jan-93 Jan-94 Jan-95 Jan-96

 

 

 

 

I; the estimated coefficient of if _ 95% confidence—intuewal

95% confidence interval—1

 

 

I72

——.m.~—_—..—— ..-~——.. ..

CHAPTER VI
FORECAST OF SHADOW EXCHANGE RATE AND PROBABILITY

OF COLLAPSE

1. Introduction

Currency crises are thought to have a significant predictable component. The first
generation models of currency crises identify fundamentals judged to be useful in the
prediction of currency crises as typified in the influential paper, Krugman (1979). A
fiscal deficit financed by domestic credit creation is considered to be the root cause of a
speculative attack. Since the monetary authority monetizes the fiscal deficit, the
oversupply of money causes a gradual decline in international reserves. Accordingly,
investors attack the fixed exchange rate with the effect being the depletion of the
government’s reserve holdings needed to defend its currency.

The currency crises in Europe and Mexico during the early 19905, however, do
not lend support to these traditional factors playing a major role in their crises. Otker and
Pazarbasioglu (1996), when thy computed the probability of an exchange rate regime
change using Blanco and Garber’s model, found the Mexican financial crisis in 1994 was
not the result of fiscal imbalances, which had previously played a major role in Mexico’s
balance of payments crises. Instead, it was the rise in private sector indebtedness and a
corresponding increase in the amount of credit owed to the banking system that built up
the pressure in Mexico’s exchange market in mid 1994. Moreover, the experiences of

several European countries in the context of the European Monetary System (Otker and

173

Pazarbasioglu, I997b) show evidence additional triggering determinants of crises - e.g.
pure speculation - that cannot be explained by the fundamentals. These studies focus on
the uniqueness of each country’s currency crisis during a specific time period.

However, while the wide range of analyses of European and Mexican currency
crises have found a consensus about the causes of the crises, the cause of the Asian
currency crisis is still under discussion among researchers. Therefore, the empirical study
in this chapter will apply a basic and an extended model, first introduced in Chapter III, to
the crises in South Korea and Malaysia. Two countries experienced severe devaluations
of their currencies during the Asian currency crisis.

The objective of this chapter is to obtain the probability of an exchange rate
regime change as a function of economic fundamentals by using the implications of the
speculative attack literature to identify the contribution of weak economic fundamentals
in both South Korea‘s and Malaysia’s currency crisis. If the probability of collapse was
high enough to cause the currency crisis in South Korea and Malaysia during the Asian
currency crisis, then support can be found that sound macroeconomic policy may have
been able to prevent the currency crisis. Hence, the arguments made by weak
fundamentalists are justified.

The remainder of this chapter is organized as follows. Section 2 outlines the
estimation procedure. Section 3 presents the empirical results and section 4 offers

concluding remarks.

2. Estimation procedure

I74

To obtain the probability of collapse, it is necessary to estimate the shadow
exchange rates. As shown in Chapter III, the shadow exchange rate derived from the

basic model is

3m = ”71+: ‘00 +0151: “azym ‘17:“ ‘1’: Hal +03)(Ez#:‘il +91“) (1)
where m,+1 = d,” +rc and Euufil + I’m =1},l 4,11.” Then. the probability of collapse,
7r, , is obtained. The probability of collapse is the probability that the shadow exchange
rate, 33+, , will exceed E, 22 in period (t+1) or 7r, = Pr[§',+l — E, )0].

However, the basic model was extended to resolve the problems caused by the
non-stationarity of variables. The shadow exchange rate derived from the extended model
is

3i+1 = "71+! “#0 ’WLMiz-i - MUAyi-i ‘WLlez-i ‘Hixz """z ‘17:“ ““m (2)

, . . ‘ . .‘i‘ 23
where mm =d,+, +rc and If,_] =l,_] -z,_,.

The probability of collapse in the
extended model is defined as same as the one in the basic model.

The estimation of the shadow exchange rate, '5, +1, requires two additional

procedures. The first procedure is to make forecasts of the variables, p. , y, i, i' , d, and u,

which are assumed to evolve according to a period-by-period systematic stationary

 

2' m, d, r, p, p*,and y are the logarithms of the money stock, domestic credit extended by the domestic
banks, central bank foreign reserves, domestic price level, foreign price level, and real output, respectively.
i is the domestic interest rate, 1'" is the foreign nominal interest rate, p is the risk premium on domestic

assets, u is the logarithms of the deviation from PPP and E, #1; is a period-by-period systematic stationary
component of d.
22 This is the time I value of the fixed rate.

23 III is the first row ofthe 17 and x,'_I = (rmH, a0, t, i,_,, y,_,, {fH ).

I75

component. E My, . and a stochastic element, a, . A forecast of these variables is made in
Chapter IV using the ARFIMA(p.d.q)-FIGARCH(P.§.Q) model. These forecasted values
are substituted into equation (1) and (2). The next procedure is an estimation of the
money demand parameters an. a], a3 and (13 from equation (I) and a(L), ,8(L), y(L) and
171 from equation (2). Under the possibility of a spurious regression, the coefficients
estimated in the OLS regression“, 510, [11,512 and [13 are initially used to derive 3', +1 in
equation (1). After testing for the number of cointegration relations and estimating the
cointegrating vectors. the estimated coefficients for the extended model,
(2(L), [3(L), w.) and [7, , are applied to the estimation of 3", +1 in equation (2).

Assuming that each stochastic part of variable, p', y, i, I", d, and u, is
uncorrelated with each other and their linear combination is normally distributed, the

probability of collapse, 7:, , in the basic and extended model can be estimated as

2

___fl_
. 2
1 2m

7r =PI‘[8 )k]= ——-——e ”'de (3)
I HI I EEtaHl‘J-i;

— o I. t
where k, = s, —d, —r(. +a0 —(al +a3)1, +a31, +a2y, +p, +u, —E,/,t,+l , Ehum and
a, H are a linear combination of the systematic stationary components and stochastic

parts of p' , y, i, z" , d, respectively, and u and 0,2,, is a conditional variance of a, +1 .

 

24 Normalized cointegrating vector which has long run equilibrium coefficients, £30 , [I] , 512 and (33 , is
also used for the comparison with the result derived by OLS estimates.

I76

3. Empirical results
3.1 Behavior of variables in the structural model

Before the results of the estimation of the shadow exchange rate and the
probability of collapse are presented, a graphical inspection of the traits of each variable
in the structural model needs to be discussed. Figures 43 and 48 show the real M225 of
South Korea and Malaysia, respectively. The real value of M2 is persistently increasing
for both countries up until the collapse. Then, for both countries, the real M2 turned
downward right after the crisis. After a few months, though, the real M2 of South Korea
began to grow again at a steeper rate than the rate before the currency crisis. By contrast,
the real M2 kept growing slowly in Malaysia.

Figures 44 and 49 display the domestic credits26 of South Korea and Malaysia.
They show that domestic credit increases at a faster rate in South Korea than Malaysia
after the collapse. This indicates that a currency depreciation pressure induced by
growing domestic credit began to increase again in South Korea. In addition, the decline
in domestic credit following the crisis in both countries shows that the second-generation
models’ view that speculators would expect an immediate increase in domestic credit
after the crisis is not supported by the data.

Figure 45 and 50 show that real GDP27 for both South Korea and Malaysia

declined for a while after the currency collapse. However, it took longer for real GDP to

 

25 Logarithm value.
26 Logarithm value.
27 Logarithm value.

177

recover than it took for the real M2 and domestic credit to begin to rise again. In addition,
even after it began to rise again, both countries’ real GDPs show an unstable tendency.
The abrupt increases in the interest rates in both countries at the point of collapse
in Figure 46 and 51 are not surprising when one considers the bottleneck that the
currency crisis caused in the financial market. Nevertheless, interest rate movements after
the collapse show stability during the recovery period from the collapse. The deviation
from PPP, the negative value of the real exchange rate, mimics, in the opposite direction,
the movement of the interest rates at the point of collapse in both countries. Whereas the
currency did depreciate in real terms by 9.5 percent in South Korea during 1996”, there
was a substantial appreciation to the real exchange rate relative to the 1980’s real
exchange rate during the 19903 prior to the collapse as shown in Figure 47. This
encourages us to anticipate a higher probability of collapse in the 19903 than in the
19803. In addition, the deviation from PPP in Malaysia shows that the real exchange rate
depreciation that continued during the 19803 turned to appreciation in early 1992 in
Figure 52. Therefore, the real exchange rate’s graphical appearance for both countries

may have signaled the upcoming Asian currency crisis.

3.2 Estimated shadow exchange rate
The shadow exchange rate, 3H1, at time t+1, is the floating rate crisis”, that

would clear the foreign exchange market if the central bank stops defending its fixed

 

28 Refer to Table 8.
29 Refer to Chapter III.

178

parity. As discussed in Chapter III, in the first generation models of currency the shadow

exchange rate is an index used by speculators to decide when to attack. This is because

the condition for profitable attack is when the postcollapse exchange rate, 2?, +1 , is larger

than the prevailing fixed rate. 3,. Profits of speculators are equal to the exchange rate

differential multiplied by the reserve stock used to defend the fixed rate regime.

The second and third columns of Table 40 report the quarterly data of actual and
shadow exchange rates30 of South Korea in 19903. The difference between the actual and
shadow exchange rates was not noticeable until March 1994. However, after March 1994,
the difference became larger and the actual and shadow exchange rates were 6.82 and
7.04 in September 1997, 3 months prior to South Korea’s currency crisis. Therefore, the
shadow exchange rate, commonly interpreted to reflect weak fundamentals, gave a
warning signal to the policy makers before South Korea’s currency crisis.

Figure 55 shows the actual and shadow exchange rates in South Korea from
1977208 to 2000:11. Three methods are used to derive of shadow exchange rates; the first
uses the coefficients from the OLS estimation; the second uses the normalized long-run
cointegrating vector; and the third uses the short-run model of real money demand
function. Although different coefficients are used to estimate the shadow exchange rates,
the different estimated shadow exchange rates exhibit no substantial differences during
the period. The graphical trend of the difference in the actual and shadow exchange rates
shows an increase in the difference before the currency crisis. This implies that the

currency crisis in 1997 may have been predictable.

I79

Similarly the third and fourth columns of Table 40 present a comparison of the
actual and shadow exchange rates of Malaysia during the 19903. The difference between
the actual and shadow exchange rate was increasing since March 1992. The actual and
shadow exchange rates became 0.93 and 1.06 just before the currency crisis in Malaysia.
One can, thus, interpret the divergence between shadow and actual exchange rates in
Malaysia as an indication that there was going to be a currency crisis.

Figure 56 provides a graphical inspection of the actual and shadow exchange rates
from 1971 :02 to 2000211 for Malaysia. It shows similar features in the1970s as in South
Korea. However, while the Won3| continued to stay its highly depreciated level
throughout the 19803, the Malaysian Ringgit converged to the shadow exchange rate
during the 19803. With Malaysia’s capital transaction liberalization still incomplete
during the 19803, its trade balance surplus helped to resist the devaluation of the Ringgit.
Then, as Malaysia’s financial system became fairly well-developed with the 19703 and
19803’ liberalization and innovation, the capital account transaction surplus matched their
growing current account deficit32 during the 19903. This happened just as it did in South
Korea. Therefore, despite Malaysia’s ongoing current account balance deficit in 19903
before the crisis, the Ringgit did not depreciate enough to end the deficit. The divergence
between the shadow and actual exchange rates similarly implies weak fundamentals in

the Malaysian economy that provided the opportunity for speculators to attack the foreign

 

30 Logarithms
3' The South Korean currency
32 See Table 1.

I80

exchange market. Hence, the trend of shadow exchange rate displayed in Figure 50 also

provides enough reason for the outbreak of the currency crisis in Malaysia, 1997.

3.3 Estimated probability of collapse

The second and third columns of Table 41 present the probabilities of a collapse
in South Korea based on the coefficients estimated by OLS and ECM, respectively. As
anticipated before the estimation, the estimated probability hovered around 1.0 for the
three years before the currency crisis in South Korea. The probabilities also show that the
devaluation of the currency decreased the probability of a currency crisis to a low level
when South Korea accepted a free foreign exchange rate regime in December 1997.

Figures 57 to 59 also show the probabilities of collapse in South Korea.
Probability I is based on the coefficients from the OLS estimation and probabilities II and
III are base on the coefficients from the normalized long-run cointegrating vector and the
short-run model of real money demand function estimated by an ECM. While the
techniques used to estimate the coefficients are quite different, the results show a similar
trend among the three probability series. The probability series in Figures 57 to 59
indicate another warning signal following the crisis. A depreciation pressure caused by
weak fundamentals began to grow again after the crisis in 1997 and, around July 1999,
reached the same level as it was at right after the Asian currency crisis. Therefore, the
current ongoing devaluation of currency in South Korea may be explained by our model
as due to continuing weak fundamentals.

Similarly, the probabilities of collapse in Malaysia based on same techniques used

for South Korea are presented in the fourth and fifth columns of Table 41. The probability

181

of collapse in Malaysia was rather high for the years, 1996-97, prior to the crisis.
Therefore, the depreciation pressure from weak fundamentals may have been cumulated
before the currency crisis as an indication of the likely onset of currency crisis. Figures 60
to 62 present the probabilities of collapse in Malaysia. However, unlike the probability
series estimated for South Korea, one of the probability series in Figure 61 estimated with
a the normalized long-run cointegrating vector does not appear to predict the 1997 crisis
as well as the other probability series. If we consider that the coefficients are normalized
and the forecasted real money demand using the normalized cointegrating vector does not
fit well generally, one would expect the estimated probability not to be completely
reliable. For about two years, following South Korea, the probability series settled at the
highest level before the regime change in Malaysia over the years 1996—97. Then, the
probability series declined to a new low level directly following the steep currency
devaluation in July 1997. This upward trend of probabilities also indicates that
Malaysia’s economic condition may have been unstable before the crisis in 1997.

While South Korea chose to move to a free exchange rate regime after the crisis,
Malaysia returned to a fixed exchange rate system with a strict capital transaction
regulation. However, unlike South Korea, there was no devaluation pressure under the
fixed exchange rate regime after the collapse in 1997 as shown by the probability series
in Figures 60 to 62. The lack of devaluation pressure implies that the Malaysian
government’s economic policies designed to stabilize their economy have been strongly

effective.

4. Conclusion

182

Using the speculative attack model previously applied by Blanco and Garber
(1986), Goldberg (1994), and Otker and Pazarbasioglu (1996, 1997b), we estimated the
shadow exchange rates and probabilities of collapse in South Korea and Malaysia
respectively in this chapter. However, the model employed in this chapter is modified to
overcome the spurious regression problem by utilizing the error correction model (ECM)
from Engel and Granger (1987).

A simple graphical investigation of important macroeconomic variables in the
model initially presents evidence that the Asian currency crisis may have been
predictable. The domestic credit of each country, which is one of the indispensable
variables causing currency crisis, showed a steady increasing tendency before the Asian
currency crisis. This implies that the depreciation pressure from the oversupply of
domestic money was cumulating gradually before the crisis. The deviation from PPP for
both countries also signaled the possibility of an currency crisis by showing that the
domestic currency had excessively appreciated in the 19903.

The estimated shadow exchange rate and probability of collapse for each country
reflects the disequilibrium between real money supply and demand. This disequilibrium
provided a signal of upcoming severe currency depreciation. In particular, the shadow
exchange rate of each country was so far above the regulated exchange rate for about two
to three years before the crises that this might attract speculators to the potential profit
attainable following the outbreak of crisis. The probability of collapse, strongly positively
correlated with the difference between the shadow and actual exchange rates, indicates
that both countries, South Korea and Malaysia, as of the early 19903 fell under severe

depreciation pressure that then lead to the Asian currency crisis likewise.

183

In conclusion. based on the evidence presented by the data, the movement over
time of the shadow exchange rates and the probability of collapses, it is confirmed that
fundamentals were weak prior to the Asian currency crisis in 1997. However, even
though weak fimdamentals are strong indicators of upcoming currency crisis, the specific
point of the outbreak of the Asian currency crisis is not predicted by the evidence given in
this chapter. Therefore, unexpected events such as bank failure, corporate failure and
political uncertainty under weak fundamentals are additional factors for the ignition of the

currency crisis.

184

Table 40. Actual and shadow exchange rates

 

 

 

 

 

 

 

 

 

South Korea Malaysia
Actual Shadow Actual Shadow
exchange rate exchange rate exchange rate exchange rate
1990- March 6.55 6.45 1.00 0.97
June 6.57 6.48 1.00 0.97
September 6.57 6.50 0.99 0.98
December 6.57 6.60 0.99 0.99
1991- March 6.59 6.59 1.02 1.00
June 6.58 6.61 1.02 1.00
September 6.61 6.62 1.01 1.01
December 6.63 6.69 1.00 1.01
1992- March 6.65 6.66 0.95 1.00
June 6.67 6.68 0.92 1.00
September 6.67 6.67 0.92 1.00
December 6.67 6.68 0.96 1.00
1993- March 6.68 6.67 0.95 1.00
June 6.69 6.68 0.95 1.01
September 6.70 6.73 0.94 l .01
December 6.69 6.72 0.99 l .03
I994- March 6.69 6.72 0.98 1.03
June 6.69 6.76 0.96 1.03
September 6.68 6.77 0.94 1.04
December 6.67 6.82 0.94 1.04
1995- March 6.65 6.80 0.93 1.05
June 6.63 6.85 0.89 1.05
September 6.64 6.85 0.92 1 .06
December 6.65 6.90 0.93 1.06
1996- March 6.66 6.89 0.93 1.06
June 6.70 6.92 0.91 1.06
September 6.71 6.96 0.92 1.06
December 6.74 6.99 0.93 1 .06
1997- March 6.80 7.02 0.91 1.05
June 6.79 7.02 0.93 1.06
September 6.82 7.04 1.16 l .08
December 7.44 7.24 1.36 1.1 l

 

185

Table 41. Probabilities of collapse

 

 

 

 

 

 

 

 

 

South Korea Malaysia
Probability of Probability of Probability of Probability of
collapse (OLS) collapse (ECM) collapse (OLS) collapse (ECM)
1990- March 0.01 0.00 0.33 0.31
June 0.01 0.00 0.39 0.38
September 0.02 0.02 0.43 0.42
December 0.47 0.87 0.48 0.48
1991- March 0.34 0.55 0.40 0.38
June 0.47 0.84 0.41 0.41
September 0.23 0.70 0.50 0.50
December 0.60 0.99 0.54 0.52
1992- March 0.37 0.61 0.75 0.74
June 0.29 0.57 0.86 0.85
September 0.36 0.59 0.89 0.87
December 0.42 0.71 0.75 0.71
1993- March 0.20 0.33 0.79 0.75
June 0.30 0.41 0.84 0.81
September 0.71 0.88 0.89 0.85
December 0.54 0.71 0.75 0.67
1994- March 0.42 0.81 0.81 0.73
June 0.87 0.99 0.89 0.83
September 0.93 0.98 0.93 0.90
December 1.00 1.00 0.94 0.91
1995- March 0.98 1.00 0.95 0.93
June 1.00 1.00 0.99 0.98
September 1.00 1.00 0.98 0.97
December 1.00 1.00 0.97 0.96
l996- March 1.00 1.00 0.97 0.95
June 1.00 1.00 0.98 0.97
September I .00 l .00 0.98 0.96
December 1 .00 1.00 0.98 0.96
1997- March 1.00 1.00 0.99 0.97
June 1.00 1.00 0.98 0.96
September 1.00 1.00 0.27 0.16
December 0.02 0.02 0.00 0.00

 

186

Figure 43. The log of real M2 (South Korea)

8.5

 

collapse point .
8
7.5 \f/

6.5
5.5

4.5

 

 

 

 

Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan-
70 72 74 76 78 8O 82 84 86 88 90 92 94 96 98 00

real W i

 

Figure 44. The log of domestic credit (South Korea)

13

collapse point _>/
12

11

 

10

 

 

 

 

Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb-
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00

 

domestic credit.

 

187

Figure 45. The log of real GDP (South Korea)

 

collapse point_.’

3.5

2.5

1.5

 

 

 

1

 

Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan-
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00

 

Figure 46. Interest rate (South Korea)

30

 

collapse point—H

25 R
20

15

 

10

 

 

 

0 . -
Aug-76 Aug-78 Aug-80 Aug-82 Aug-84 Aug-86 Aug-88 Aug-90 Aug-92 Aug-94 Aug-96 Aug-98 Aug-00

 

 

 

, interest rate

188

Figure 47. Deviation from PPP (South Korea)

 

Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- JJn- Jan-
70 72 74 76 78 so 82 84 as 88 9o 92 94 96 98 00

-6.6 collapse point _fi

 

 

 

 

 

 

 

deviation from???

 

 

Figure 48. The log of real M2 (Malaysia)

8.5

 

collapse point __’

7.5

6.5 .

5.5 .

 

 

4.5 A

 

 

Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan— Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan-
70 72 74 76 78 80 82 84 86 88 9O 92 94 96 98 00

‘ _’ real MZ ;

189

Figure 49. The log of domestic credit (Malaysia)

13

12.5 collapse point a“

12

 

11.5

11

10.5

10

9.5

9

8.5

 

 

 

8 . .
Feb- Feb- Feb- Feb- Feb— Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb-
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00

 

T'Ido—niefiic credit

Figure 50. The log of real GDP (Malaysia)

5.5

 

collapse point —.

4.5

 

3.5 l

 

2.5

 

 

 

Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan-
71 73 75 77 79 81 83 85 87 89 91 93 95 97 99

8‘1'651’669

,..
l

 

 

190

 

 

18

Figure 51. Interest rate (Malaysia)

 

16

14

12

10

 

0

collapse point —->

 

 

 

 

 

 

 

Jan- Jan- Jan-
70 72 74

Jan-
76

Jan-
78

Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan— Jan-
80 82 84 86 88 90 92 94 96 98 00

 

__ interest rate 1

 

Figure 52. Deviation from PPP (Malaysia)

 

0 72 74

-1.2

-1.4

 

-1.6

76

0 . . .
ngn- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- llan- Jan-
7?

78

80 82 84 86 88 90 92 94 96 98 00
collapse point—p

 

 

 

 

 

deviatTon from PPP—f

 

 

191

"—3

 

Figure 53. US Interest rate

18

 

16

14

12

10

 

 

2

 

0

Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan-
70 72 74 76 78 80 82848688909294969800

 

_‘ ‘ ‘ 'US-Tritérési are '

Figure 54. us CPI

5

 

4.8
4.6
4.4
4.2

4
3.8
3.6
3.4
3.2

 

 

3 . . a
Jan- Jan— Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan-
70 72 74 76 78 80 82 84 86 88 90 92 94 96 98 00

 

 

Figure 55. The actual and shadow exchange rates (South Korea)

 

collapse point
7.5 _

6.5

 

5.5

 

 

 

5 . n i

 

 

Aug-77 Aug-79 Aug-81 Aug-83 Aug-85 Aug-87 Aug-89 Aug-91 Aug-93 Aug-95 Aug-97 Aug-99

__ actual exchange rate _ shadow exchange rate I

-.__._ shadow exchange rate ll .-._..-.___ shadow exchange rate Ill

Figure 56. The actual and shadow exchange rates (Malaysia)

1.6

 

1.5
1.4
1.3
1.2

collapse po'nt __u

1.1 '5; 4w»
1 4;.- 5. .i- ,5
0.9 . M - -Aflkwy

0.8 .
0.7 '
0.6 .
0.5
0.4 o

 

 

 

 

 

Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb-
71 73 75 77 79 81 83 85 87 89 91 93 95 97 99

actual exchange rate __ shadow exchange rate I

 

 

__ shadow exchange rate ll shadow exchange rate III

193

Figure 57. The probability of collapse 1 (South Korea)

1.2

 

Collapse point _>

0.8

 

0.6

 

0.4

0.2

 

0 A _L A

 

 

Aug-77 Aug-79 Aug-81 Aug-83 Aug-85 Aug-87 Aug-89 Aug-91 Aug-93 Aug-95 Aug-97 Aug-99

 

';__ Pl'obability of collapse (OLS) *

 

Figure 58. The probability of collapse 11 (South Korea)

1.2

 

Collapse point-u

0.8

0.6

 

 

0.4

 

0.2

 

0 . .A. r A

 

 

Aug-77 Aug-79 Aug-81 Aug-83 Aug-85 Aug-87 Aug-89 Aug-91 Aug-93 Aug-95 Aug—97 Aug-99

 

1;.Pmbability of collapse (Normalized cohtegrating vector) .

194

Figure 59. The probability of collapse 111 (South Korea)

1.2

 

Collapse point —p

 

0.8 .

 

0.6

0.4

 

 

0.2 .

 

 

 

 

o . . AL . - .
Aug-77 Aug-79 Aug-81 Aug-83 Aug-85 Aug-87 Aug-89 Aug-91 Aug-93 Aug-95 Aug-97 Aug-99

 

 

""___ Probability of collapse (ECM) ”g

 

Figure 60. The probability of collapse I (Malaysia)

1.2

 

Collapse point +

0.8

0.6

 

0.4 l

0.2

 

 

 

o . A g .
Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb—
71 73 75 77 79 81 83 85 87 89 91 93 95 97 99

 

 

"__”:Probabillly of collapse (0le l

195

 

Figure 61. The probability of collapse 11 (Malaysia)

 

Collapse point—b
0.8

0.6 .

0.4

0.2 .

 

 

0

 

 

Feb- Feb- Feb- Feb— Feb- Feb- Feb- Feb— Feb— Feb- Feb- Feb- Feb— Feb- Feb-
71 73 75 77 79 81 83 85 87 89 91 93 95 97 99

 

 

 

Probability of collapse (Normalized cointegrating vecEF) i
1

Figure 62. The probability of collapse III (Malaysia)

1.2

 

Collapse point 2,

0.8

0.6

 

0.4

0.2

 

 

 

 

Feb- Feb- Feb- Feb- Feb- Feb- Feb- Feb Feb- Feb- Feb- Feb- Feb- Feb- Feb-
71 73 75 77 79 81 83 85 87 89 91 93 95 97 99

 

2: Probability of collapse (ECM) l

 

196

CHAPTER VII
COMMON FUNDAMENTALS AND CONTAGION EFFECT IN

CURRENCY CRISIS

1. Introduction

While the traditional approach to currency crisis stresses the decline in
international reserves leading to a collapse of a fixed exchange rate, more recent models
focus on additional variables and the possibility of a contagion effect.

In the aftermath of the 1994 Mexican crisis and the 1997 Asian crisis there was a
widespread contagion between several emerging markets. Clearly, there may be
numerous reasons to expect contemporaneous crises. First, there may be common
external factors. An example would be how policies undertaken by industrial countries
may have similar effects on emerging markets. This is frequently called the "Monsoon
effect". For example, a rise in US. interest rates in 1994 or the devaluation of the yen in
1995 would be common external factors. Second, a crisis in an emerging market may
affect the macroeconomic fundamentals in other emerging markets. This is usually
caused by trade linkages or spillovers related to third market competition. The third factor
of the contagion effect is market sentiment. The herd mentality of investors could explain
part of the contagion. If investors pay little heed to countries’ economic fundamentals,
fail to discriminate properly among countries, a crisis in a neighboring country may

threaten a future domestic crisis.

197

Sachs, Tomell and Velasco (1996) and Tornell (1999) seek to identify
macroeconomic variables that can help explain which countries were vulnerable to
“contagion effects” following the Mexican crisis and the Asian crisis. In addition,
Corsetti, Pesenti and Roubini (1998b), and F urman and Stiglitz (1998) try to explain the
spread of the Asian crisis by finding the common fundamentals for the eruption of the
crisis. Glick and Rose (1998) find that countries with important trade linkages to the
country that first experienced a crisis were more likely to experience a crisis. Masson
(1998) suggests that the contagion effect is unexplained by the common external effects
and that trade linkages played a major role in the Mexican and Asian crises. Table 55
summarizes the findings of 7 selected empirical studies on currency crisis that focus on
the contagion effect.

The first objective of this chapter is to see whether the macroeconomic variables
that explain the cross-country variation in the severity of the crises in Mexico and Asia
generalizes to countries other than emerging market countries. To this end, this study
looks at whether the estimated coefficient in each macroeconomic variable is still
significant in a regression taken from sample of 29 industrialized and developing
countries covering years of the European, Mexican and the Asian currency crises.

The second objective is to examine the extent of contagion effect in all aspects of
common external effects, trade linkages and market sentiment. For common external
effects, a real exchange rate reflecting the devaluation of yen before the Asian crisis and
an interest rate differential indicating the high interest rate of US. before the Mexican
crisis are controlled for in the model. To control for trade linkage, a trade linkage index,

which captures the degree to which the initially attacked country and the home country

198

compete in other markets and the degree to which the two countries directly trade with
each other, is added to the model. To control for market sentiment, a dummy variable is
included in the model.

The third objective is to predict the currency crisis based on the contagion effect
as well as weak fundamentals. To this end, I check for which countries were vulnerable to
speculative attack without contagion effects prior to the Asian currency crisis based on
the result of prediction of a crisis using estimated coefficients from the Mexican currency
crisis in the benchmark regression. Then, the coefficients of variables in the benchmark
regression with all the contagion effect variables are estimated using the countries chosen
in the first step as the initially attacked countries of the contagion instead of Mexico.
Finally, each country’s vulnerability before the Asian crisis is predicted using the
coefficients of variables for the contagion effect and other variables estimated in the

second step.

2. Theoretical framework

To respond to a speculative attack which requires a large supply of foreign
exchange in a market, a country runs down reserves; increases its interest rate; and
depreciates its currency. The first option may be the least costly politically, but it is an
option only available to governments with sufficient reserves to respond to an attack. As
such, a country whose short-run liabilities exceed by far their reserves must choose
between monetary contraction and currency depreciation. Thigh monetary policy that
increases the domestic interest rate makes speculation attack against the currency more

costly in the short run. However, those effects may come at the cost of recession. The

199

extent of recession tends to be severe with rapid lending boom. When the banking
system has a large share of bad loans because of a lending boom, a higher interest rate
leads to a full-scale banking crisis. The existence of both low reserves and a weak
banking system may force the government to close the external imbalance through
currency depreciation. The more currency has previously appreciated, the more the
government should depreciate the currency. This is because it is more likely that firms in
the tradable sector have moved to the non-tradable sector. The movement between sectors
then lowers the response of tradable sector to a real depreciation.

Therefore, the countries with low reserves, 8 rapid lending boom and a severe real
appreciation are more likely to face a speculative attack. In addition, trade linkages with
an initially attacked country or a differential between the domestic and foreign interest

rates may signal to speculators the increased likelihood of currency crisis.

2.] A simple model

Consider an open economy where there are many identical investors who initially
hold an aggregate stock M of deposits denominated in domestic currency that pay an
interest rate 1'. The model is static, with the focus on the interaction between an investor’s
expectation of devaluation and the government’s management of the external account in
the very short run.

In the model, each investor initially selects the stock of domestic deposits she
wishes to hold and the amount she hopes to convert into foreign currency. Then, the

government responds to the capital outflow by running down its reserves, increasing

200

 

interest rate. or depreciating the country’s currency. Finally. investors cash their deposits

plus the interest accrued.

+i

A risk neutral investor will hold domestic deposits so long as 21+i.,

 

'6
1+3]

where t" denotes the foreign interest rate and 3';- denotes the devaluation rate expected by

investor j. In other words, investor j will hold domestic deposits only if the expected

devaluation rate is no greater than the threshold value,

0 It
1—1

In

 

.3 =

(1)

Under this assumption, each investor, who initially holds a stock m of deposits,
can either continue to hold the deposits or withdraw everything. Hence, an investor j’s

strategy would be

my: 0 ”’13:” (2)
—m ifs; >s(i)

where E (i) > 0. In this model, an increase in the interest rate i will make it more likely

that s; is less than 3' . In a symmetric equilibrium, all investors derive the same

conclusion from this common information. Thus, the change in aggregate deposits

A M d is equal to either -M or 0, where M denotes the aggregate initial stock of deposits.

The government has an initial stock R of international reserves. By taking the

behavior of investors. AM d, as given, the government chooses the change in reserves

201

 

t f
“I!
‘

 

 

AR , the depreciation rate 3 , and the unemployment rate u, to minimize the following

social loss function33 (3). subject to equations (4), (5), (6) and (7).

 

min (u + a3) (3)
AR..§',u
d
CA+AM, =AR, ARz—R (4)
l + 3

CA 2 go(rer)3 + Muff) — F(rer) — T(Irade), (p' > 0, F' < 0 (5)

0 < u < 176%), 57w}? < 0 (6)

where CA = current account, rer = real exchange rate, wf= weakness of financial system
and trade = trade linkage with the initially attacked country“.

Equation (3) says that the government minimizes the sum of the depreciation rate

and the unemployment rate, but does not care about the changes in reserves. The

parameter a captures how sensitive the government is to nominal depreciation. Equation

(4) is the identity linking the current account balance, CA, and the capital account,

d
AM . . .
1 . , to the changes in reserves. As shown in equatlon (5), the current account balance
+ s

 

is positively affected by nominal depreciation and unemployment but negatively related
to the real appreciation, F (rer). The term, — F (rer), captures the negative effect of today’s
service on the debt associated with past current account deficits caused by previous real
appreciation. The coefficient (p(rer) indicates how effectively a normal devaluation

would improve the current account. The more the real exchange rate appreciated, the

 

33 This social loss function is quoted from Obstfeld (1997)-

34 We assume that there is a country which was already attacked by the speculative agents and have been in
currency crisis.

202

lower (p. The term —T(trade) reflects the negative effects from the trade linkage with the
initially attacked country. The more the currency of the initially attacked country
depreciates, the weaker the home country’s competitiveness in the international goods
market will be. The existence of an upper bound on the unemployment rate, 17(Wf), in
equation (6) captures the idea that one cannot indefinitely increase unemployment
without causing bankruptcies and a melt down of the payment system. The share of bad
loans in the banks’ portfolio caused by a lending boom is one indicator of the weakness

of a financial system.
There are three possible solutions to the government’s problem, depending on the

size of the reserves. First, if international reserves are sufficient to cover any potential

capital outflow plus the current account deficit or R 2 AMd +F(rer)+T(trade), the

government will be able to close the external gap by spending its reserves. That is,
AR“ = AM" + F(rer) + T(trade)

.5" = 0 (7)

Secondly, when reserves may cover the current account deficit, but are not sufficient to

cover both the current account deficit and the potential withdrawal of deposits or
F (rer)+T(trade) S R SAMd +F(rer)+T(trade), then, a government’s policy will
depend on investors’ expectations of devaluations. If 5"" $3 . then investors will not

attack the currency. Thus, AR*, 3*, u*are the same as in (7). If 5*" > 3 , investors will

attack the currency. Since the government prefers to close the external imbalance by

203

 

spending its reserves, it will wait until R = 0 before pursuing the alternatives when
reserves are not sufficient to close the external gap. Then, the deficit not covered by
reserves must be closed by either depreciation or a recession. Let 45 be the unemployment
rate chosen by the government given the existence of an upper limit of the unemployment
rate. When (D is less than the maximum feasible unempolyment rate (d) S E), the choices
of unemployment and devaluation are given by (9). However, when 05 > 17, 05 is not
feasible any more. Then, as shown in (10), the government sets unemployment at 17 , and

needs to close the external deficits through additional depreciation than would be the case

 

ifCDSE.
AR*=—R (8)
.. M
7r = —l
(p—a if¢327 (9)
36:00

 

. _ —((0+Z)+\[((0+Z)2 —4(p(M+Z)
” ’ 2¢ ifd5>z7 (10)

 

 

(o(./go—a -2JM)+a +
.lrp—a

Third, when reserves are too low to cover the current account deficit or

whereZ'=-17+R—F—Tand(DE F+T—R.

R < F (rer)+T(trade), then, reserves will not be sufficient to close an external gap

regardless of the fluctuations in the demand for money, AMd. In this scenario, the

government will deplete its reserves and then close the external gap through a

204

combination of depreciation and manipulation of the unemployment rate as expressed in
equations (9) or (10).
The symmetric rational expectation’s equilibria are found by combining the

investor’s withdrawal policy (2) and the government strategies from (7) to (10). There are
three cases. First, if $'*( — M) S s: , then there is a unique symmetric equilibrium where
AM" =0 and 3:370).
Second, if 3 e [55 (O). SEX—111)], then there are two symmetric equilibria.
AMd = 0 and E: {(0)
AM" =—M and §=s’(—M).
Third, if E < 3*(0 ) , then there is a unique symmetric equilibrium,
AMd = —M and i =.s"*(—M).

In the first case, an attack never occurs since either reserves are high or

fundamentals are strong.35 When reserves are high, the government will respond to any

AM d by spending its reserves and setting 3 = 0. The second and third cases occur when
reserves are low and fundamentals are weak. In the second case, there are multiple

equilibria. In the crisis equilibrium, investors believe that the devaluation will be greater
than 3 and consequently withdraw their deposits. As a result of the withdrawal, the
devaluation is indeed greater than 3 . In the non-crisis equilibrium, investors believe do

not withdraw their deposits and depreciation is not greater than 3 . In the third case, the

205

 

fundamentals are so weak that the government will have to depreciate more than sT
regardless of investor’s expectations.

In sum, an individual money manager will attack a currency only if it is
anticipated that the country will respond with a sizeable depreciation. A sizable
depreciation is more likely to occur in countries with low international reserves, a severe

current account deficit and a rapid lending boom.

3. Empirical analysis

The theoretical model in the previous section suggests that the countries most
vulnerable to a speculative attack have a severe real appreciation, a strong trade linkage
with the country initially attacked by currency speculation, 8 rapid lending boom and low
international reserves. Investors concentrate their speculative attacks in countries more
likely to respond with an excessive depreciation.

This section shows that the European, Mexican and Asian currency crises did not
spread randomly across industrial and emerging markets during the years 1993, 1995 and
1997. In addition, the extent of role of other determinants of currency crisis is examined.
These include high government consumption, slowdown in real GDP growth rate,

excessive capital inflows and increasing foreign liabilities.

3.1 Defining variables in the empirical model

 

35 There are neither many bad loans nor a sever current account deficit due to a real appreciation or a
strong trade linkage with an initially attacked country.

206

Currency crisis index

 

The first issue confronted in the analysis is how to measure devaluation pressures
on the foreign exchange market. Eichengreen, Rose, and Wyplosz (1996), Sachs, Tomell
and Velasco (1996), Frankel and Rose (1996) and Kaminsky and Reinhart (1999) used
analogous crisis indices for the measurement of pressures on the market. Their indices are
weighted averages of the percentage of depreciation in the nominal exchange rate with
respect to the US. dollar and the percentage decrease in reserves. The rationale for these
indices is as follows. If capital inflows reverse, then the government can depreciate the
exchange rate. Alternatively, it can defend the currency by spending its reserves or
increasing interest rates. Since the authors assert that there is no reliable and comparable
cross-country interest rate data, their indices are constructed without an interest rate.

The extent of currency crisis here is measured with a crisis index (denoted
“Crisis”),

Crisis” 2 WAS %Asi, - WAR%AR,-, (11)
where %As,, and %AR,-, are percentage change of the nominal exchange rate and the
international reserves respectively. The weights used to derive the crisis index are

constructed as

WA. =—‘+— and WAR =1—wAS,

where 0'45 and GAR are the standard deviation of the change rates of nominal exchange

rate and international reserve. The initial point for the percentage change is the month

207

 

before the onset of the crisis (August 1992, November 1994 and June 1997). Then. the
terminal month is varied over a period of six months starting in October 1992, January
1993, and August 1997. This index is similar to indices used in the extensive prior
literature. However, the standard deviation of the change rate of the variable instead of
the level to preclude an excessively small weight being given to the international reserves
due to its unusually high volatility compared to that of a nominal exchange rate. In
addition, unlike Kaminsky and Reinhart (1999) who assign the weight of the nominal
exchange rate to be one, Wm, which is less than or equal to one, is used to weight the
change of nominal exchange rate to prevent an excessively large weight being given to
the nominal exchange rate. The values of Crisis are listed in Table 42. A higher value of
Crisis means either higher level of devaluation or a greater fall in reserves. With the
exception of Brazil in 1993, all of the countries that experienced a currency crisis in

1993, 1995 or 1997 have higher a crisis index than other countries.

Lending boom

 

A broad cross-country set of comparable bank balance sheets does not exist.
Hence the weakness of the banking sector cannot be assessed directly by comparing the
ratio of non-performing loans to total assets. Instead, an indirect measure of financial
system vulnerability is used: the magnitude of the increase in bank lending as measured
by the percentage change in the ratio of claims on the private sector by deposit money
banks and monetary authorities (line 32d) to GDP (line 99b) during the periods 1988-92,

1990-94 or 1992-96. This variable is LB and its values are listed in Table 43. The first

208

column of the table shows the LES for European currency crisis. Unlike as was the case
with the Mexican and Asian currency crisis in the second and third columns, the LBS
values in European countries in the first column are not in the high rankings. This may
imply that lending booms did not play so crucial of a role in the currency crises of
industrial countries relative to the role it played in the currency crises of the developing

countries.

Real exchange rate depreciation

 

A real exchange rate depreciation index is constructed as a weighted average of
the bilateral real exchange rates of a given country with respect to the US dollar, the Yen
and the Mark. The weights add up to one and are proportional to the shares of bilateral
trade in the given country with the US, Japan and Germany, respectively. The extent of
real exchange rate misalignment is then measured with the percentage change in this
index over the four years prior to the onset of the crisis. This variable is RER. A positive
value of RER signifies that the real exchange rate depreciated relative to the base period,
while a negative value indicates appreciation. Table 44 offers the values of RER for the
currency crises in 19903. The visual inspection of the first column in the table also shows
that the negative relationship between the currency crisis and the RER is weaker in the
European currency crises, although the rankings of Italy and Spain in 1992 are higher

than others in 1994 and 1996.

Eserves adequacy

 

 

The government’s liquidity is proxied by the ratio of M2 to reserves in the month
preceding the onset of the crisis (August 1992, November 1994 or June 1997). The ratio
captures the extent to which the liabilities of the banking system are backed by
international reserves. If the central bank is unwilling to allow the exchange rate to
depreciate, then it must be prepared to cover all the liabilities of the banking system with
reserves. Hence, it is MZ, and not simply the monetary base, that is the relevant proxy for
the central bank’s contingent liabilities. The values of the ratio of M2 to reserves for both
industrial and developing countries are listed in Table 45 and 46, respectively. The
industrial countries’ reserve adequacies in Table 45 are excessively lower than the
developing countries’ in Table 46. This implies that a country whose financial market is
well developed and stable is allowed to maintain a larger monetary base than emerging
market countries. who are more likely to be exposed to unexpected speculative attack.
However, the reserve adequacy of Italy in 1992, Mexico in 1994 and Thailand in 1997,
when they were initially attacked, is lower than the reserve adequacies of other countries

in their group of sample.

Contagion effects

 

Contagion effects are the most recent contribution of second-generation models.
There are several channels through which they may be transmitted across countries. First,
contagion can be explained by common external factors, 30 called "Monsoon effect". For
example, a rise in US. interest rates in 1994 or the devaluation of the yen in 1995 could
be common external factors. Second, it is also caused by trade linkage or third market

competition-related spillovers. The trade linkages between countries with geographic

210

 

 

 

proximity help to explain spillover effects. In addition, an indirect trade linkage due to
third market competition may be instrumental in encouraging repeated rounds of
competitive devaluation. The third factor of the contagion effect is market sentiment. The
herd mentality of investors also may contribute to the contagion effect.

For the common external factors of contagion, this section considers the
contagion effect from the devaluation of the major currencies as reflected in the change of
the real exchange rate depreciation index, RER and a decline in the differential between
domestic and US. interest rates, Itrdus. To control for the trade linkage or third market
competition-related spillovers, a trade linkage index (denoted “T rade”) between the
initially attacked and home country is constructed following the same method used in

Glick and Rose (1998). Trade is a weighted average index between the third market

competition index,

x . + x-. x ' - X .
Indirect = z{[-9A——'l’] * [1 - lL—m—ll}
k 9‘0. + xi. xik + ka
and direct trade linkage index,
Direct = 1 ———l '0 0"
x20 '1” x0,-

where x,k denotes aggregate bilateral exports from country i to k (k ¢ 1', 0) ;x,-0 denotes

aggregate bilateral export from the home country i to the initially attacked country 0; xi.

denotes aggregate bilateral exports form country i. Indirect is the weighted average of the
importance of exports to country k for countries 0 and i. The relative importance of
country k is strongest when it is an export market of equal importance to both countries 0

and i. The weights are proportional to the importance of country k in the aggregate trade

21]

'
.I . r
A

LUZ-i -‘~ .1;

 

I Polar: -

of countries 0 and i. Direct measures the equality of bilateral exports between countries 0

and i. A measure of total trade. Trade, is the weighted sum of Indirect and Direct. The
weight on the latter term is (xm + x(,,.)/(x(,' + x,). Table 47 lists the values and rankings

of the countries. For the European and Asian currency crises, the trade linkages between
the initially attacked countries, Italy and Thailand, and home countries are higher than
those of other countries. However, for the Mexican currency crisis, the trade linkages
between the first victim, Mexico, and home countries are not very high. Based on the
lower levels of competition in the third market between the Latin American countries.
Indiect is low while Direct is high. However, the weights for direct trade linkages are low
since there is very little direct trade in their whole trade volumes. Finally to control for a
market sentiment, Pure, a dummy variable equal to one when one of the regional

countries is attacked by speculative agents, is included in the regression.

Additional determinants of currency crisis

 

While the first generation models of currency crisis proved that high government
consumption levels (denoted GOVC) was a crucial factor for the onset of currency crisis
before 19903, additional factors such as a slowdown in real GDP growth rate (denote
GDP), excess capital inflows (denoted CAPI) and increasing foreign liabilities (denoted
by F ORLB) were identified as important determinants of currency crises in the19903 by
the second generation models. Therefore, it is necessary for us to analyze whether these
variables help to explain the cross-country variation in the crisis indices after controlling

for a lending boom, a real appreciation, and 8 reserves adequacy ratio. Each variable is

IQ
Ix)

 

measured as the average ratio to GDP (GOVC, CAP], and F ORLB) or the change rate of
real GDP growth rate (GDP) over the four years to the onset of the crisis (1988-92, 1990-

94, and 1992-96).

3.2 Data set

A three-period panel data is used from three different episodes of important and
widespread currency crisis in 19903. The three episodes are: 1) the European currency
crisis of 1992-93; 2) the Mexican currency crisis of 1994-95; and 3) the Asian currency
crisis of 1997-98. All the variables except bilateral trade are taken from the CD-ROM
version of the lntemational Monetary Fund’s International Financial Statistics (IFS). For
the bilateral trade, the IMF’s Direction of Trade is used. The data set includes data from
29 countries”. The countries are grouped as European, Asian and Latin American
countries or industrial and developing countries. The sample was chosen based on the
existence of free convertibility and financial markets. The sample was also selected to
ensure that both industrial and developing countries were included.

For the currency crisis index, monthly data of nominal exchange rate (line rf) and
international reserves (line 11.d) were collected from January 1985 through January 1998.
To estimate the impact of lending booms, it was necessary to collect annual data of
claims on the private sector by deposit money banks and monetary authorities (line 32d)

and nominal GDP (line 99b) for the years 1988 through 1996. For the real exchange rate,

 

36 The countries are U.S.A., UK, France, Italy, the Netherlands, Norway, Sweden, Canada, Japan,
Finland, Spain, Australia, Germany, Argentina, Brazil, Chile, Colombia, Mexico, Jordan, Sri Lanka, India,
Indonesia, Korea, Malaysia, Pakistan, the Philippines, Thailand, Turkey, and Venezuela.

213

 

 

 

the CPI annual data (line 64) over the 1988-96 period as well as nominal exchange rate
were collected. To measure reserve adequacy, monthly data of money (line 34) and quasi-
money (line 35) as well as international reserves for August 1992, November 1994 or
June 1997 were collected. To capture contagion effect, annual data of money market
interest rates (line 60b) or discount rates (line 60) were collected for the years 1988
through 1996. In addition, annual bilateral trade data is used to estimate trade linkage
index over the same period as the interest rate. For additional determinants of the
currency crises, annual data of government consumption (line 911), capital accounts (line
78bc), financial accounts (line 78bj), net errors and omissions (line 78 ca), real GDP (line

99 br), and foreign liabilities (line 26c) were collected for the years 1988 through 1996.

3.3 Regression analysis

As discussed in the theoretical framework, a currency crisis occurs when the
investors launch an attack because of the weak fundamentals of the country and its
relatively low reserves level. The targeted countries for a speculative attack are those
countries that are most likely to respond with an excessive depreciation. Following Sachs,
Tomell and Velasco (1996), an empirical implementation of these ideas is made by
classifying observations into four groups: high and low reserves cases, and strong and
weak fundamentals cases. However, since the classification system includes both
industrial and developing countries, the country-years with high reserves and strong
fundamentals are different from theirs. A country is defined to have a high level of
international reserves if the ratio of its M2 to its reserves is in the lowest quartile for

either industrial or developing countries. The dummy variable for high reserves, dhr, is

214

 

7'1

 

up up,

 

equal to one for countries whose money-to-reserves ratio is in the bottom quartile for its
group. By contrast, a country has strong fundamentals if its real depreciation is in the
highest quartile of sample and its lending boom is in the lowest quartile of sample. The
dummy variable for strong fundamentals, dsf, is equal to one for countries that have

strong fundamentals.

3.3.1 Country effects

In the sample, there are three observations per country for the European, Mexican
and Asian currency crisis. As such, we need to check the existence of country effects to
determine the correct specification of benchmark regression model. Based on this, the
following regression is estimated using pooled OLS. fixed effects and random effects

models. The specification consists of
Crisis,, 2 I30 + [ilLB,-, + /)’2 RER,, + 63dhr - L8,, + [34dhr ' RER,, + ,85de - LB,,
+/)’6dsf-RER,-, +v,, (12)
where i indexes the country and t indexes time; v,, = a, +u,, and a,- is an unobserved

country effect.

To test the null hypothesis of no country effects against the alternative of fixed

effects, the unobserved effect a, is replaced by 28 terms of the form a,- * d, in equation
(12), where d ,- is a dummy that equals to one if the observation corresponds to country i.

Then, the model is estimated using the pooled OLS model and an F test is performed.

Under the null, all coefficients of [1’0 and a, ’s are equal. The F statistic is

 

 

(0.3607 — 0.1225) / 28 _

F 28.52 =
[ ] (l — 0.3607)/52

0.6920.

 

Since the 5% critical value is 1.69, we failed to reject the null hypothesis of no fixed
effects.

Next, to test the null hypothesis of no country effects against the alternative of
random effects, a Breusch Pagan test is performed after the model in equation (12) is first
estimated using a random effects model. The null hypothesis means that the variance of

a,- is zero. The test statistic for a Breusch Pagan test is 1.29 and we failed to reject the

null hypothesis at the 5% critical level.

The model in equation ( 12) is then estimated using the pooled OLS and fixed
effects models.37 As shown in Table 48, the point estimates of LB and RER have the same
signs regardless of the specification of the models. However, while the estimated LB
coefficient is significantly different from zero at 5% level in the pooled OLS model, it is
not in the fixed effects model.

The results of the tests indicate that a pooled OLS model is an appropriate
specification. Based on this, the pooled OLS model will be the benchmark regression in

the remainder of this chapter.

3.3.2 Benchmark regression
In the benchmark regression, 87 observations for the 1992, 1994, and 1997 crises

are stacked and the following regression using ordinary least squares is estimated.

. .f 7.]

 

Crisis,-, = flu + B,LB,, + [12 RER,, + 8,211” - L3,, + 5,611” . RER,, + 85w - L8,,
+ B6dsf - RER,-, + u,, (13)
The effects of a lending boom and real appreciation with weak fundamentals and low
reserves are reflected in B, and B2 , respectively. The signs of B, and B2 are expected to
be positive and negative, respectively. In addition, B, + B3 and B2 + B, indicate the
effects of a lending boom and real appreciation with high reserves. The effects of a
lending boom and real appreciation with strong fundamentals are likewise indicated by
B, + B5 and B2 + B6 , respectively. This study’s expectation is that B, + B3 =0, B, + B,
=0, B, +B5 =0and B2+B6 =0.
The currency crisis index used here is obtained with data from five months after
the eruption of the crisis. For the European crisis, the time period is from September 1992
through January 1993. For the Mexican and Asian crises, it is from November 1994

through April 1995 and June 1997 through November 1997, respectively. The estimated

regression is

Crisis,, = 3.90 + 0.08LB,, — 0.18RER,, — 0.03dhr - L8,, + 0.01dhr - RER,-, - 024de - LB,-,

(1.85) (0.04) (0.13) (0.11) (0.27) (0.21)
—— 0.21de . RER,,
(0.24)
R2=O.12, R2=O.O6, N=87 (14)

HI

 

 

37 I do not present the results of estimation using the random effects model since they are as same as the
pooled OLS estimation.

217

 

 

Heteroskedasticity robust standard errors are presented in parentheses. The point
estimates in equation (14) indicate that the estimated coefficients of LB and RER have
positive and negative signs as expected. A one unit increase in the LB or a one unit
decrease in the RER for a country-year with low reserves and weak fundamentals leads to
0.08 or 0.18 unit increase in the crisis index, respectively. In addition, the estimated
coefficients of LB and RER are significantly different from zero at the 5% and 10% level,
respectively. This justifies the inclusion of both variables in the equation (14). The fourth
column of Table 49 presents the results of estimation with the sample only including
industrial countries. In this sample, the signs of LB and RER are negative and both
variables are not significant at the10% level. In the strong and stable financial systems of
industrial countries, a lending boom does not contribute to the variation of the crisis
index. By contrast, the estimation results from the sample of developing countries in the
fifth column shows that the signs of LB and RER are positive and negative and only LB is
significant at the 5% level. Therefore, the existence of a lending boom has a stronger
impact on the crisis index in developing countries than the industrial countries. Since the
lending boom before the currency crisis is a common experience in the developing
countries,38 this result is not puzzling. As an additional check for this, the terms LB *ddev
and RER *ddev were added to equation (13) where ddev takes the value of one for
observations that correspond to the years, 1994 and 1997. The sixth column of Table 49
shows that the estimated coefficient of LB *ddev is significant at the 5% level whereas the

RER *ddev’s coefficient is not significantly different from zero.

 

38 See Chapter 11 for details.

218

C. nA"-.
'.

C‘AIsJVG—‘ji '2! 3'."
. .
.3

 

Since the lending boom’s effect on the currency crisis is different depending on
the crisis episodes. we need to check whether the same model that explains the crises of
industrial countries in 1992-93 also explains the cross-country variation in the 1994-95
and 1997-98 crises. To test the hypothesis that the coefficients in the equation (13) are the

same in both periods, I perform a Chow test. The test statistic is

(17606 — 5946 - 6599) / 7 _

F 7, 73 =
l l (5946 + 6599) / 73

4.21

 

Since the critical value at the 5% level is 2.14, we can reject the null hypothesis that the
coefficients are the same for the crises of industrial and developing countries.

Following a confirmation of the initial theoretical implications, F-tests indicate
that the hypothesis B, + B, = 0 and B2 + B, = 0 failed to be rejected in the third column
of Table 49. Therefore, in countries with higher levels of reserves, neither LB nor RER
affect the severity of a crisis. In addition, for the countries with strong fundamentals,
B, + B5 = 0 and B2 + B6 = 0 cannot be rejected. Hence, neither changes in LB or RER

affect the severity of a crisis in the countries with strong fundamentals.

3.3.3 Contagion effects
To find contagion effects at the onset of the currency crisis, a regression with the
trade linkage index variable, Trade, interest rate differential, Itrdus, and a pure contagion

variable, Pure that reflects market sentiment. The estimated regression is

219

Tr-VH-
n
I

 

32'". 2A 0 m _ - .

1.1-.ij
~ I

 

Crisis, = —6.58 + 0.08LB,, — 0.0411511, + 0.08dhr . deg . L8,, — 0.13dhr - deg - 111511,,
(3.04) (0.03) (0.12) (0.10) (0.22)

— 0.28dsf - deg - LB,, — 0.01dsf-dcg . RER,, + 26.6STrade,, + 6.48Pure,-,

(0.29) (0.29) (9.30) (3.90)
— 0.4OItrdus,,
(0.47)
R2=O.35, [73:028. N=87 (15)

A country is defined as not exposed to contagion effects if its trade linkage index is in the
lowest quartile of the sample or its pure contagion index is zero. In addition, if the
interest rate differential is in the highest quartile of the sample, a country is not under the
influence of the contagion effects. Thus, the dummy variable for the contagion effect,
deg, is equal to one for countries not exposed to the contagion effects. Therefore, the
point estimates of the coefficients of LB and RER in equation (15) reflect the effect of a
unit increase of LB and RER on the crisis index under all conditions for countries who do
not have high reserves, strong fundamental, and have been exposed to contagion effects.
In equation (15), the estimated coefficients of Trade and Pure have positive signs and
Itrdus coefficient has a negative sign as expected. The scale of Trade and Pure goes from
zero to one. In this framework, 8 0.1 unit increase in Trade and Pure leads to 2.67 and
0.65 unit increases in the crisis index, respectively. A one unit decrease in Itrdus also
increases the crisis index by 0.40 units. However, while the estimated coefficients for
Trade and Pure are significantly different fiom zero at the 5% level, Itdrus coefficient is
not significantly different from zero at the 5% level. This indicates that the contagion

effects from the trade linkage and market sentiment played a crucial role in the currency

220

l—ahflmn-I“ -44a:-.i-'v 1. ill 21“,
..

 

 

crises after 1990. The third column of the Table 50 indicates that the estimated
coefficients of LB and RER are still significant at the 5% and 10% level even with the
dummy deg. The fifth column of the Table 50 indicates that the estimated coefficients of
RER, Trade, Pure and Itrdus has the expected signs but all the coefficients except Itrdus
and Constant are no longer significantly different from zero. Thus, with the exception of
Itrdus, the relationships between the crisis index and explanatory variables in the
currency crisis of industrial countries are not reliable. By contrast, the sixth column,
where the estimated coefficients of Trade and Pure are significant at the 5% level, shows
that the contagion effects played a key role in the onset of the Mexican and Asian

currency crises.

3.3.3 Additional determinants of currency crisis

In this subsection, it is analyzed whether higher government consumption, a
slowdown in real GDP growth rate, excess capital inflows, and increasing foreign
liabilities help to explain the cross-country variation in the crisis indices after controlling
for a lending boom, real appreciation, reserves adequacy and contagion effects. Table 51
presents the estimated coefficient for each variable.

The third column of Table 51 presents the estimated coefficients of government
consumption, denoted by GOVC. The regression results indicate that government

consumption does not significantly effect the crisis index. This coincides with the

 

t mi 5. Ti.

25--27.- m1. -’ .' 1...;

literature”). The literatures has found that government consumption had a weaker effect
on the currency crises in 19903 relative to the crises in 19803. The insignificant
coefficient on GO VC may be explained by the changing nature of crises.

Similarly, a slowdown of real GDP growth rate is assumed to increase the
policymaker’s incentive to switch to a more expansionary policy, which can be achieved
through a nominal devaluation of the currency. Therefore, the real GDP growth rate,
denoted by GDP, should capture the escape-clause interpretation developed in various
second-generation models of currency crisis."0 The fourth column of Table 51 shows that
the estimated coefficient of GDP has an expected negative sign but it is not significantly
different from zero at the 5% level. Hence, a decline in the real GDP growth rate does not
appear to contribute to the currency crises. Table 2, in Chapter II, shows how Asian
countries continued to have relatively high GDP growth rate in the 19903 before the 1997
crisis although the growth rate slowed slightly prior to the crisis

For the capital inflows, denoted by CAP], the estimated coefficients are reported
in the fifth column of Table 5]. Excessive capital inflows are regarded as a main factor
for the onset of currency crisis. This is because the short time span of excessive inflows
prevents them being efficiently channeled to productive projects and eventually lead to a
shortage of returns to repay investors. The fifih column of Table 51 shows that the
estimated coefficient of CAP] has positive sign as we expect. It is also significantly

different from zero at the 5% level. Hence, even after controlling for all the other

 

39 Sachs, Tomell and Velasco (1996), Pazarbasioglu and Otker (1996), and Corsetti, Giancalro, Pesenti,
and Roubini (1998)

4° For example, Obstfeld (1994, 1996).

222

I.” “I

‘ "-1; “1‘..- K‘k'h‘x‘j‘. .Ll

 

contributors to a crisis. a one unit increase in the capital inflow index leads to 0.56 unit
increase in the crisis index.

The sixth column of Table 51 presents the estimated coefficients of foreign
liabilities, denoted by F ORLB. We expect the bank’s foreign liabilities as a ratio of GDP
to represent the extent to which the banking system is exposed to international capital
flow. However, the point estimate sign goes against expectations but also is not
significant at the 5% level. Hence, the foreign liabilities do not contribute significantly to

the cross-country variation of the crisis index.

3.4 Robustness

To analyze whether the results are robust over the periods in which the crisis
index is measured, equation (13) is estimated again using six different crises indices. For
all indices, the starting point is the month preceding the onset of the crisis (i.e. August
1992 for the European crisis. November 1994 for the Mexican crisis, and June 1997 for
the Asian currency crisis). Then, the terminal month is varied over a period of six months
starting with October 1992, January 1995, and August 1997. As Table 52 shows, in
columns three, four, six and eight, the point estimates of LB are similar to the benchmark
regression point estimates (the fourth column). Moreover, they are significantly different
from zero at both the 5% or 10% level with the exception of the coefficient reported in
the eighth column. The point estimates for RER show some variation across
specifications but is always significantly different from zero at the 5 or 10% level. Other

variables show the same tendency in their values and significances.

In the benchmark regression. a country-year is classified as having high reserves if
its ratio of M2 to reserves is in the lowest quartile for industrial or developing countries at
the onset of the crisis. The threshold values are 8.0 (10 country-years) and 2.8 (12
country-years), respectively. A country-year is also classified as having strong
fundamentals if lending boom index is in the lowest quartile of sample and real
appreciation index is in the highest quartile of the sample where the threshold value for
the lending boom is 8.0 and for the real appreciation is —9.0 (9 country-years). The fourth
and fifth columns of Table 53 show the estimates for different thresholds concerning the
high reserves dummy, while keeping the strong fundamentals dummy unchanged. The
thresholds are 6.7 (8 country-years) and 2.0 (8 country-years) for the fourth column and
9.5 (11 country-years) and 3.2 (14 country-years) for the fifth column. Column 6 and 7
also indicate the estimates based on different thresholds for the strong fundamentals
dummy, while keeping the high reserves dummy stayed at the same level. The thresholds
of RER and LB are 12.6 and -l 0.9 (6 country-years) for the sixth column and 3.8 and -4.0
(15 country-years) for the seventh. The last column shows the estimates after all the
thresholds for the dummies are changed. The thresholds of the ratio of M2 to reserves,
RER and LB are 6.7 (8 country-years) and 2.0 (8 country-years), 12.6 and —10.9 (6
country-years) respectively. As seen in the columns of Table 53, the coefficient’s values
and significance levels do not substantially differ from each other based on the threshold

values selected.

3.5 Predicting the Asian currency crisis

3.5.1 Prediction without the contagion effects

224

Based on the benchmark regression’s estimates, we may predict the currency
crisis indices in the Asian crisis, 1997. To this end, the following regression can be
estimated with data from the 1994 crisis:

C risis,, 2 B0 + B,LB,-, + BZRER” + B3CAPI,-, + B4dhr - LB,-, + Bsdhr - RER,, + B6dsf - L8,,

+ B7dsf- RER,, + u,-, (16)
This equation includes the capital inflow index denoted by CAP]. Its inclusion is based on
its ability to explain the cross-country variation of the currency index. Then, an out-of-
sample predicted crisis index is constructed by substituting into equation (16) the
estimated coefficients of a regression that are significantly different from zero and the
values of the explanatory variables that correspond to the Asian currency crisis. The
fourth column of Table 54 presents the resulting predicted crises indices without the
contagion effects according to descending order.

First, the predicted crisis indices is a dotted line in Figure 63. As shown in Figure
63, the predicted crisis indices follows closely the actual crisis indices well. Second, each
5 countries is then grouped in descending order of predicted crisis indices. This chapter
then checks how many rankings of countries in each group of predicted crisis indices
coincide with the rankings of the countries in the matched group ranked by the actual
indices as reported in column 2 of Table 54. A total of 8 countries have rankings that
match in column 2 and 4. Third, the actual crisis indices of 1997 are regressed on the

predicted out-of-sample crisis indices. The regression result is

225

'F2nmfio-l. M. -na..‘L “at -‘w
~ i

 

Actual 97erisis, = 6.55 + 0.42 -[Pr edicted 97erisis,]
(2.17) (0.12)

123:0.26, 172:0.24, N=29 (17)
The regression coefficient is 0.42, and significantly different from zero at the 5% level.

Fourth. a Root Mean Square Errors, denoted by RMSE', is estimated;

 

N
RMSEf = JLEXAetuaI 97crisis, — Predicted 97erisis,-)2 , N = 29
N

i=1
The estimated RMSEf is 16.02. This value will be compared with RMSEC, the Root

Mean Square Errors when the predicted crisis indices includes the impact of contagion

effects as shown in the following subsection.

3.5.2 Prediction with the contagion effects

The previous section’s predicted currency indices for the Asian currency crisis
was obtained with the benchmark regression. The benchmark regression proved to be
dependable in predicting the out of sample movements of the Asian currency crisis
indices. However, as shown in Table 54, the predicted crisis indices of the countries,
Indonesia and Korea, which were severely attacked during the Asian crisis, were not
predicted well. To improve our ability to predict the crisis indices, the following

regression is estimated
Crisis,-, 2 B0 + B,LB,-, + B2 RER,, + B3CAPI,, + B4dhr -dcg - LB,-, + Bsdhr ~deg . RER,,

+ B6dsf-deg - L8,, + B7dsf-deg - RER,, + B8Trade,, + BgPure,, + B,01trdus,-, + u,,(18)

226

. '..; “1.01
I

sP-auml-M‘O «.21
i
.l

 

 

where the contagion effects are captured by deg, Trade, Pure and Itrdus. After the
Philippines and Thailand were predicted as the countries that are most likely to be
attacked by the speculators by the benchmark regression without the contagion effects,
the trade linkage index, Trade, and market sentiment variable, Pure, for the Asian crisis
are reconstructed based upon the prediction of the benchmark regression."I Then, an out-
of-sample predicted crisis index with contagion effects is constructed with the same
procedure used in the previous subsection.

The sixth column of Table 54 presents the resulting predicted crises indices with
the contagion effects according to descending order. As shown in Figure 63, the new
predicted crisis indices closely follows the actual crisis indices. In addition, the new
predicted crisis indices appears to fit the actual crisis indices better than the prediction
made from the benchmark regression in Figure 63. Furthermore, a total of 11 as opposed
to 8 countries have the same ranking based on the new predicted series as when they are
ranked with the actual crisis indices.

Next, the actual crisis indices of 1997 are regressed on the new predicted out-of-

sample crisis indices. The regression result is

Actual 97crisis, = 4.23 + 0.51 -[Pr edieted 97crisisf]
(1.41) (0.13)

R2=0.56, 172:0.53, N=29 (19)
The point estimate of the predicted crisis indices with contagion effects is now 0.51

instead of 0.42. It is also significantly different from zero at the 5% level. A larger

 

4' The new trade linkage is a weighted sum of the trade linkages with the Philippines and Thailand which

227

’_~__ 7m]
V

77-: '. _I_' i

. nap-a 9.421» :..in'&

 

 

portion of the variation in the actual crisis indices is explained by the new predicted crisis
indices. This is shown by how the adjusted-R2 increases from 0.24 to 0.53 and the
estimated RMSE goes from 16.02 to 12.08. This indicates that the prediction based upon

contagion effects improves our ability to predict an upcoming currency crisis.

 

are weighted by the predicted crisis indices of the Philippines and Thailand, respectively.

228

 

 

 

 

1m

Table 42. Currency crisis index:t

 

 

Country 1993 Country 1 995 Country 1997
Brazil 69.7 Mexico 70.9 Thailand 47.9
Finland 29.7 Brazil 19.1 Indonesia 33 .5
Spain 25 .2 Argentina 16.7 Malaysia 32.0
Sweden 22.8 Philippines 7.2 Philippines 29.7
UK. 21.5 Italy 5.8 Korea 16.4
Italy 18.7 Spain 5 .7 Colombia 16.3
Norway 1 8. 1 Venezuela 4.3 Turkey 12.4
Venezuela 14.7 Colombia 1 .8 Pakistan 8.4
Turkey 1 1 .4 Indonesia 1.1 Brazil 7.3
Australia 1 l .1 Pakistan 0.7 Australia 4.6
U.S.A. 9.7 Sri Lanka 0.1 Japan 4.4
France 9.4 Canada -0. 1 India 3 .8
Sri Lanka 8.9 Malaysia -0.6 Finland 3.8
Canada 8.8 India -O.9 Canada 2.9
Jordan 3 .9 Australia -1 .2 Netherlands I .9
Colombia 3 .2 Jordan -1 .4 Germany 1.3
Pakistan 3 .1 Thailand -2.4 USA 1.2
India 2.9 UK. -2.7 UK. 0.8
Malaysia 2.1 Sweden -3.3 Sri Lanka 0.3
Indonesia 1 .9 Korea -4.5 Sweden 0.3
Germany 1.8 France -7.3 Mexico 0.1
Thailand 0.0 Germany -7.4 Chile -0.5
Chile -0.3 Chile -7.5 Argentina -0.7
Japan -O.5 Finland -7.8 France -1 .5
Korea -0.7 Norway -9.5 Jordan -2.3
Philippines -1 .6 Netherlands -9.5 Spain -2.9
Netherlands -2 .3 Turkey - 1 2 .4 Norway -4.5
Mexico -2.4 Japan ~18.7 Italy -7.7
Argentina - I 2.3 U.S.A -23 .4 Venezuela -1 1.0

 

* The currency crisis index (Crisis) is a weighted average of the percentage depreciation of nominal
exchange rate with respect to the US. dollar and the percentage decrease in reserves.

229

Table 43. Lending boom.

 

 

Country 1988-1992 Country 1990-1994 Country 1992-1996
Mexico 231 .1 Mexico 124.6 Philippine 13 7.2
Turkey 181 .0 Philippines 51 .0 Colombia 40.3
Indonesia 67.7 Thailand 40.9 Thailand 38.4
Australia 44.4 Brazil 30.8 Turkey 38.3
Thailand 41.4 Colombia 25 .5 Malaysia 25.6
Philippines 28.1 Sri Lanka 24.7 Chile 23.3
Malaysia 21 .6 Canada 20.7 Netherlands 22.8
Canada 20.0 Argentina 1 7. 1 Jordan 22.3
Korea 1 7.7 Netherlands 10.7 Indonesia 21 .8
Finland 17.0 Indonesia 10.6 Canada 20.2
UK. 14.9 Chile 8.0 Argentina 16.7
France 10.0 Italy 7.1 Germany 15.7
Italy 5 .8 Korea 6.4 Korea 14.3
Netherlands 4.5 Australia 6.1 Sri Lanka 12.1
Germany 4.3 Malaysia 4.9 Australia 7.6
Japan 4.3 Germany 3 .9 UK. 3 .6
Sweden 4.2 Jordan 1.9 U.S.A 3.1
Sri Lanka 1.3 Spain -0.9 Norway 0.8
Spain 1 .1 Pakistan -3.l Pakistan 0.5
Colombia -1 .6 Japan -4.2 Spain -3.9
India -2.9 UK. -4.8 Japan -0.7
Norway -10.2 India -5 .9 India 40
Jordan -10.8 Turkey -8.6 Italy -9.5
Pakistan -1 1.9 France -10.6 France -15.5
Chile -20.0 U.S.A -1 1.3 Finland -29.2
Brazil -13.1 Finland - 1 3 .3 Sweden -32.1
U.S.A. -16.0 Norway -13.5 Mexico -39.6
Argentina -23.1 Sweden -30.1 Brazil -51.6
Venezuela -3 7.9 Venezuela -44.5 Venezuela -56.7

 

* Lending boom (LB) is the percentage change in the ratio of claims on the private sector by deposit money
banks and monetary authorities (line 32d) to GDP (line 99b).

230

Table 44. Real depreciation’

 

 

Country 1988-1992 Country 1990-1994 Country 1992-1996
Argentina -50.0 Argentina -41.1 Colombia -33 .5
Mexico -27.6 Brazil -27.4 Brazil -24.1
Turkey -25.2 Japan -27.0 Philippines -19.6
Spain -21.2 Colombia ~25.0 Chile -1 6.1
Chile -18.2 Philippines - l 9.7 Thailand -1 1.8
Sweden - l 7.3 Mexico -19.3 Indonesia -1 1 .7
Brazil -17.3 Chile -12.4 Sri Lanka -10.8
Philippines -16.6 Venezuela -9.5 Japan -10.1
Italy -14.2 Sri Lanka -8.8 Malaysia -9.3
Germany -1 1 .2 Malaysia -8.5 Australia -9.2
UK. -10.2 Thailand -7.9 Korea -9.2
France -9.7 Indonesia -5.8 Argentina -8.5
Korea -8 . 5 Germany -2.2 Venezuela -8. 1
Netherlands -7.4 Jordan -1 .9 Germany -5 .3
Sri Lanka -7.3 Korea -1 .6 Netherlands -5.0
Thailand —6.6 USA. -0.9 Jordan -3 .9
Norway -5.3 Netherlands 0.6 U.S.A -2.5
Canada -4.9 Pakistan 4.3 France -2.4
Malaysia -3.6 France 5.3 Pakistan 0.8
USA -2.6 Australia 1 1.0 Norway 3.8
Finland 1.4 UK. 14.6 India 4.2
Australia 1 .4 Norway 15.3 Finland 5.3
Japan 1 .6 Canada 18.9 Turkey 9.6
Indonesia 2.4 Spain 20.5 UK. 10.7
Colombia 7.8 Sweden 22.2 Sweden 12.7
Pakistan 10.7 Italy 25 .2 Spain 12.8
Venezuela 1 7.9 Turkey 32.0 Italy 14.3
Jordan 22 .3 India 34.5 Canada 14.4
India 44.5 Finland 39.2 Mexico 22.8

 

* Real depreciation of the exchange rate (RER) is the percentage change in the real exchange rate index
over the four years prior to the onset of the crisis

23]

Table 45. Reserve adequacy (Industrial countries)m

 

 

Country Aug. 1993 Country Nov. 1994 Country June 1997
U.S.A 57.9 U.S.A. 63.6 U.S.A 81.5
Japan 54.0 Japan 42.1 UK. 40.6
Italy 33.7 France 34.7 France 32.4
France 26.3 UK. 24.9 Japan 22.3
UK. 24.4 Canada 24.6 Canada 17.7
Canada 19.9 Italy 23 .0 Germany 16.6
German 1 7.9 Australia 1 8.4 Australia 16.5
Netherlands 1 5 .9 German 16.3 Italy 14.9
Finland 13.4 Spain 9.7 Netherlands 1 1.3
Australia 1 3 .2 Netherlands 8 .6 Sweden 7. 1
Sweden 6.9 Finland 5.9 Spain 6.3
Spain 6.6 Sweden 4.2 Finland 5.9
Norway 4.7 Norway 3.7 Norway 2.8

 

* Reserve adequacy is the ratio of M2 to reserves in the month preceding the onset of the crisis (August
1992, November 1994 or June 1997).

232

Table 46. Reserve adequacy (Developing countries).

 

 

Country Aug. 1993 Country Nov. 1994 Country June 1997
India 19.8 Mexico 9.0 Pakistan 20.9
Pakistan 1 8.9 Pakistan 8.7 India 7.5
Korea 6.9 India 7.5 Korea 6.8
Jordan 6.3 Korea 6.7 Indonesia 6.2
Turkey 6.3 Indonesia 6.1 Thailand 4.9
Philippines 5.7 Philippines 4.8 Philippines 4.9
Indonesia 4.8 Turkey 4.4 Jordan 4.3
Mexico 4.4 Argentina 4. 1 Mexico 4. 1
Thailand 4.0 Brazil 4.0 Malaysia 4.0
Brazil 3.7 Jordan 3.9 Brazil 3.7
Sri Lanka 3.3 Thailand 3.8 Argentina 3.6
Argentina 3 .2 Norway 3.7 Turkey 3.2
Malaysia 2.8 Malaysia 2.1 Norway 2.8
Chile 1.7 Colombia 1.9 Sri Lanka 2.6
Venezuela 1.6 Sri Lanka 1.9 Colombia 2.0
Colombia 1 .0 Venezuela 1 .8 Chile 1.8

 

* Reserve adequacy is the ratio of M2 to reserves in the month preceding the onset of the crisis (August
1992, November 1994 or June 1997).

233

Table 47. Trade linkage’

 

 

Country 1992 Country 1994 Country 1996
Italy 1.00 Mexico 1.00 Thailand 1.00
France 0.86 Canada 0.54 Malaysia 0.83
Netherlands 0.80 Korea 0.45 Indonesia 0.81
UK. 0.79 Japan 0.36 Australia 0.70
Germany 0.62 Malaysia 0.34 India 0.68
Spain 0.56 UK. 0.34 Brazil 0.66
Japan 0.44 Venezuela 0.32 Philippines 0.65
Sweden 0.43 Brazil 0.31 Korea 0.63
U.S.A 0.38 Thailand 0.30 Finland 0.47
Korea 0.38 U.S.A 0.26 Sweden 0.46
Norway 0.35 Germany 0.25 Chile 0.44
Brazil 0.32 Italy 0.23 Norway 0.37
Malaysia 0.29 Philippines 0.23 Turkey 0.37
Thailand 0.27 Indonesia 0.22 Venezuela 0.37
Canada 0.25 India 0.22 Colombia 0.33
Finland 0.24 France 0.21 Argentina 0.3 1
Australia 0.23 Sweden 0.20 UK. 0.30
Mexico 0.23 Colombia 0.18 Spain 0.30
Indonesia 0.22 Chile 0.18 Italy 0.29
India 0.27 Australia 0.17 Pakistan 0.29
Venezuela 0.16 Spain 0.16 France 0.27
Turkey 0. 14 Argentina 0.1 5 Canada 0.26
Philippines 0. l 3 Norway 0.15 Netherlands 0.24
Chile 0.12 Finland 0.14 Mexico 0.24
Argentina 0. 1 2 Netherlands 0. l 4 Japan 0.22
Colombia 0.09 Pakistan 0.1 1 Germany 0.20
Pakistan 0.07 Turkey 0.11 Sri Lanka 0.17
Sri Lanka 0.03 Sri Lanka 0.07 U.S.A 0.12
Jordan 0.01 Jordan 0.00 Jordan 0.03

 

B1

‘3; r v- CTZ'Aifl-f‘ 4'1.“-
x

..--

 

* Trade is a weighted average index between the third market competition index, Indirect. and direct trade
linkage index, Direct.

234

 

Table 48. Country effects

 

Dependent variable: Crisis

 

 

 

 

Estimated Independent Pooled OLS Fixed effects
coefficients variable
[1, LB 0.084“ 0.054
(0.041) (0.110) ._
[)2 RER 0183‘ -0.211 P
(0.134) (0.196)
[)3 LB*dlir 0035 0.161
(0.106) (0.320) .
[1, RER*dhr -0.012 -0.056 1
(0.268) (0.484)
55 LB*d.sf -0241 -0230
(0.209) (0.579)
[)6 RER *dsf -021 1 -0.393
(0.241) (0.476)
flu Constant 3.901 .. 4.854”
(1.853) (2.640)
Sample
5‘“ 87 87
R2 0.123 0.361
11‘ 2 0.060 0.000

 

Note: Heteroscedasticity robust standard errors in parenthesis.
Significance at the 10 percent level is denoted by *; at the 5 percent level by "

235

Table 49. Benchmark regression

 

Dependant variable: Crisis

 

 

Estimated Independent Benchmark Industrial Developing Changes of LB
variable countries crisis countries crisis and RER
coefficients (1992) (1994, 1997)
)1, LB 0.08? -0043 0.330“ -0014
(0.041) (0.063) (0.112) (0.052)
'32 RER -0.183' -0.009 -0.180. -0.179
(0.134) (0.293) (0.123) (0.253)
[33 LB *dhr -0.035 0.425 -0.236" -0.224”
(0.106) (0.677) (0.114) (0.126)
fl, RER *dhr -0.012 -0.388 0.030 0.089
(0.268) (0.469) (0.192) (0.276)
[35 LB *dsf -0.241 -1.073 -0.343" —0.225"
(0.209) (0.839) (0.183) (0.121)
[36 RER *dsf -0.211 -0.945 -0.192 -0. 123
(0.241) (0.554) (0.153) (0.198)
B7 LB*ddev+ 0.316"
(0.112)
[)8 RER*ddev 0.036
(0.220)
[30 Constant 3.901" 10.156" -0.026 2.904.
(1.853) (3.843) (1.737) (1.993)
Sample size 87 87 87 87
R2 0.123 0.068 0.492 0.290
132 0.060 0.000 0.397 0.217
5, + 53 _ 0 0.31 [0.58] 0.33 [0.57] 7.54 [0.01]
[)2 + ,3, 2 0 0.55 [0.46] 1.79 [0.19] 0.01 [0.93]
)3, + 35 _ 0 0.00 [0.91] 1.76 [0.20] 1.64 [0.21]
g, 5. [B6 2 0 1.75 [0.20] 3.69 [0.06] 2.78 [0.11]

 

Note: Heteroscedasticity robust standard errors in parenthesis.
Significance at the 10 percent level is denoted by *; at the 5 percent level by **

P-values in brackets.

+ Dummy variable which is one when year is 1994 or 97

Table 50. Contagion effects

 

Dependant Variable: Crisis

 

 

Estimated Independent Benchmark Contagion Industrial Developing
coefficients variable with deg effects countries crisis countries crisis
(1992) (1994,1997)
fl, LB 0.083" 0.076“ -0.051 0.217"
(0.037) (0.028) (0.028) (0.080)
[72 RER -0. l 73’ -0.037 0.075 -0.004
(0.124) (0.117) (0.316) (0.108)
[33 LB*dhr*deg 0.010 0.076 0.619 0.027
(0.122) (0.101) (0.583) (0.098)
’84 RER *dlir *dcg -0.066 -0.127 -0.227 -0.019
(0.265) (0.218) (0.379) (0.178)
,35 LB *dsf*deg -0.267 -0.279 -1 .169. -0.273"
(0.235) (0.293) (0.705) (0.099)
flb RER *dsy‘*deg -0.184 -0.006 -0.736' -0.048
(0.256) (0.286) (0.458) (0.176)
’37 Trade 26.650" 7.331 32.952"
(9.304) (9.677) (7.650)
58 Pure 6.478‘ 7.441 7.076“
(3.892) (6.522) (3.891)
[19 Itrdus -0.399 -2.253" -0.293
(0.470) (1.175) (0.394)
:60 Constant 3.904" -6.582" 8.617” 41.636"
(1.706) (3.040) (6.613) (2.453)
Sample size 87 87 87 87
R2 0.123 0.352 0.184 0.724
1??- 0.057 0.276 0.000 0.672

 

Note: Heteroscedasticity robust standard errors in parenthesis.
Significance at the 10 percent level is denoted by *; at the 5 percent level by **

237

Table 51. Additional determinants

 

Dependent variable: Crisis

 

 

Estimated Independent GOVC GDP CAP] FORLB
co- variable (Government (Real GDP) (Capital inflow) (Foreign
efficients consumption) liability)
,7}, LB 0.080“ 0.07555 0.062“ 0.072“
(0.027) (0.028) (0.022) (0.029)

[)2 RER -0.043 -0.040 -0.051 -0.023
(0.117) (0.119) (0.112) (0.110)

’83 LB *dhr 0.073 0.085 0.058 0.080
*dcg (0.093) (0.107) (0.101) (0.111)

'84 RER *dhr -0.134 -0.107 -0.029 -0.114
*dcg (0.210) (0.235) (0.197) (0.225)

,85 LB *dsf -0.293 -0.289 -0.329 -0.256
*dcg (0.296) (0.320) (0.295) (0.296)

[)6 RER *dsjf -0.038 0.033 -0.l62 0.010
*dcg (0.285) (0.323) (0.253) (0.289)
[)7 Trade 26.434" 26.232" 26.745" 27.527"
(9.572) (9.545) (8.979) (9.452)

fig Pure 6.481. 6.236. 6.649. 6.656"
(3.871) (3.917) (3.971) (3.861)

[jg Itrdus -0.321 -0.462 -0.545 -0.464
(0.514) (0.502) (0.494) (0.500)

1;, 0 Added 0.157 -O.176 0.563“ -0059
Variable (0.210) (0.327) (0.319) (0.047)
[)0 Constant -8.938" -6.246" -7.916" -5.769"
I (3.452) (3.383) (3.488) (3.102)

Sample # 87 87 87 87
R 2 0.354 0.354 0.370 0.359
172 0.269 0.269 0.287 0.275

 

Note: Heteroscedasticity robust standard errors in parenthesis.
Significance at the 10 percent level is denoted by *; at the 5 percent level by **

238

Table 52. Robustness for the crisis index

 

Dependant Variable: Crisis

 

 

Estimated lndepen- Aug- Aug-Nov Aug-Dec Aug-Jan Aug-Feb Aug-Mar
co- dent Oct'" Nov-Feb Nov-Mar Nov-Apr Nov-May Nov-June
efficients variable Nov-Jan June-Sep June-Oct June-Nov June-Dec June-Jan
June-Aug
)3, LB 0.071“ 0.074“ 0.101“ 0.084“ 0.069‘ 0.086
(0.018) (0.016) (0.034) (0.041) (0.054) (0.081)
)3, RER 0051‘ -0122” -0.179” -0.183’ -0.256’ -0.350‘
(0.033) (0.049) (0.093) (0.134) (0.171) (0.268)
[)3 LB*dhr -0.039 -0.068 -0.088 -0035 -0.038 -0.068
(0.050) (0.081) (0.094) (0.106) (0.123) (0.144)
34 RER*dhr 0.086 -0043 -0054 -0.012 0.106 0.228
(0.1 16) (0.173) (0.220) (0.268) (0.329) (0.434)
[)5 LB*dsf -0.1 17 -0.048 -0.082 -0.241 -0204 -0.201
(0.095) (0.153) (0.173) (0.209) (0.217) (0.201)
[36 RER*dsf -0101 0.013 -0.060 -0211 -0.286 -0401
(0.1 12) (0.168) (0.209) (0.241) (0.277) (0.324)
30 Constant 1.595" 2.489“ 3.023“ 3.901 " 6.635” 10.114”
(0.559) (0.812) (1.457) (1.853) (2.569) (3.953)
Sample # 87 87 87 87 87 87
R2 0.145 0.167 0.165 0.123 0.081 0.055
172 0.080 0.105 0.102 0.060 0.012 0.000

 

 

 

 

Note: Heteroscedasticity robust standard errors in parenthesis.
Significance at the 10 percent level is denoted by *; at the 5 percent level by **
*"1992, 1994 and 1997 crisis, respectively.

239

 

r‘tl. 41.4. 3'1 G

 

.... iii... 2..... .21..

a... 3 326. 222828 m 65 an a... .3 88:86 «.2 36. E8222 2 2: 2m 858E525
£85528 5 £226 BEES... 8252 56223688286: ”362

 

 

 

83 83 228.22 Sod 82; 822.22 E
2.8.22 £2.22 83 2.8.22 222 23 N22
S S S 5 3 mm 22 62.2.5.6.
$9. 8 252. 8 882.8 :8. 8 6:. 8 232. 8 a
:23. :83. :23 :83. :43... .. Sam 2.8288 n
82.8 642.8 83.8 5&8 62.8 222.8 a
$3. .2328- ES. 82.22- $3- :22- 8.28.252 a
232.8 A: 2 .8 32.8 ASN.8 232.8 88.8 a
.822- ms 2 .o- :2- ammo- .263- 22.2- 8.26.2: -n
ASN.8 938 88.88 88.8 82.8 22288 a.
58.22 822.8 832 Red. $23 N222- 2.228.. «me n
38.8 2828 Go 2 .8 E 2.8 2.8.8 688 a
2.8.8 232. N88 63o- Rod- 98.22- 8226...“: a
82.8 83.8 62.8 2.28 :28 22.: .8 a
.322- 62.0- .32.? .2223- .2822- .328 «$2 a.
83.8 88.8 A 23.8 A 23.8 83.8 28.8 2
:83 :83 read :53 :82; tamed m: e.
5.2:- v m.— oé- v m.— a.o_- v m.— od- v m.— od- v m2 od- v m.—
o.~_ A mm... war. A amy— ofi A mud cm A 1mm o.” A Mix cm A amy—
odvoytdz 22.99252 @9932 m.vam\m2 899.282 8.9952 6322? 3826568
he vERE 9m v ENE o.» vER—z 00 v 2282 he v ENE ow v ENE 2820:8202: neaénm
3.28.8 m: 388 2.5228
8% mm? J2EE m.— ..w mm; m._ a. Mix @8on ~32 @eomv ~52 8228:5225

5th 8522223 “SN—62595

 

82:22:36 65 20.2 mmosznom .mm 2%...

240

Table 54. Actual and predicted currency crisis index.

 

 

 

 

 

 

 

Actual crisis index Predicted crisis index Predicted crisis index with
without contagion effects contagion effects

Country 1997 Country 1997 Country 1997
Thailand 47.9 Philippines 68.3 X" Philippines 75.6 X
Indonesia 33.5 Thailand 15.] X Thailand 33.2 X
Malaysia 32.0 Turkey 15.1 Indonesia 26.2 X
Philippines 29.7 Malaysia 8.3 X Malaysia 25.9 X
Korea 1 6.4 Netherlands 6.8 India 21 .3
Colombia 16.3 Jordan 6.5 Korea 17.8
Turkey 12.4 Indonesia 6.2 Pakistan 14.3 X
Pakistan 8.4 Canada 5.4 Turkey 14.0 X
Brazil 7.3 Argentina 4.5 Sri Lanka 13.1
Australia 4.6 Colombia 3.4 X Colombia 10.0 X
Japan 4.4 Venezuela 3.3 Chile 8.5
India 3.8 Germany 2.9 Australia 4.4
Finland 3.8 Korea 2.9 Canada 2.9 X
Canada 2.9 Australia 1.0 Japan 2.6 X
Netherlands 1 .9 UK. -1 .4 Argentina 2.6
Germany 1 .3 Chile -1 .5 Netherlands 0.7
USA 1.2 Sri Lanka -2.7 X Norway -1.8
UK 0.8 USA -3.6 X UK. -1.8 X
Sri Lanka 0.3 Sweden -3.7 X Spain -2.8
Sweden 0.3 Finland -4.7 Jordan -3.4
Mexico 0. 1 Pakistan -5.1 Germany -4.1
Chile -O.5 Italy -6.5 Italy -7.5
Argentina -0.7 Norway -7.3 USA. -9.3
France -1.5 Japan -7.6 Finland -10.7
Jordan -2.3 India —7.7 Sweden -10.9
Spain -2.9 Spain -10.6 X France -13.6
Norway -4.5 France -1 3 .8 Mexico -1 7. 1
Italy -7.7 Mexico -20.3 Brazil -20.6
Venezuela -11.0 Brazil -33.3 Venezuela -22.0 X

 

* The currency crisis index (Crisis) is a weighted average of the percentage depreciation of nominal
exchange rate with respect to the US. dollar and the percentage decrease in reserves.

** The country which can be matched with the one in the group of 5 countries which is clustered by the
descending order of crisis index

241

x0922 mmmco 5222.228 88:55 98 322232. .mo 85?..—

503282282

.38»? =2w8228 523 can 225523 386822 2m 22023 3.05 2n = Ea 2 282:: are 38:65 ..

 

 

 

2
2 .822 mm; 8228.". ....... .822 328 222o<|II|2

 

 

 

 

 

 

__ .822 828 8202.8:
4 v / «v 291V
00 ... 4v v Ar 0 00.. o o b 0.. A/ o, o
o J. 9 av a e co . n o o y» 0 1. o o w. «u .
00.2. 9&0 09%.. 010 )mm, «V9 920.1 «o90 00/ 06) eavo 494%0/ 40 8.0 cows 92/ 8% 900.. 94%. 9%.» 009 one 9% 090 000,. 09% .0 .®0
9/ x. 0 so 09 ex. 9 ex. ex. 9.7 0 do by. ex. A. e 1/ ex. 0;. no 0 so 0 f ....o s f 11 ......
. mm-
/ .. _ .. 2.).
.. 2 \\
/ .oc .. .. ..o. o/... ... o // \\
. . . . o . .. . k c I. . as t . \o.. . mu
. . .. . X ,. . .. . .\ .\./..~ 5. /.l. . . ’..4\..
. o u . I \c o . /\i . ...
. . \. m2
,/ .
x .. .
/ .. ..
. .. mm
/.. ... 3
”Q
mm

 

 

mm

242

 

 

 

 

 

 

 

 

 

 

 

 

 

98222: 2.5002
22%on :2 2202w8000 .20 0222025 8:30.222 22222 822222200 2.32
.25 $22.2 022.20 02258 2.8 ”82.22 5% 22...: - W20 2282.25. 22223 h .2... - £2 .32 :3 22 82.2;
6222m228>222
2m 2202232200 .20 88:» 2222228 220252200 2:. - 2222225
022222228 2a 23. 2025 00:82 :2 2220 82:22:00 23222: 228 22222202
2: :0 b22322 2: =2 2202322a> 22225008820 2:. - mqo w=2w22=o mm 302 - £02 800 22 2222.20...
200332832
2a :02w5000 :0 98:0 292.2208 00:22:00 2:. -
22:0 :2: 2220 2222228 230:
2282082 2: .3 228:2. 20:2 2025 32:22:00 82:22:00 Hoe—SE 228 22222202 00mm2o> :28
2220... .2223 222222222qu :0 2 222822an 2:.- mac 22292220 cm .302 - 02202 .22uE0.2. .2203
.8220 222223. :28
2282232 2: 222 0202 20.3222 22 2.0.3222 3:222 2022 28
300:0 222.2208 005800 2: .3 26022222222: 23:0 82:22:00 282.88 222225
2202232200 2: 22: amowwsm .2203» .20 :32 2:. - 3222222222 2222; 22292220 2 .0002 - 302 $00 22 22882
2222293022
222 822222 20a: :5 38.2.2» 22222228 2208800 2:. -
22228223. :28.
2a 3:23:53 02222000000202: :5 32:22 25:. - 80022 50222.23
38.22: owcanoxo 82:22:00 322.2220 205 332
22320.2 :2 220382200 20.2 $2 0: 0202.. 22:. - 200222 3:22:02 cm .802 - 032 2202922205
62:23
2a 8222: 2022 :5 $02.20 EESxo 008600 2:. -
.2022: 2220222225 .3 22:03: 200: 82:22:00 2222222224.. 300 22
.20 22.22.2020 28:22 2220 2222.50 22: 3022... 22:. .. mAO :52 222022 8222222200 202 .2202 -2 2.02 832 2028 020220
22.23 2228 3322
2222.228 2: mats: 22280556 262222 0: 22220.2 £002 .2 222.22.220.00
2m 82222.23 wcosa 2202322200 2222222093020 - m<> 82:22:00 m .32 - 302 .2 .2222 2022a wfim
2022222002:
3222220200 202:2: 28222925 022030 E22500 :5 0226mm 22an

 

 

82:05 28222an 220322 .3 2an

243

CHAPTER VIII

CONCLUSION

This dissertation has studied the causes of the Asian currency crisis and improved
the performance in predicting actual currency crises. The empirical analyses presented in
this dissertation examined the Asian currency crisis using higher frequency data and more
refined models than previous studies.

First, an extension of the structural currency crisis model is presented to derive
shadow exchange rate and the probability of an exchange rate regime change. The model
is a stochastic version of the monetary approach to exchange rate determination. While
the nonstructural studies’ results are not robust and do not forecast crises well, the results
of our structural study indicate that weak fundamentals in the selected Asian countries,
South Korea and Malaysia, prior to the Asian currency crisis already had been predicting
the upcoming currency crisis.

Second, to analyze currency crises extensively, a pooled OLS using panel data is
estimated, focusing on the importance of contagion effects on the eruption of a currency
crisis. The empirical results show that a lending boom and contagion effects sufficiently
explain the cross-country variation in the severity of the crisis of emerging markets. In
addition, the prediction of currency crisis based upon the contagion effect is found to
improve our ability to predict an eruption of currency crisis.

In Chapter 11, an overview of the beginning and development in the Asian crisis is
presented with a focus on the movements of the macroeconomic variables and the

structural conditions of financial system. The evidence given in the overview indicates

244

that deterioration in macroeconomic fundamentals and poor economic policies were a
root cause of the crises. Nonetheless, the evidence is not convincing enough to establish
that fundamentals had deteriorated so severely that the outbreak of the Asian currency
crisis was inescapable.

Chapter 111 provides a survey of the theoretical and empirical literature on
currency crises and introduces an extended structural currency crisis model. The
theoretical literature about the currency crises contains a number of models which either
support the ‘fundamentalist’ or ‘non-fundamentalist’ views as to the causes of currency
crises. These two views are represented by first and second generation models of
currency crises.

Based upon the logics of the theoretical models, the empirical literature has
attempted to determine the actual sources of various cm'rency crises. The empirical literature
grouped into two categories: nonstructural and structural analyses. Nonstructural analyses
exploited the high variability associated with cross-country information with the
limitation by the lack of robustness to various sensitivity tests and poor performance in
predicting actual crises. Beginning with the Blanco and Garber’s (1986) study, structural
analyses have presented strong evidence suggesting that domestic macroeconomic
indicators play a key role in determining a currency crisis.

To determine whether the Asian crisis was distinct relative to other crises and
improve the performance in the prediction of crises, the currency crisis is modeled with a
structural model. As a result of the modeling, the influence of pure macroeconomic
fundamentals on exchange market pressures for the Asian currencies can be evaluated.

Furthermore, a spurious regression problem caused by the non-stationarities of relevant

245

processes is resolved by using an error correction model (ECM) for the extended
structural model.

Chapter IV then offers an analysis of the time series properties and forecasts of
each variable of the structural currency crisis models introduced in Chapter III for the
derivation of shadow exchange rates and probabilities of collapse. Unlike most previous
studies, to capture the properties of economic and financial time series that exhibit long
memory in both their conditional mean and variances, the ARFIMA(p,d,q)-
FIGARCH(P,5,Q) model is included in the selection of models in Chapter IV. Based on
the Wald tests, the US. inflation rate, the percentage changes of deviations from PPP in
Indonesia, the percentage changes of domestic credits in Malaysia and Thailand and the
change rates of real GDP in Malaysia appear to have estimated long memory parameters
d and 6 which lie in the ranges of — 0.5 < d < 0.5 and O < 6 <1.0 , respectively. However,
Wald tests do not find evidence for dual long memory behavior in other processes.
Following the analysis of time series properties, the estimated parameters are used to
forecast each variable. The forecasted variables are used to generate shadow exchange
rates and probabilities of collapse in Chapter VI.

In Chapter V, long and short-run real money demand functions are estimated in
the structural model used to derive the shadow exchange rates and probabilities of
collapse. The previous structural analyses of currency crises estimated a real money
demand function without taking into consideration the variable’s non-stationarity.
Therefore, the estimated money demand functions faced a spurious regression problem
whereby conventional t-ratio and F significance tests could not be applied. To avoid these

problems. cointegration and error correction techniques are applied to model a real

246

money demand. The unit-root tests presented in the following sections detect non-
stationarity of real money balance, real GDP. interest rate for South Korea and real
money balance for Malaysia. Then, a residual based test was used to test for the number
of cointegration relations and to estimate the cointegrating vectors. These analyses
suggest that both long and short-run models can be specified in South Korea and
Malaysia. Nonetheless, we cannot determine whether a model without a deterministic
time trend can explain real money demand in South Korea. The use of monthly data
enables this study to determine that long-run real money demand is unstable when
modeled as a cointegrating relationship without a deterministic time trend in South
Korea.

Chapter VI derives the shadow exchange rates and the probabilities of an
exchange rate regime change for South Korea and Malaysia. Two countries experienced a
severe devaluation during the Asian currency crisis. The shadow exchange rates and
probabilities of collapse rely on earlier forecasts for relevant variables and estimates of
the real money demand functions.

As shown in Figures 49 and 50, the estimated shadow exchange rate of each
country signals the possibility of an upcoming severe depreciation by following the
behavior of fundamentals. In particular, South Korea’s shadow exchange rate was far
above the regulated exchange rate since 1995. For Malaysia, the gap between the shadow
and controlled exchange rates became noticeable around 1994. It seems that the gap
between the shadow and actual exchange rates attracted speculators by the potential profit
to be gained following a change in exchange regimes. The changes in the probability of

collapse, which is positively correlated with the difference between shadow and actual

247

exchange rate, indicate that both South Korea and Malaysia were under severe
depreciation pressure starting in the early 19905 and continuing up until the Asian
currency crisis.

In Chapter VII, a pooled OLS is estimated on panel data with a focus on the
influence of contagion effects on the currency crises. The first result found is that lending
booms impact the crisis index much more among developing countries than industrial
countries. Under the strong and stable financial systems of industrial countries, the
lending booms appears not to contribute to the variation of the crisis index. The second
result is that contagion effects as represented by trade linkage between the initially
attacked and home country as well as market sentiment are a significant source of the
eruption of currency crises during the 19903. Furthermore, the contagion effects are
stronger in emerging markets than in industrial countries. Thirdly, evidence such as
adjusted-R2, RMSE , and matched number of countries supports that controlling for
contagion effects improves the ability to predict a currency crisis.

Currency crises in the 19703 and 19805 were rooted in the dynamics of domestic
credit extended by the central bank to the government. However, in the 1997 Asian
currency crisis, domestic credit to the government did not play a crucial role. As shown
in Chapter II, the domestic credit to the private sector enhanced by the domestic banks
fueled by the foreign liabilities played a pivotal role in the latter currency crisis. In
particular, as shown in Table 9, domestic bank lending to the private sector increased
among Asian countries prior to the crises, leading to a sever lending boom. To make
matters worse, many of the loans made by banks were invested in risky and low profit

projects or used for real estate, property and the purchase of equity funds due to the moral

248

hazard problem. The shadow exchange rates and probabilities of collapse derived in
Chapter VI using M2 as a money supply and the empirical results in Chapter VII that
indicate the significance of lending booms confirm that lending booms played a key role
in the eruption of the Asian currency crisis.

The 1994 Mexican crisis and the 1997 Asian crisis had a widespread contagion to
several emerging markets. Currency crises were a regional phenomenon in the 19905. As
expected, the contagion effects are found to play a crucial role in the Asian currency
crisis based on the empirical results in Chapter VI. In particular, strong trade linkage and
market sentiment were essential in the eruption of Asian currency crisis.

In conclusion, given the evidence presented by the estimated shadow exchange
rates and the probabilities of collapse of South Korea and Malaysia and by the empirical
results in Chapter VII, weak fundamentals and contagion effects in the Asian countries
had been indicating that a currency crisis could erupt whenever unexpected events such
as bank failure, corporate failure and unstable political condition may trigger it in the
Asian countries. Therefore, the implications for economic policy based on the Asian
currency crisis are that countries that support sound macroecOnomic and monetary
policies are not readily vulnerable to a currency crisis and a country that has an initially
attacked neighbor should perform more stable macroeconomic and monetary policies

while trying to weaken the channel through which the contagion effects are delivered.

249

BIBLIOGRAPHY

Agenor, P.-R., J. S. Bhandari, and R. P. Flood (1992), “Speculative attacks and models of
balance of payments crises,” IMF Staff Papers 39, 357-394.

Akerlof, G. A. (1980), “ The short run demand for money,” The Economic Journal 90,
885-900.

Akerlof, G. A., and R. D. Milboume (1980), “Irving Fisher on his head 11: The
Consequences of the timing of payments for the demand for money,” The Quarterly
Journal of Economics 95, 145-57.

Amsler, C. (1999), “Size and power: Lower tail KPSS tests and anti-persistent
alternatives,” Applied Economics Letters 6, 693-95.

Arize, A. C. (1994), “A Re-Examination of the demand for money in small developing
economies”, Applied Economics 26, 217-28.

Ariz, J ., F. Cararnazza, and R. Salgado (2000), “Currency crisis: In search of common
elements”, IMF Working Papers No. 00/67.

Baillie, RT. (1996), “ Long memory processes and fractional integration in
econometrics”, Journal of Econometrics 73, 5-59.

Baillie, R.T., C.-F. Chung, and MA. Tieslau (1996), “Analyzing inflation by the
fractionally integrated ARFIMA-GARCH model”, Journal of Applied Econometrics
11, 23-40.

Baillie, R.T., and RA. Pecchenino (1991), “Equilibrium relationships in international
finance”, Journal of lntemational Money and Finance 10, 582-593.

Baillie, R.T., and T. Bollerslev (1992), “ Prediction in dynamic models with time-
dependent conditional variances” , Journal of Econometrics 52, 91-113.

Baillie, R.T., and T. Bollerslev (1994), “ Cointegration, fractional cointegration and
exchange rate dynamics” , Journal of Finance 49, 737-745.

Baillie, R.T., T. Bollerslev and H0. Mikkelsen (1996), “ Fractionally integrated
generalized autoregressive conditional heteroskedasticity” , Journal of Econometrics
74, 3-30.

Barro, R. J ., and D. B. Gordon (1983), “Rules, discretion and reputation in a model of
monetary policy”, Journal of Monetary Economics 12, 101-21.

250

Barro, R. J., and S. Fischer (1976), “Recent development in monetary theory”, Journal of
Monetary Economics 2, 133-67.

Baumol, W. J. (1952), “The transactions demand for cash: An inventory theoretic
approach,” The Quarterly Journal of Economics 66, 545-56.

Bollerslev, T. (1986), “Generalized autoregressive conditional heteroskedasticity”,
Journal of Econometrics 31, 307-27.

Blanco, H., and P. M. Garber (1986), “Recurrent devaluation and speculative attacks on
the Mexican Peso,” Journal of Political Economy 94, 149-166.

Boorman, J. T. (1976), “The evidence on the demand for money: theoretical and
empirical results,” in current issues in monetary theory and policy, ed. by T. M.
Havrilesky and J. T. Boorrnan (Arlington Heights, Illinois: AHM Public Corporation,
1976), 315-60.

Boughton, J. M. (1979), “Demand for money in major OECD countries,” OECD
economic outlook: occasional studies (Paris: Organization for economic cooperation
and development, January 1979).

Boughton, J. M. (1981), “Recent instability of the demand for money: An international
perspective,” Southern Economic Journal 47, 579-97.

Calvo, G. A., and E. G. Mendoza (1997), “Rational herd behavior and the globalization
of securities markets,” mimeo, University of Maryland.

Chiang, C. Y. (1999), “The long-run demand for money functions in Taiwan (1961:4-
199723) : Cointegration and stability,” Ph.D. dissertation, Department of Economics,
Michigan State University.

Chow, G. C. (1966), “On the long-run and short-run demand for money,” Journal of
Political Economy 74, 111-31.

Cooper, R. (1996), “Comment,” Brookings Paper on Economic Activity 1, 203-208.

Corsetti, G., P. Pesenti, and N. Roubini (1998a), “Fundamental determinants of the Asian
crisis: A preliminary empirical assessment,” mimeo.

Corsetti, G., P. Pesenti, and N. Roubini (1998b), “Paper tigers?: A model of the Asian
crisis,” NBER Working Paper No. 6783.

Corsetti, G., P. Pesenti, and N. Roubini (1998c), “What caused the Asian currency and
financial crisis? Part I and II,” NBER Working Papers No. 6833 and 6834.

251

Craig, B., M. Eichenbaum, and S. Rebelo (1998), “Perspective deficits and the Asian
currency crisis,” NBER Working Papers No. 6758.

Cumby, R. E., and S. V. Wijnbergen (1989), “Financial policy and speculative runs with
a crawling peg: Argentina 1979-1981,” Journal of lntemational Economics 27, 111-
127.

Davidson, R., and J. G. MacKinnon, (1993), Estimation and inference in econometrics
(Oxford: Oxford University Press, 1993).

Dickey, D.A., and WA. Fuller (1979), “Distribution of the estimators for autoregressive
time series with a unit root”, Journal of American Statistical Association 74, 427-31.

Dickey, D.A., and WA. Fuller (1981), “Likelihood ratio statistics for autoregressive time
series with a unit root”, Econometrica 49, 1057-72.

Dombusch, R. (1998), “Asian crisis themes,” mimeo, MIT.

Eichengreen, B., A. K. Rose, and C. Wyplosz (1996), “Speculative attacks on pegged
exchange rates: An empirical exploration with special reference to the European
monetary system,” in M. B. Canzoneri, W. J. Ethier, and V. Grilli (eds), The New
Transatlantic Economy, New York: Cambridge University Press, 191-235.

Engel, RF. (1982), “Autoregressive conditional heteroskedasticity with estimates of the
variance of UK Inflation”, Econometrica 50, 987-1008.

Engle, RF, and C.W.J. Granger (1987), “Co-integration and error correction:
Representation, estimation, and testing,” Econometrica 55, 251-76.

Engel, RF, and T. Bollerslev (1986), “Modeling the persistence of conditional
variances”, Econometric Reviews 5, 1-50.

Fair, R. C. (1987), “International evidence on the demand for money,” The Review of
Economics and Statistics 69, 473-80.

Feige, E. L., and D. K. Pearce (1994), “The substitutability of money and near-monies: A
survey of the time-series evidence,” Journal of Economic Literature 15, 439-69.

Fisher, 1. (1922), The purchasing power of money, its determination and relation to
credit interest and crises by I. Fisher assisted by H. G. Brown. (New York:
Macmillan, 1922).

Fischer, A. M. (1993), “Is money really exogenous? Testing for weak exogeneity in
Swiss money demand,” Journal of Money, Credit, and Banking 25, 248-58.

252

 

Fischer, S. (1975), “The demand for index bonds,” Journal of Political Economy 83,
509-34.

Flood, R. P., and R. J. Hodrick (1986), “Real aspects of exchange rate regime choice with
collapsing fixed rates,” Journal of lntemational Economics 21, 215-32.

Flood, R. P., and N. Marion (1998), “Perspectives on the recent currency crisis
literature,” NBER Working Paper No. 6380.

Flood, R. P., and P. M. Garber (1984a), “Collapsing exchange-rate regimes: Some linear
examples,” Journal of lntemational Economics 17, 1-13.

Flood, R. P., and P. M. Garber (l984b), “Gold monetization and gold discipline,” Journal
of Political Economy 92, 90-107.

Frankel, J. A., and A. K. Rose (1996), “Currency crashes in emerging markets: An
empirical treatment,” Journal of lntemational Economics 41, 351-66.

Friedman, M. (1956), “The quantity theory of money-a restatement,” in studies in the
quantity theory of money, ed. by M. Friedman (Chicago: University of Chicago
Press, 1956).

Fuller, W.A. (1976). Introduction of statistical time series, (Wiley: New York, NY.
1976).

Furrnan, J., and J. E. Stiglitz (1998), “ Economic crises: Evidence and insights from
Asia”, Brookings papers on economic activity 0, 1-114.

Ghysels, E.(1990), “Unit-root tests and the statistical pitfalls of seasonal adjustment: The

case of US. postwar real gross national product”, Journal of Business and Economic
Statistics 8, 145-348.

Glick, R., and A. K. Rose (1998), “Contagion and trade: Why are currency crisis
regional?”, NBER Working Paper No. 6806.

Goldberg, L. S. (1988), “Collapsing exchange rate regimes: A theoretical and empirical
investigation,” Ph.D. dissertation, Department of Economics, Princeton University.

Goldberg, L. S. (1994), “Predicting exchange rate crises: Mexico revisited,” Journal of
lntemational Economics 36, 413-430.

Goldfeld, S. M. (1973), “ The demand for money revisited,” Brookings Papers on
Economic Activity 3, 577-646.

253

 

Goldfeld, s. M., and D. E. Sichel (1990), “The demand for money,” in Handbook of
Monetary Economics, Volumn 1, ed. by B.M. Friedman and PH. Hahn
(Amsterdam: Elsevier Science Publishers B.V, 1990), 300-56.

Gonzalo, J. (1994), “Five alternative methods of estimating long-run equilibrium
relationships”, Journal of Econometrics 60, 203-233.

Goodfriend, M. (1985), “Reinterpreting money demand regressions,” Carnegie-Rochester
Conference Series on Public Policy 22, 207-42.

Gordon, R. J. (1984b), “The short-run demand for money: A reconsideration,” Journal of
Money, Credit, and Banking 16, 403-34.

Granger, C.W.J. (1980), “Long memory relationships and the aggregation of dynamic
models”, Journal of Econometrics 14, 227-238.

Granger, C.W.J. (1981), “Some properties of time series data and their use in
econometric model specification”, Journal of Econometrics 16, 121-130.

Granger, C.W.J. (1986), “Developments in the study of cointegrated economic
variables”, Oxford Bulletin of Economics and Statistics 48, 213-28.

Granger, C.W.J., and R. Joyeux (1980), “An introduction to long memory time series
models and fractional differencing”, Journal of Time Series Analysis 1, 15-39.

Hansen, BE. (1992), “Efficient estimation and testing of cointegrating vectors in the
presence of deterministic trends”, Journal of Econometrics 53, 87-121.

Hansen, BE. (1992), “Efficient estimation and testing of cointegrating vectors in the
presence of deterministic trends”, Journal of Econometrics 53, 87-121.

Hendry, D. F. (1986), “Econometric modeling with cointegrated variables: An
overview”, Oxford Bulletin of Economics and Statistics 48, 201-12.

Hurst, HE. (1951), “Long-term storage capacity of reservoirs”, Transactions of the
American Society of Civil Engineers 116, 770-799.

Hylleberg, S. (1995), “Tests for seasonal unit roots: General to specific or specific to
general?”, Journal of Econometrics 69, 5-25.

Jarque, C.M., and AK. Bera (1980), “Efficient tests for normality, homoscedasticity and
serial independence of regression residuals”, Economic Letters 6, 255-59.

Jeanne, O, (1997), “Are currency crises self-fulfilling? A test,” Journal of lntemational
Economics 43, 263-286.

254

Johansen, S. (1988b), “Statistical analysis of cointegration vectors.” Journal of Economic
Dynamics and Control 12. 231-54.

Johansen, S. (1991), “Testing weak exogeneity and the order of cointegration in UK
money demand data”, Journal of Policy Modeling 14, 313-334.

Johansen, S. (1992), “Estimation and hypothesis testing of cointegration vectors in
gaussian vector autoregressive models”, Econometrica 59, 1551-80.

J ohansen, S., and K. Juselius (1990), “Maximum likelihood estimation and inference
on cointegration — with application to the demand for money”, Oxford Bulletin of
Economics and Statistics 52, 169-210.

J ohansen, S., and K. Juselius (1992), “Testing structural hypotheses in a multivariate

cointegration analysis of the PPP and the UIP for the UK”, Journal of Econometrics
53, 211-44.

Judd, J. P., and John L. Scadding (1982), “The search for a stable money demand
function: A survey of the post- 1973 Literature,” Journal of Economic Literature 20,

993-1023.

Kaminsky, G. L., and C. M. Reinhart (1999), “The twin crises: The cause of banking and
balance-of-payments problems,” American Economic Review June, 473-500.

Kaminsky, G. L., S. Lizondo, and C. M. Reinhart (1997), “Leading indicators of currency
crisis,” IMF Working Papers No. 97/79.

Keynes, I. M. (1930), A treatise on money (London and New York: Macmillan, 1930).

Keynes, J. M. (1936), The general theory of employment, interest, and money (London
and New York: Macmillan, 1936).

Kim, S. (2000), “What caused the Asian currency crisis?,” Ph.D. dissertation, Department
of Economics, Michigan State University.

Klein, M. W., and N. P. Marion (1997), “Explaining the duration of exchange-rate pegs,”
Journal of Development Economics 54, 387-404.

Krugman, P. (1979), “A model of balance-of-payments crises,” Journal of Money, Credit
and Banking 11, 311-325.

Krugman, P. (1998), “What happened to Asia?,” mimeo, MIT.
Kwiatkowski, D., P.C.B. Phillips, P. Schmidt, and Y. Shin (1992), “Testing the null

hypothesis of stationary against the alternative of a unit root: How sure are we that
economic time series have a unit root?”, Journal of Econometrics 54, 159-178.

255

 

Kwon, T. (1997), “Long memory modeling of conditional means and variances with
applications to asset pricing,” Ph.D. dissertation, Department of Economics,
Michigan State University.

Kydland, FE, and EC. Prescott (1977), “Rules rather than discretion: The inconsistency
of optimal plans”, Journal of Political Economy 85, 473-91.

Laidler, D. E.W. (1985), The demand for money: Theories and evidence (New York:
Dunn-Donnelley, 3rd edition, 1985).

Lee, H.S., and C. Amsler (1997), “Consistency of the KPSS unit root test against
Fractionally integrated alternative,” Economic Letters 55, 151-60.

Ljung, G., and G. Box (1979), “ On a measure of lack of fit in time series models”,
Biometrika 66, 265-270.

Lucas, R. E., Jr. (1980), “Equilibrium in a pure currency economy,” Economic
Inquiry 18, 203-20.

Marshall, A. (1923), Money, credit and commerce (London: Macmillan, 1923).

Masson, P. R. (1998), “Contagion effects: monsoonal effects, spilllovers, and jumps
between multiple equilibria,” mimeo, IMF.

Milboume, R. (1983), “Optimal money holding under uncertainty,” lntemational
Economic Review 24, 685-98.

Milboume, R. (1988), “Disequilibrium buffer stock models: A survey,” Journal of
Economic Surveys 2, 187-208.

Milboume, R., P. Buckholtz, and MT Wason (1983), “A theoretical derivation of the
functional form of short-run money holdings,” The Review of Economic Studies 50,
531-41.

Miller, M. H. (1968), “ The demand for money by firms: Extensions of analytic results,”
Journal of Finance 23, 735-59.

Miller, M. H., and D. Orr (1966), “ A model of the demand for money by firms,” The
Quarterly Journal of Economics 80, 413-35.

Morris, 8.. and HS Shin (1995), “Informational events that trigger currency attacks,”
Working paper No.95-24, Federal Reserve Bank of Philadelphia.

Nelson, CR. and GI. Plosser (1982), “Trends and random walks in macroeconomic time
series”, Journal of Monetary Economics 10, 139-162.

256

 

Obstfeld, M. (1986), “Rational and self-fulfilling balance-of-payments crises,” American
Economic Review 76, 72-81.

Obstfeld. M. (1994). “The logic of currency crises,” Cahiers Economiques et Monetaires
43, 189-213.

Obstfeld, M. (1996), “Models of currency crises with self-fulfilling features,” European
Economic Review 40, 1037-1047.

Obstfeld, M. (1997), “Destabilizing effects of exchange-rate escape clauses,” Journal of
lntemational Economics 43, 61 -77.

Otker, 1., and C. Pazarbasioglu (1996), “Speculative attacks and currency crisis: The
Mexican experience,” Open Economies Review 7, 535-552.

Otker, I., and C. Pazarbasioglu (1997a), “Likelihood versus timing of speculative attacks:
A case study of Mexico,” European Economic Review 41, 837-845.

Otker, I., and C. Pazarbasioglu (1997b), “speculative attacks and macroeconomic

fundamentals: Evidence from some European countries,” European Economic
Review 41 , 847-860.

Park, D., and C. Rhee (1998), “Currency crisis in Korea: Could it have been avoided?,”
mimeo.

Park, Y. C. (1996), “East Asian liberalization, bubbles, and the challenges from China,”
Brookings Papers on Economic Activity 2, 357-371.

Patinkin, D. (1965), MOney, interest, and prices: An integration of monetary and value
theory (New York: Harper and Row, 2nd edition, 1965).

Perron, P. and T]. Vogelsang (1992), “Nonstationarity and level shifts with an
application to purchasing power parity”, Journal of Business and Economic Statistics

10, 301-320.

Phillips, P.C.B. (1987), “Time series regression with a unit root”, Econometrica 55, 277-
301.

Phillips, P.C.B. and BE. Hansen (1990), “Statistical inference in instrumental variables
regression with 1(1) Processes”, Review of Economic Studies 57, 99-125.

Phillips, PCB. and P. Perron (1988), “Testing for a unit root in time series regression”,
Biometrika 75, 335-346.

257

Phillips, PCB. and S. Ouliaris (1990), “Asymptotic properties of residual based tests for
cointegration”, Econometrica 58, 165-193.

Pigou, A.C ( I917), “The value of money”, The Quarterly Journal of Economics 37, 38-
65.

Radelet, S., and J. D. Sachs (1998a), “The east Asian financial crisis: Diagnosis,
remedies, prospects,” Brooking Papers on Economic Activity 1, 1-74.

Radelet, S., and J. D. Sachs (1998b). “The onset of the east Asian financial crisis,” NBER
Working Paper No. 6680.

Rose, A. K. (1985), “An alternative approach to the American demand for money”,
Journal of Money, Credit, and Banking 17, 439-55.

Sachs D. J., A. Tomell, and A. Velasco (1996), “Financial crises in emerging markets:
The lessons from 1995,” Brookings Papers on Economic Activity 1996, 147-198.

Salant, S.W., and D.W. Henderson (1978), “Market anticipation of government gold
policies and the price of gold”, Journal of Political Economy 86, 627-648.

Samuelson, P. A. (1958), “ An exact consumption loan model of interest with or without
the social contrivance of money”, Journal of Political Economy 66, 467-82.

Schwert, G.W. (1989), “Test for unit roots: A Monte Carlo investigation”, Journal of
Business and Economic Statistics 7, 147-59.

Sowell, RB. (1992), “Maximum likelihood estimation of stationary univariate
fractionally integrated time series models”, Journal of Econometrics 53, 165-188.

Sriram, S. S. (1999a), “Demand for M2 in an emerging-market economy: An error-
correction model for Malaysia”, IMF Working Paper No. 99/173.

Sriram, S. S. (1999b), “Survey of literature on demand for money: Theoretical and
empirical work with special reference to error-correction models”, IMF Working
Paper No. 99/64.

Stock, J .H., and M.W. Watson (1988), “Testing for common trends”, Journal of the
American Statistical Association 83, 1097-1107.

Tan, E.U. (1997), “Money demand amid financial sector developments in Malaysia,”
Applied Economics 29, 1201-1215.

Tobin, J. (1956), “The interest-elasticity of transactions demand for cash,” The Review of
Economics and Statistics 38, 241-47.

258

 

Tobin, J. (1958), “Liquidity preference as behavior towards risk,” The Review of
Economic Studies 25, 65-86.

Tomell, A. (1999), “Common fundamentals in the tequila and Asian crises.” NBER
Working Paper No. 7139.

Whalen, E. L.(1966), “A rationalization of the precautionary demand for cash,” The
Quarterly Journal of Economics 80, 314-24.

Willman, A. (1988), “The collapse of the fixed exchange rate regime with sticky
Wages and imperfect substitutability between domestic and foreign bonds,”
European Economic Review 32, 1817-1838.

Wooldridge, J. M., (2000), Introductory Econometrics: A modern approach. South-
Westem College Publishing.

Yoshida, T. (1990), “On the stability of the Japanese money demand function: Estimation

results using the error correction model,” Bank of Japan Monetary and Economic
Studies 8, 1-48.

259

 

          

       

M IIIIIIIIIIIIIIIIIIIII

Illljllllllllllllllllolllllllllllllll

93 O 72 9910