. up... 6.5%”! dung .kfiafldhflélrfimud . pLHWVWnWmmMmq . .. 61 f. .at .. «am? .333 q. ll ii. To 6 i . Evita. :6: ..Iu . rakluaatl. I6. Jillian 5 In I.‘ unfr- .f' .‘i ..Ial. 3.1,: I... 2:651. . 6.9 x. «sit-Pvtvfieiyt 330‘ . . 541:2..3‘? (0.47.! ult:l.7...v . t . . 32b. . 6.3.5.65. ll . . . II ....I 3.9-35‘661. . . 6:: 13.1.76 .vlotsillfi) u C . . . .76 :2. .. \ixtvxl 6.1.: . ... . . 3:56:31 . . 6... . . . : .... . . . I6 ‘9 7. .5. 6.. . . . . . .I 16:312. . fl.v.=rv. 566 .' . . .... . filio‘l 1 .2 .. . .,|6|6|'v...toval6:(. I.” . . .. . .:.vlfn..!6.6|f..l¢|(.» . . . . . .. .........561.r! 9 .. . . . . . - . ..Yl......i.. .. . . .. ...I..I!:.A .. . . ...... . . .) .i.|(..- . -....6. . q 36.. . ‘ ... . ‘II I I III ‘III II, | . | I l Micmoms 7/ OZVI'I/X .9 ; JI/ IIIII I IIIII II IIIII III II III II IIIIIIIIII —_._ I' fl LIBRARY Michigan State University J _ II Id This is to certify that the dissertation entitled MONEY DEMAND RELATIONSHIP AND MONETARY TARGETING IN A CHANGING FINANCIAL ENVIRONMENT: THEORETICAL CONJECTURE AND EMPIRICAL EVIDENCE FROM KOREA presented by IBYUNG HAN SEO has been accepted towards fulfillment of the requirements for PH. D. degree in ECONOMICS é/fl Major professor Date 6/22/90 MSULt an Affirmative Action/Equal Opportunity Institution 0-1277} PLACE N RETURN BOX to remove this checkout from your record. TO AVOD FINES return on or before due due. DATE DUE DATE DUE DATE DUE ” ‘| ___I II ——II_—II II MSU Is An Afflrmdlve Action/Equal Opportunity Institution chS-pt MONEY DEMAND RELATIONSHIP AND MONETARY TARGETING IN A CHANGING FINANCIAL ENVIRONMENT: THEORETICAL CONJECTURE AND EMPIRICAL EVIDENCE FROM KOREA By Byung Han Seo A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Economics 1990 ABSTRACT MONEY DEMAND RELATIONSHIP AND MONETARY TARGETING IN A CHANGING FINANCIAL ENVIRONMENT: THEORETICAL CONJECTURE AND EMPIRICAL EVIDENCE FROM KOREA BY Byung Han Seo This paper investigates the implications of financial innovation and deregulation for the money demand relationship and monetary targeting policy. The major question is whether the public's demand for some measure of monetary aggregates is a relatively stable function of real income, interest rates, and prices, so that the monetary aggregates can continue to be desirable intermediate targets in the conduct of monetary policy. To address this issue, the theoretically possible effects of financial changes on money demand stability are examined and empirical tests for various hypotheses are conducted by using the Korean experience during past two decades. As the traditional distinction between monetary and financial assets becomes increasingly blurred, the central issue on the stability of money demand relationship is reduced to the question of whether transactions and asset demand balances can be effectively separated in the face of financial innovation. Such a separation should continue to exist because of the existence of extra costs inherent in the bank's asset transformation process and of the public's willingness to hold transactions assets of relatively low yields and fixed nominal value available on demand in the absence of perfect synchronization of receipts and expenditures. Then the demand for transactions balances is a stable function of income and prices. Empirical evidence presented.here shows that once the interest-bearing transactions (or demand) deposits are added to the conventional money, the M13 demand function in Korea, with one notable exception of interest elasticity, is quite stable throughout a changing financial environment. In the financially repressed economy, it is not surprising that instability of the interest rate coefficient is responsible for observed marginal instability of the money demand function. While significant financial innovations occurred during last decade and had great impacts on M1 as well as M2, there appears to be little evidence that those innovations caused the underlying relationship of M18 demand to change fundamentally. Therefore, it can be concluded that most of the prevalent assertions about money demand behavior and monetary targeting policy in association with financial changes are not supported by the Korean experience. To my Grandparents, the late, and Parents iv ACKNOWLEDGEMENTS I owe a special thanks to my dissertation committee Chairman and a devoted monetarist scholar, Professor Robert Rasche, for his patient readings of various drafts of this paper. Without his valuable comments and time-consuming guidance, it was impossible to make clear what this dissertation discusses. Furthermore, I benefited from helpful comments by Professor Mark Ladenson and Professor Paul Strassmann. In addition, I am indebted to Professor Peter Schmidt for numerous econometric consulting of the empirical tests. In spite of their excellent guidance, any remaining faults in the dissertation are mine. Finally, the everlasting support of my parents, elder brother and younger sisters, and the love of my wife, Insun, and daughter, Anne, made these six years studying in the foreign country much easier. I dedicate this dissertation to all of them as a small token of my gratitude. TABLE OF CONTENTS Page LIST OF TABLES ........................................................ viii LIST OF FIGURES ......................................................... IX Chapter I. INTRODUCTION ...................................................... 1 II. BACKGROUND AND CAUSES OF FINANCIAL INNOVATION ..................... 4 2.1. The Search for a Hypothesis 2.1.1. Hypothesis of a Complement to Real Change 2.1.2. Hypothesis of Constraint-Induced Innovation 2.1.3. Hypothesis of Circumventive Innovation 2.1.4. Hypothesis of Transactions Cost 2.2. The Experience of Financial Innovations in Korea 2.2.1. Overview of the Financial Structure and Regulation 2.2.2. Major Financial Changes in the 19703-19803 2.2.3. A Simple Credit Market Model III. THEORETICAL HYPOTHESES OF FINANCIAL INNOVATIONS ON THE MONEY DEMAND IV. RELATIONSHIP ..................................................... 39 3.1. Theoretical Effects of Interest Rate Deregulation 3.2. Theoretical Effects of Financial Innovation 3.3. Theoretcial Effects of a Change in Monetary Policy Regime FINANCIAL INNOVATIONS AND MONETARY TARGETING ..................... 65 4.1. A Framework for Analyzing Financial Innovations and Monetary Targeting 4.2. Adjustment of the Target Monetary Aggregate 4.2.1. Base Drift and Target Growth Rate Change 4.2.2. Redefinition of the Monetary Aggregate 4.3. Overview of Alternative Intermediate Targets vi 4.4. Changing Transmission Mechanism of Monetary Policy V. EMPIRICAL EVIDENCE ON THE STABILITY OF THE SHORT-RUN MONEY DEMAND FUNCTION IN KOREA ................................................ 95 5.1. Overview of Specification of the Short-Run Money Demand Dynamics 5.2. Empirical Estimates of the Short-Run Money Demand Function in Korea . Stability Test of the Short-Run Money Demand Function . Cusum-Squares Test . Chow and LR Test . Recursive Regression and Stimulation 3 5. 5. 5 5 . Dummy Variables Test WUUW kwNH 5.4. Evaluation of Alternative Hypotheses VI. SUMMARY AND CONCLUSIONS ......................................... 168 APPENDICIES ........................................................... 177 BIBLIOGRAPHY .......................................................... 194 vii Table LIST OF TABLES Page Financial Structure of Korea 1973-1988 .............................. 28 Likelihood Ratio Test Statistics ................................... 113 Quarterly MlB Demand Functions(l973/I-1989/11) ..................... 116 Double-Log Specification of Quarterly MlB .......................... 122 Demand Function(l973/I-1989/II) Cusum-Squares Test Results ......................................... 131 Chow and LR Test Results ........................................... 139 Recursive Regression and Simulation Results(Level) ................. 141 Recursive Regression and Simulation Results(First-Difference) ...... 142 Stability of Error Structure ....................................... 144 Post-Sample Static Forecast Errors of M13 Demand ................... 148 Dufour Test Results ................................................ 153 Interactions Test Results .......................................... 157 viii LIST OF FIGURES Figure Page 2 Process of Integration of Curb Market into Official Market .......... 33 4 Adjustment of Base and Growth Rate of Money Supply .................. 76 5.1 M13 Forward Cusum-Squares .......................................... 133 5.2 Ml Forward Cusum-Squares ........................................... 133 5.3 M2 Forward Cusum-Squares ........................................... 133 5.4 Quandt's Log-Likelihood Ratio(Forward) ............................. 136 5.5 Residuals of M18 First-Difference Regression ....................... 137 5.6 Residuals of M1 First-Difference Regression ........................ 137 5.7 Residuals of M2 First-Difference Regression ........................ 137 ix CHAPTER I INTRODUCTION Since the late 1970s, many countries, including Korea, have set and attempted to control the annual growth rates of some measure of monetary aggregates consistent with a desirable growth rate of output and prices. In conducting the monetary targeting policy, the aggregate money (particularly transactions balances; M1) demand is implicitly assumed to be a relatively stable function of real income, interest rates, and prices. Ironically, the period during which this strategy for monetary policy was implemented coincides with a breakdown of the standard money demand function. Simultaneously those countries have experienced numerous financial changes. It is alleged that financial innovation and deregulation can cause the money demand relationship to become so unstable or unpredictable that the achievement of a desired nominal income growth through control of the growth of monetary aggregates will be unsuccessful. This study examines the implications for the money demand relationship and monetary policy of financial deregulation and innovation. After addressing purely theoretical conjectures, the study conducts empirical tests for the hypotheses using the experience of Korea during the past two decades. The dissertation is organized as follows. In chapter II the general background and causes of financial innovation are briefly reviewed from several specific and individual 2 propositions. Then the experience of financial innovation and deregulation in Kbrea over the last two decades is outlined, and the underlying process of major financial changes is analyzed in a simple credit market framework. In chapter III the theoretical hypotheses of financial innovation on the money demand relationships are exhaustively investigated. A simple inventory-theoretic model of the transactions demand.for'money is employed to examine the likely effects of deposit rate deregulation. A more general asset demand model is used for the analysis of the potential effects of reductions in transactions costs and of a proliferation of new money substitutes. In addition, a money demand model that takes account of Friedman's long-standing argument is developed to show how rational agents react to a change in monetary policy regime when determining their demand for real balances. The analysis corresponds to Lucas'(l976) critique in the context of the short-run money demand dynamics. In chapter IV the implications of'a changing financial environment for monetary policy are addressed. The chapter begins with an examination of how financial changes can generate the difficulties for policymakers who are attempting to control the growth of money stock. The discussion considers how short-run.monetary stabilization policy should be conducted if some measure of monetary aggregates is used continually for a preferred intermediate target variable. The third section of this chapter enumerates alternative candidates for an intermediate target variable to a transactions monetary aggregate or M1, and evaluates the eligibility of each alternative as a good target variable. In chapter V empirical tests are performed to examine whether the 3 money demand function in Korea is stable throughout a period in.which the financial environment is rapidly changing. First, some specification problems inherent in a large portion of the recent money demand literature are discussed. Then we specify an appropriate quarterly money demand function in Korea for the period 1973/I-1989/II, taking some special institutional characteristics into consideration. Once the basic model is specified and estimated, we conduct a set of stability tests for the estimated money demand relationship, and attempt to identify the extent, nature, and causes of observed statistical instabilities. Finally, the theoretical hypotheses and/or the conventional wisdom regarding the money demand relationship and monetary targeting policy in the face of financial changes are assessed based on the empirical results. The empirical evidence presented.here suggests that, with one notable exception, the money demand function in Korea is remarkably robust in the face of financial innovation and deregulation over the past two decades. The exception is the tendency for the interest sensitivity of money demand to gradually decrease. This instability is largely due to the distorted and turbulent credit market conditions for which pervasive financial repression is ultimately responsible, although it is partly attributable to the recent financial innovation and deregulation. This finding provides empirical evidence that most of the prevalent assertions about money demand. behavior and. monetary targeting 'policy in a changing financial environment are not supported by the Korean experience. In chapter VI main theoretical arguments, major empirical findings, tentative conclusions, and some avenues to further research are summarized. CHAPTER I I BACKGROUND AND CAUSES OF FINANCIAL INNOVATION 2.1. The Search for a Hypothesis Financial innovation (FI) refers to several phenomena. It includes new financial instruments which are the objects of financial transactions, new financial markets which are the fields for trading in financial assets, and. new financial practices to carry out the transfers of financial assets. F1 is neither a new phenomenon nor just an ephemeral phenomenon. However, until lately little attention has been paid to F1. The predominant catalyst of the current considerable interest in F1 is the difficulties experienced in the conduct of monetary policy. F1 is frequently alleged as an explanation of the empirical monetary puzzles. Since the mid-19708, the ”standard” money demand function has exhibited instabilities in several advanced countries. The natural question of why such instabilities occurred has been raised, and F1 is among the plausible potential explanations. As a result, there is a great need for sound knowledge about the various aspects of F1. However, we lack serious understanding of overall FI, of factors which induce it, processes which diffuse it, and effects which follow it. Given our existing imperfect knowledge, it is very difficult to fully understand the processes inducing innovation and to deal adequately with F1 in the conduct of monetary policy. Although 5 particular attention has been given to the background and causes of F1 and to the broader characteristics of the innovation process, there is no generally acceptable framework for dealing*with the subject” The existing hypotheses briefly reviewed below are not general, but rather specific propositions that are only appropriate to some individual innovations. Few attempts have been made to integrate these individual hypotheses systematically within macroeconomic theory. 2.1.1. Hypothesis of a Complement to Real Change As seen later, there is much evidence to suggest that F1 is a "demand- induced", in contrast to an. autonomous or external events—induced, activity. That is, FI arises from some financial adversity over (or a decrease in) profit opportunities and is calculated to overcome the constraints imposed on the financial sector. But it is doubtful whether FI would occur without an underlying demand from the real sector for financial services, such as monetization and payments, intermediation, and asset transformation. In this respect, the Schumpeterian school of economic thought which stresses a constant interaction (or interdependence) between real and financial sectors is more suited to the study of F1 than the mainstream monetary theory'which tends to dichotomize real andumonetary relationships particularly in the context of the long-run neutrality of money. According to the Schumpeterian approach, financial innovations respond to a stimulus in the real sector and in turn influence the potential path of real economic activity (Minsky, 1957; Silber, 1975). More recently, Hart claims that ”in a large part, financial innovations are an adaptation to 6 changes in technology and management outside the financial field."(1982, p. 90) The hypothesis is consistent with some observations of a secular trend of the parallelism in the long-run.development of real and financial structures. Schumpeter noted that over business cycles, "expansion initiated by technical innovations is financed not by saving, but by credit creation considered to be the monetary complement of innovation."(l964, p. 85) In this light, money and financial structures adapt to the "needs of trade" and thus essentially respond to the requirements of the real sector. The complement hypothesis or financial view of money therefore denies the proposition that monetary sector can exercise an independent influence on economic processes. But the connection of credit creation with real economy can, be easily broken down when capricious and unreliable expectations induce purely financial and speculative expansions. This aspect leads Minsky(l969, 1982) to the hypothesis of ”financial (or credit) instability". The hypothesis can be criticized in that it does not explicitly consider FI as an integral part of general economic evolution. The emphasis on the innovative entrepreneur contrasts with the conservative and adaptive banker. The innovator in the realm of'money and finance does not appear to be any less ingenious in seeking profit or self-interest than.a typical Schumpeterian.entrepreneur; The view of passive adaptation of financing thus cannot properly explain FI arising from an environment of deregulation and fierce competition. However, it is reasonable to assume that the expansions of credit and money, together with institutional changes in.the banking sector, are related to expansions in 7 the real sector brought about technological innovation. The hypothesis is particularly suitable for the secular economic growth patterns that the long-run financial and real developments in general exhibit considerable similarities and a strong interdependence. An interesting feature of the F18 since the 19708 is that they have occurred in generally depressed economic conditions, but in the environment of structural and technological changes, inflation and deregulation. 2.1.2. Hypothesis of Constraint-Induced Innovation A general and microeconomic theory of F1 has been provided by Silber(l975, 1983). He accepts that financial firms seek to maximize utility (profits) subject to some given constraints--balance sheet constraint, government regulation, and self-imposed or market-imposed constraints, and claims that "innovation of financial instruments and practices occurs in an effort to remove or lessen the financial constraints imposed on firms."(l975, p. 64) When an exogenous change in the constrained optimization of the financial firms occurs, the firms are induced to undertake the search costs of innovations so they can remove or circumvent the constraints by altering the opportunity set they face.1 As the main. externally imposed. constraints, Silber(1983) refers to inflation, interest rate variability, internationalization, technological innovation, and government regulations. 1 Silber distinguishes two types of such changes; first, exogenous changes in constraints force a decrease in the utility (profits) of the firm, so the firm innovates to return to its previous level of utility and second, increases in the cost (shadow price) of adhering to the existing constraints also stimulate financial innovation (1975, p. 66). 8 Whether a particular F1 is actually initiated depends upon the relationship between the cost that customers of financial services have to bear because of the existence of financial constraints and the cost to competing financial institutions of providing innovations. If the former exceeds the latter, there always exist profit opportunities to be exploited by introducing new financial instruments or practices. The condition and timing of innovation has been formally analyzed in a linear programming (LP) framework. For a financial firm's portfolio allocation, the LP model of the firm's objective function. (profit maximization) and constraints can be formulated. By using historical data of the firm's asset selection, the LP model can be solved period by period. The "shadow prices" (or dual values) of the constraints are derived as a by-product of the optimization problem, and can be used to measure profit opportunities and their changes over time. Rising shadow prices imply an increase in the cost of adhering to the existing constraints and in turn increased profit opportunities that could be exploited by altering the constraints faced by the firm. The course of a reaction to profit opportunities then leads to new financial instruments and practices. Such a profit opportunity approach can predict the likely timing, of innovations or an. increase in the rate of' diffusion. of innovations by financial firms.2 There is a criticism that paradoxically Silber's "general theory" of F1 is "too general” and "too specific".(Podolski, 1986, p. 186) It is too 2 Ben-Horim and Silber(1977) formally applied the LP model to the study of innovations by large commercial banks, and found some evidence that the shadow prices of constraints tend to rise prior to introduction of a new financial instrument and drop immediately thereafter. 9 general in that financial firms innovate to maximize profits; the stress is mainly on "adversity innovation? when an externally imposed constraint such as government regulation needs to be avoided only for maximizing profit. On the other hand, it is too specific in that the hypothesis applies to innovation of financial instruments and practices by the "existing" firms and thus may not be suited to dealing with the emergence of new markets, new firms, or new monetary standards. Nevertheless, the analysis of innovation of financial instruments and practices by financial firms provides an important aspect of our overall understanding of financial innovationary processes. By applying a LP model of the portfolio behavior of a financial firm to the study of innovation, the microeconomic approach contributes useful advances in the neglected yet important subject of monetary economics. 2.1.3. Hypothesis of Circumventive Innovation On many occasions, FI constitutes a response to monetary regulations, which ultimately tends to make obsolete the existing statutory framework. It is in the nature of many forms of monetary and regulatory controls that they represent at once an obstacle to the pursuit of existing profitable activity and on the other hand an opportunity to profit from the discovery of a way to pursue a closely related activity lying just outside the boundary of such devices. FI thus serves as a device to circumvent various regulatory controls. As a typical model of this hypothesis, Kane(l981) has developed a framework of "regulatory dialectic" in which the "invisible” hand of the market interacts with the "visible" hand of the regulatory authorities by 10 incessantly adapting to each other; ”Market institutions and politically imposed constraints reshape themselves in a Hegelian manner, simultaneously resolving and renewing an endless series of conflicts beWeen economic and political power. The approach envisions repeating stages of regulatory avoidance (or "lOOphole mining") , and re- regulation, with stationary equilibrium virtually impossible."(p. 355) Indeed, the capacity and flexibility to adapt to changing conditions, including a deliberate avoidance of regulations and policy imposed by monetary authorities, have long been observed in the world of money and finance. For example, Gould, et. al.(1981) note that the whole history of money is the continual invention of new kinds of money in the face of greater inconvenience of the existing money or a shortage of an official medium of exchange. It is also well-known that usury laws or interest rate ceilings were effectively circumvented by commission payments over and above allowable interest rates or by various compensatory arrangements. Moreover, rational expectations school economists (Lucas, 1976; Kydland and Prescott, 1977; Barro and Gordon, 1983) and Goodhart (1981, 1984) emphasize, as discussed later, that any change in policy will bring about a policy-induced change in the agent's optimal behavior. The notion that innovation is induced by monetary actions needs to be qualified by considering the underlying conditions and the environment in which circumventory processes would develop. Circumventive innovations occur normally when a combination of forces is at work. The forces include high and variable interest rates, changes in the regulatory regime, ambiguities in regulations, and technological progress in information processing . In one aspect, Hester notes that "financial institutions innovate ll whenever customer relationships are jeopardized by slow monetary growth."(l981, p. 183) According to his argument, such.a condition occurs predominantly during restrictive monetary policies based either on market actions manifested in high interest rates or restrictive regulatory measures. On the other hand, Wojnilower identifies "credit bottlenecks“ as the most important condition for the Circumventive innovations to occur. He denies that the demand for credit is interest-elastic, and argues that "The growth. of' credit: is therefore essentially’ supply- determined.'(1980, p. 277) Once the credit market (particularly credit supply side) is interrupted for some reasons (e.g., regulatory obstacles such as interest rate ceilings or credit controls and a loss of confidence consequent upon the failure of a major institution or market), both authorities and.private markets deliberately undertake measures which are designed to prevent a future interruption in the flow of credit. Thus FI triggered by regulatory inadequacies can be seen as a product of the restricted credit availability, rather than the high cost of credit. This view contrasts with the proposition that high and variable interest rates are an essential part of Circumventive innovations. It is not clear whether a creative monetary response is more likely to be induced by quantity controls (credit crunches) or price constraints (high interest rates). But both conditions often occur simultaneously, particularly at a time of restrictive monetary action. In sum, there is considerable evidence of interaction between monetary control and F1. Monetary regulations are highlighted as key factors inducing avoidance behavior, yet they seem to be only one of the necessary conditions for innovation to take place. From the evidence available so 12 far, we cannot deduce that creative financial response is inevitably and exclusively a reaction to monetary restriction. Innovation continues to occur and to affect monetary control even in the countries relatively free from continuous monetary restrictions. Although regulation-induced innovations are of a special interest to policymakers, we do not understand the nature of Circumventive innovation sufficiently enough to be able to construct arrangements in the policy design anticipating such creative reactions. Little is known about the conditions necessary for its "take-off”, the rate of diffusion, and the factors governing diffusion . 2.1.4. Hypothesis of Transactions Cost This hypothesis focuses on the importance of transactions costs in the demand for money.3 According to the proposition, the reduction of transactions costs is the dominant factor in F1 and F1 is essentially a response to the potential for cost reduction offered primarily by technological advances . Hicks(l967) considered the rationale for holding money in terms of lowering transactions costs. He envisages the development of ever more sophisticated ways of reducing transactions costs as one way of looking at monetary evolution. The generation of different media serving as money, the development of new, better organized markets, the genesis of such devices as the check and clearing system, and the evolution of liquidity management (Hicks, 1967, pp. 31-6) are all linked by the drive 3 The role and meaning of transactions costs in the money demand function are discussed in chapter III. 13 to lower transactions costs. More recently, Niehans has emphasized that "the primary (though not the only) motor of financial innovations is the gradual decline in transactions costs."(1982, p. 27) Thus financial evolution is interpreted as a reflection of technological progress which reduces the costs, though it is admitted that we do not have a full analytical understanding of this process. Indeed, a view is expressed that the only true change in financial industry is "technological innovation" which reduces transactions costs (of storage, retrieval, and transmission of information); all other changes are merely "adaptive innovations" which consist of new ways of "bundling” financial services that remain fundamentally the same in nature (Niehans, 1983, pp. 537-9). By reducing transactions costs, technological innovation has increased the sensitivity of holders of liquid assets to changes in the financial environment, such as changing interest rate differentials or alterations in. monetary regulation or control. Thus it has enhanced the Circumventive power of financial institutions and. of corporations, and. contributes towards financial markets becoming more contestable and more competitive through easier entry and exit. In conclusion, technical innovations and the accompanying reduced transactions costs are a common denominator of all major F1. The phenomenon may have had strong impact on the demand for narrowly defined money. But the explanation of PI exclusively in terms of a response to declining transactions costs, whether induced by technical change or by competition, seems to be an over-simplification of the complexity of F1. Before leaving for the next section, some concluding remarks to the 14 search for a hypothesis of F1 are in order. As stated in the review of the existing literature, the overwhelming impression in our attempt to examine the innovation-inducing factors is that a combination of circumstances seems necessary for F1 to occur. It is unlikely that a simple hypothesis will suffice to explain innovation-generating influences and.processes and to predict how the financial system reacts to a changing socio-economic environment. F1 is essentially a manifestation of interdependent influences, and its process seems complex and as yet not fully understood. FUrthermore, the processes underlying macroeconomic changes are essentially microeconomic. Yet the microeconomics of F1 is in its infancy and in general confined to analyses of factors or conditions leading to innovations in financial instruments, markets, and institutions. The process of innovation by financial institutions and the nature of diffusion of innovations are almost unresearched. Certainly, the subject deserves more systematic and statistical investigation. 15 2.2. The Experience of Financial Innovation in Korea The governments of most developing countries intervene extensively in the operation of their financial system. In a system where the financial markets are less developed. and. highly fragmented, resulting in an inefficient flow of funds, government intervention is often justified on the assumption that it has a better chance of success in channelling credit for the desired allocation of resources. Financially "repressed" and/or underdeveloped economies in general are characterized by interest rate regulations, direct domestic credit controls, international capital controls, and so forth. In Korea it is often asserted that such highly government-controlled monetary and financial policies, particularly in the transitory stage, helped insure the success of the government's economic development plans.‘ During the 19605 and the early 19705, direct government controls over interest rates and credit allocation allowed the continued increases in investment and rapid economic growth.5 In the mid-19703 as the economy became more sophisticated and more dependent on the rest of the world, this system of heavy controls began to produce the harmful side-effects which could damage the economy's growth potential. A chronic high inflation rate resulting from the process of rapid.development, substantial variations in the world interest ‘ For an overall description of financial development in Korea 1945- 1978, see Cole and Park(1983). 5 Korea's annual growth rate measured by real GNP has averaged over 9% in the 25 years since the First Five Year Economic Development Planuwas introduced in 1962. As for other statistics relevant to the discussion in this section, see Table 2 unless otherwise specified. l6 rates and economic activity, and a huge amount of external debt inevitably called for a change to less regulated financial systems. As in many developed countries, the move toward liberalization has focused on reducing the government's role in directing resource allocation and on permitting market forces to overcome domestic distortions and inefficiencies . 5 Over the past two decades, financial reforms undertaken by the government have included relaxation of interest rate restrictions, reduction in the use of direct credit rationing or in the provision of special credit at preferential rates, and development of open markets in the primary securities. Although the government has taken various steps for gradual liberalization, it still maintains relatively restrictive controls (implicit or explicit) over almost all the fields of the financial system. This section provides an overview of Korean experiences of F1. The structural characteristics of Korean financial markets are briefly reviewed, and then major financial changes during the past two decades are outlined. Finally, a simple financial sector model is presented to analyze the background and causes that are considered most significant in the process of F1 outlined. Since our principal interest in F1 is in its impact on the money demand relationship and the measurement of monetary aggregates, and hence on the monetary targeting policy, our discussion is confined primarily to those innovations which have a direct effect on the 5 The "neo-liberal" school of development economics stresses the need for "financial liberalization" as an important factor of growth strategy. For example, McKinnon(l973) and Shaw(l973) argue that financial restrictions have hindered industrialization of most developing countries. 17 monetary aggregates, putting aside innovations whose effect might be indirect or negligible. 2.2.1. Overview of the Financial Structure and Regulation For the investigation of the course of F1, it is necessary to understand the underlying financial structure and general policy attitude. One of the most important financial characteristics of Korea has been the ”financial dualism”. Its origin may be traced in part to the dualistic economic structurencoexistence of modern industrial sector and traditional agriculture and service sectors. The financial dualism refers to the financial system in which modernized (regulated) financial institutions exist together with traditional (unregulated) financial markets such as the completely free curb market.7 The unregulated financial (curb) market is a market for short-term primary securities which are close substitutes for the short-term indirect securities issued by regulated financial institutions. As in the regulated money and capital markets, the ultimate borrowers and lenders are firms and households. But the two markets are fundamentally different on the ground 7 Referring to Korea's financial system, the terms "organized vs unorganized”, ”official vs unofficial", ”formal vs informal”, and "regulated vs unregulated" markets are used interchangeably. The concept of "unorganized money market" was first introduced by Wai(l957) to describe financial markets in which institutional intermediaries such as commercial banks, are not involved in the exchange of either existing or new financial claims. However, Cole and Park claim that the last terminology describes best the curb and "KYE" (a form of rotating credit association) markets which are quite organized, recognized by the government as an important part of the country's financial system, and have been very efficient in allocating credit. For more detailed characteristics of the various unregulated financial markets, see Cole and Park(1983). 18 that interest rates in the curb market are market—determined, while deposit and lending rates in the regulated institutions are government- mandated far below the free market level. This feature suggests that the curb market rate is likely to be sensitive to the stance of monetary policy so the rate fluctuates over a wide range in the short run. However, the curb market rate, as discussed later, cannot provide valuable information for an indicator of monetary policy. Neither the volume of the transactions nor the interest rate is readily available because the transactions are carried out illegally in the underground economy.a More importantly, the interest rate largely reflects a default-risk premium rather than the direct influence of monetary policy. Although the government has made great efforts to suppress the curb market,9 the unregulated market has expanded throughout the years meeting the financial needs of large or small borrowers and lenders both in the modern and traditional sectors. The financial repression is primarily responsible for such continual existence and expansion of the curb market, in spite of rapid growth of the economy and.modernization of the economic structure. The regulated institutions have been so closely controlled that they have not been able to adapt to a changing economic environment. ° The survey of Bank of Korea (BOK) asks sample firms to quote the interest rates bearing on their curb market loans, regardless of the very heterogenous terms, sizes and.collateral requirements of those loans. The quoted interest rates are then weighted by the simple number of loans of different rates to get an average rate. °.Amonggmany reasons to eradicate the curb market, the main objection to its existence is that it is ”unregulated” or out of the monetary authorities' control and thereby it may disturb policy interventions. l9 Naturally there exists an important role for the unregulated institutions. Specifically, the low interest rate policy” for the banking sector has generated a chronic excess demand for the bank credit while discouraging the public from holding savings deposits in banks. Under such a circumstance, the monetary authorities have had to exercise stringent credit rationing. As might be expected, the priority for credit distribution has been given to such strategic sectors for economic development as exporters and larger producers with very little credit allocated to others. Those who have not been accommodated by the banks have had to raise funds outside the regulated institutions, mostly in the curb market. Therefore, the low interest rate policy can be regarded as the most direct cause of continued expansion of the curb market and persistence of the financial dualism. The real interest rate on. deposits and loans in the regulated institutions has, on average, been close to zero or even negative because of downward adjustment of regulated nominal interest rates and high rate of inflation. The growth of regulated sector has slowed accordingly and much of domestic savings has shifted into the curb market or into real assets, such as real estate, houses, and.consumer durables. Nevertheless, the monetary authorities have been reluctant to normalize interest rate 1° As the rationale for the low interest rate policy, there was a presumption that low interest rates (cost of capital) are necessary to stimulate the high level of investment required for a target rate of growth. The possibility that such a policy may discourage savings and thereby encourage inflation was less emphasized. Paradoxically, Korea was one of the countries that demonstrated the potential effect of interest rate on the demand for bank savings and time deposits during the high official interest rate period 1965-71. For more details, see Cole and Park(1983) especially chapter V. 20 for the fear that the resulting high nominal interest rates would induce cost-push inflation and undermine the competitiveness of Korean exports in the world market. 2.2.2. Major Financial Changes in the 19708-19808 In the early 19708,11 concerns over high inflationary pressures and lack of direct control over credit flow through the curb market led the government to authorize the formal establishment of short-term investment and finance companies (STFCs)3l2 which stand between commercial banks and unregulated curb markets. In order to assure competitiveness with the curb market, the government allowed these non-bank financial institutions to be less regulated with respect to their asset and liability management and interest rates than commercial banks. These intermediate groups of non-bank financial institutions are thus permitted to set their interest rates between unregulated curb market rate and regulated bank rate in an attempt to shift the underground finance activity into the official finance market. However, the growth of STFCs was limited during the early period 11 In August 1972, the government announced Presidential Emergency Decree for Economic Stability and Growth. The decree attempted to suppress the curb market that has served useful function in the economy and to relieve the financial stress on large borrowers in that market; but it also lowered bank rates by a large amount (e.g., time deposit rate from 17.4% to 12.6%). The failure to eradicate the curb market provided some insight into its role and the limit to government regulatory powers over the free market. 12 Their main business is to deal in commercial bills which they issue, but they also buy and sell, accept and guarantee the paper of corporations. The government thus expected STFCs to be effective substitute for the unregulated short-term lending institutions. 21 mainly because their interest rates were still controlled far below the rates offered in the curb market. The emergence of STFCs, nevertheless, was the harbinger of all the FI that occurred in the following years. In the middle of 19708, the inter-bank "call money" market (similar to the U.S. federal funds market), CD8, and RPs were introduced as the first step toward gradual interest rate liberalization. CD8 offered by commercial banks provided an outlet whereby large depositors can receive relatively higher interest rates than official deposit rates even within the banking system. RPs offered by security companies also allowed short-term investment funds to earn yields more closely related to the free market rate. Both innovations have added some degree of flexibility to the financial markets without any formal lifting of the existing regulations. Another dramatic innovation was the introduction of "overall- households-deposits" (similar to the U.S. NOW account) on which personal checks can be written with a relatively high interest rate earned.13 The account has contributed to enhancing the public's access to banks and thus to reducing its preference for currency. A8 inflation subsided in the early 19808, the monetary authorities pegged interest rates of regulated institutions at very low levels. Thus the disintermediation phenomenon at commercial banks became severe. In 13 ”Savings-deposits" was introduced at the same time. Except for a small difference in the maximum limit of deposits, both accounts are similar in that they can be withdrawn on demand without any penalty. What matters in categorizing deposits into narrow and broad money, particularly in the society where currency is more widely used than checks, should be the availability on demand rather than the checkability of deposits. Nevertheless, the former is included in M1 while the latter is classified into non-M1 component of M2. For policy purposes, however, it does not make any difference because the monetary authorities have targeted on M2. 22 an effort to overcome the stringent credit availability constraint, corporations and financial institutions at last innovated the unofficial ”WANMAE“ (similar to the U.S. RPs) marketi‘.a8 a means of circumventing regulated interest rates. It played an active complementary role for regulated financial markets from its introduction late in 1982 until the authorities outlawed the transactions in November 1984.15 In spite of government regulation, the WANMAE transactions resulted from the workings of a free market within the institutionalized arrangements and spread rapidly from the very beginning. The new informal market forced the authorities to gradually relax interest rate restrictions. Therefore, it can be claimed that the WANMAE transactions played a leading role for the FI that followed in the 19808. During 1984-1986, the interest rates on ”call-money" market, negotiable CD8, CPs, and government bills were deregulated. Finally, all the interest rates except for some bank deposit rates were formally 1‘ The most direct background of that innovation was the so called "financial scandal” of May 1982, which contracted considerably short-term financial markets, especially the curb market. WANMAE was different from the already-institutionalized WHANMAE in that the latter was used to raise funds for security companies themselves while the former was used primarily for providing corporations with short-term funds as an instrument of intermediation with various bonds collateral. 15 The official reason for outlawing WANMAE was that it weakened the official institutions' financial intermediation and disturbed secondary bond markets. But the actual reason was probably the fact that the existence of a financial instrument earning returns higher than the market yields of corporate bonds became a direct barrier to the banks' competitiveness. By prohibiting the transactions, the authorities expected that the partial deregulation of bank rates would enhance the banks' competitive position. 23 deregulated in December 1988.15 However, this does not mean that the formal deregulation goes into effect directly. So long as the authorities still desire to keep implicit or explicit controls over various interest rates and credit allocation and so long as the bank officials are accustomed to the atmosphere of pervasive government direction with little ability to cope with a free market system, we anticipate that a transition to the free market rate system will take a considerable period of time. STFCs were allowed to deal with "securities investment trust“ and “CMA” (similar to the U.S. MMMFs) in 1981 and 1984 respectively. With the introduction of these instruments, by pooling small savings STFCs can invest the funds raised primarily in the short-term financial instruments, such as CD8, CPs, and government or corporate bonds, and thereby pay the savers a share of operating profits. Although these accounts are not directly checkable, there are no limits on withdrawals as long as required minimum balances are met. Corporations and households can put their temporary idle or savings balances into these attractive accounts and earn relatively high yields even for short time periods. These new money market funds attracted a rapid and sizable flow of funds primarily out of interest-bearing bank deposits and curb market funds. Given the problem of high risk in the curb market and of low deposit and borrowing rates at banks, it is not surprising that STFCs have been rapidly gaining market share at the expense of banks and also attracting some savings from the curb market. In a parallel with such changes, comercial banks were authorized to introduce "free-savings-deposits" 15 For detailed statements of interest rate liberalization, see BOK's Weekly Demeeeie end International Eeenom, No. 1400, (Dec. 1988). 24 (similar to the U.S. NOWs) for households in April 1985. Funds in this account pay a relatively high interest rate and are withdrawable on demand, so to some extent commercial banks can compete with STFCs. Another innovation that deserves mention is the improvement of cash management techniques through the rapid spread of telecommunications technology and electronic funds transfers. Since the mid-19708, on-line banking system, payment of public utility charges through GIRO, wide use of credit cards, automatic teller machines, and automatic depositing of monthly salaries and automatic overdrafts by overall-households~deposits have been introduced. In principle, all these developments are expected to decrease the demand for conventional money as well as to alter interest rate sensitivity. Improved information flows and forecasting procedures have reduced uncertainty about cash flows and lowered precautionary balances; new money market instruments (CD8, RPs, CMA, and free-savings- deposits) provide new profitable outlets for liquid funds; new electronic devices enable switches of funds into high interest rate assets to be made quickly at lower transfer costs. However, the public may place little weight on such developments because there always exists an excess demand for the bank credit under the financial dualism or the bank loans rate far below the curb market rate. Although there is a hypothesis that high nominal interest rates are a primary factor of innovations in cash management (Porter et al, 1979), there is no indication in Korea to suggest that episodes of falling rates in the 19808 discouraged the technological innovations in.cash management. Apparently cash management is a more complex phenomenon than a simple reaction to high interest rates. While interest rates may be important, 25 cash management innovations are also related to other factors, such as transactions costs, profitability, and other innovations in the financial system. It should be stressed that the developments of cash management are an integral part of interrelated innovationary processes involving the development of money markets, new financial instruments, and new technology. Finally, it should be pointed out that monetary control regime has changed. The monetary authorities, at one time or another, have used all the usual instruments of monetary control, but the primary instruments have'been direct control of interest rates and credit allocationar7 Early on such direct controls were considered necessary in the attempt to effectively mobilize capital for the development programs. In recent years, however, the underdeveloped. money and capital markets have precluded an active use of openimarket operations. On the other hand, the low interest rate policy to stimulate investment has brought about a chronic excess demand for the bank credit making discount rate policy ineffective as well. Nevertheless, the above mentioned innovations have created new concerns for the authorities who are used to relying on direct control of interest rates and credit allocation in conducting monetary policy. The growth of close substitutes for the bank credit through STFCs and the consequent decline in the market share of commercial banks have weakened the effectiveness of credit rationing. The direct control of bank loans 13.As recently as the early 19808, the government allocated 50% to 70% of the domestic credit, directly or indirectly, depending on the classification of "policy loans". Furthermore, the deposit money banks exercised very little control over the remaining part of the credit. 26 became less effective in limiting credit expansions when regulated banks ceased to be the only source of large amounts of funds and STFCs provided a source of credit outside direct control of the monetary authorities. As the authorities permitted some degree of flexibility in the determination of interest rates as well as more freedom in the financial transactions, the emphasis has shifted to controlling monetary aggregates. This is largely the result of increasing desire to use monetary policy to achieve price stability rather than to allocate credit in the economy. The monetary authorities finally abandoned formal direct control over lending by individual banks in 1982 and ended preferential rates for specified borrowers in 1984.18 The new regime called "indirect" monetary control places more emphasis on control of monetary aggregates by the use of formal instruments of monetary policy rather than direct domestic credit controls. Since 1979, the authorities have set growth targets for M2 consistent with the desired target rates of income growth and inflation and also used Ml as a supplementary indicator. Both the instability of demand for M1 and the lack of control over broader monetary aggregates are the reasons mentioned by the BOK's research staffs for the choice of M2 as the principal intermediate target; "With structual changes (or recent financial changes, such as the raising,of interest rates on.some demand deposits relative to other callable deposits), the relationship between M1 and economic activity has become less stable, while that between M2 or M3 and economic activity has become more evident and reliable. On the other hand, because the flow of funds from non-bank financial institutions to private curb money markets 1‘ In reality the monetary authorities still considerably make use of credit allocation rules as an instrument of monetary policy. For example, BOK continues to have preferential discounting facilities to banks which lend to exporters and other designated industries and maintain credit allocation requirements for banks as well. 27 is also highly volatile, M3 and'broader credit aggregates have become much less easily controllable by the monetary authority."m However, the present classification of M1, M2, and M3 seems to be quite arbitrary since the accounts closely related to transactions balances are included in M2 or M3 but excluded from M1. The current debate focuses primarily on which monetary aggregate should be a target variable.2° 1° This argument can be easily found in various issues of the BOK's Menehly_fielleein, Special Studies, and also in Shin(l986) "The Money-GNP Relationship in Korea" (Paper presnted at the Seventh Pacific Basin Central Bank Conference on Economic Modelling, Sydney, Australia, Dec. 1986). 20InSpecial Studies (mimeo), the BOK's research staffs recently“have claimed that the monetary aggregates based on the degree of "turn-over" ratio of liquid financial assets as well as various bank deposits should be considered a target variable regardless of the institutions issuing those financial liabilities. 28 Table 2 Financial Structure of Korea 1973-88 deposit yields curb GNP deposit credit bank STFC of market share share rate rate coporate rate a b c d e f g h i j bond 1973 14.0 12.2 78.0 22.0 73.2 26.8 - 12.6 13.4 - 20.0. 33.3 1974 8.5 30.4 76.9 23.1 75.1 24.9 - 15.0 15.4 - 21.3 40.0 1975 6.8 24.6 77.9 22.1 73.2 26.8 - 15.0 16.4 - 20.3 41.3 1976 13.4 21.0 75.9 24.3 71.4 28.6 - 15.5 16.4 - 20.4 40.4 1977 10.7 15.9 74.8 25.2 65.8 34.2 - 16.7 16.0 - 20.0 38.1 1978 11.0 21.6 74.0 26.0 64.4 35.6 - 18.6 16.4 - 21.1 41.7 1979 7.0 20.0 71.9 28.1 63.1 36.9 - 22.9 16.4 - 26.7 42.4 1980 -4.8 25.3 68.1 31.9 60.6 39.4 - 18.5 20.4 - 28.8 44.9 1981 6.6 15.4 67.0 33.0 58.9 41.1 14.4‘11.0 18.3 39.7b23.6 35.3 1982 5.4 6.7 62.8 37.2 58.1 41.8 8.0 8.0 9.0 17.0 17.3 30.6 1883 11.9 3.9 59.1 40.9 57.0 43.0 8.0 10.0 8.5 13.0 14.2 25.8 1984 8.4 3.8 55.5 44.5 54.2 45.8 6.0 10.0 8.5 13.6 14.3 24.7 1985 5.4 4.1 52.6 47.4 54.3 45.7 6.0 10.0 8.5 13.4 13.9 24.0 1986 12.3 2.7 50.7 49.3 54.5 45.5 6.0 10.0 8.5 13.3 12.8 23.1 1987 12.0 3.7 47.9 52.1 51.9 48.1 6.0 10.0 8.0 12.9 12.8 23.0 1988 12.2p 4.5p 44.9 55.1 51.1 48.9 6.0 10.0 8.0 12.8 14.5 22.7 a real GNP growth b GNP deflator c depositoty banks d non-bank financial institutions e depository bank f non-bank financial institutions g household checking account h l-year time deposit 1 commercial bills(60-90 days) j CPs GNP: % change based on 1980 constant prices interest rates: % annual average per year ‘ effective from July 1981 b effective from 1981 and deregulated from October 1985 P preliminary estimate Sources: BOK, Eeonomie §taeistics Yearpeok and Monthly Bulitin, various issues BOK. W. 1984 29 2.2.3. A Simple Credit Market Model As seen in the previous sections, F1 in Korea can be viewed as a process of regulatory avoidance by which the free market has responded to the low interest rate policy and the direct credit control of regulated institutions. F1 was promoted further when authorities recognized an obsolescence of the existing regulations and relaxed them. This section develops a framework to analyze the process through which the curb market has been integrated into the institutionalized market--the structural change that is considered most significant in the process of F1 over the past two decades. We set up a simple financial sector model based on the financial dualism” The model specifies the supply of and the demand for credit both in regulated and unregulated financial markets. The regulated market consists of banks and STFCs, while the unregulated market represents the curb market. Although there exist some differences between banks and STFCs, with co-ercial banks more heavily regulated than STFCs, this classification simplifies the credit market model without altering the essential aspects of interaction of the financial dualism or the basic conclusion of our analysis. First, the regulated (or institutionalized) credit market is formulated as follows: (2.1) RC' -£,(I, k, BR, I, 11,) i - + + ? (2.2) RC‘ -fz(i, r, I, U2) - + + 7 (2.3) RC -RC' < R0“ The supply of loanable funds in the regulated market (RC‘) depends on 30 regulated interest rate (i) , reserve ratio (k) ,21 bank's borrowing from the central bank (BR),22 diversity of financial instruments (I), and other factors not explicitly specified (U1) such as expected inflation and conditions of general economic activity. The supply function can be derived from banks' and STFCs' balance sheets given by the identity of RC”-(l-k)D+BR, where D represents the public's demand for the liabilities of regulated institutions. The demand for credit in the regulated market is function of I, r(curb market rate), I, and U2. Finally, equation (2.3) indicates the existence of an excess demand for the credit and thus its actual (realized) quantity is supply-determined. Second, the unregulated (or free) curb market is represented.as below: (2.4) PC“ -f3 (I, r, R, I, 0,) ; + - - 7 (2.5) FC‘ -f,, (1, r, I, U.) (2.6) FC' -FC"- - - 7 The supply of the curb market loans (FC'), or the demand for the informal securities by wealth-owners, depends on I, r, R(degree of risk), I , and U3. On the other hand, the demand for the curb market credit (FC‘) is a function of I, r, I, and U,. In the model RC, FC, and r are endogenously determined, while i, BR, 1, R, and U1 are exogenously determined outside the system. Finally, signs in the equations represent a positive or 2‘ Since there exists a chronic excess demand for credit in banks and STFCs, excess reserves are almost close to zero. Very often, commercial banks have suffered a shortage of reserve requirements, which called for a ”special emergency credit" from the lender of last resort. STFCs are, on the other hand, not subject to reserve requirements, but are required to hold a proportion of their liabilities with the secondary (liquid) reserve assets. 22 STFCs have no access to the BOK's discount window. 31 negative effect that the respective independent variables have on each dependent variable. To analyze the processes of F1 that occurred during the past two decades, the above model can be represented graphically in Figure 2. If the banks and STFCs credit market were competitive and subject to no government intervention, the market would clear at 1'0 and the curb market would disappear. But the government set the loans and deposit rate at i far below the market equilibrium rate and thereby generate an excess demand for the credit measured by Q1Q3. In order to maintain I, the authorities have to exercise direct credit rationing among borrowers using various techniques to suppress the excess demand. The borrowers who are excluded from such credit rationing and the ultimate lenders who are not happy about the low regulated deposit rate, spill over into the direct and informal finance market as shown by FC‘:1 and FC'.23 It is important to recognize that money lenders in the curb market are subject to a relatively high degree of default risk, the risks associated with violation of the "usury law“, income-tax evasion, and possibly high costs of information due to market imperfections. These money lenders ask for a corresponding risk premium as the compensation for their informal loans. This characteristics of the curb market is captured by r0 higher 23 FCd is basically related to RCd. If regulated and unregulated market loans are perfect substitutes (i.e. , borrowers would be indifferent to the sources of credit so long as the cost of borrowing is the same), PC‘1 can be derived by subtracting 0Q1 at all levels of i from RCd schedule. FC' also can be obtained similarly from RC' curve if banks and STFCs deposits and informal securities are perfect substitutes in lender's portfolios. This traditional view of substitutability between bank loans and informal credit reflects the views of Mckinnon(1973) and Shaw (1973) on the interaction between the two markets. 32 than i. Given the demand for and the supply of the informal credit, the curb market reaches an equilibrium at r‘o. If the regulated market represents banks only, this situation correponds to the financial repression prior to the formal establishment of STFCs in 1972. Since the new quasi-banking (short-term) financial institutions were authorized, the integration of curb market into official market began to appear. The curb market disappeared for a while after the August 1972 reforms, but the reforms failed to eradicate the institutional factors or market forces that led to the existence of the curb market. This is primarily because the STFCs' interest rate was still pegged at much lower rate than the market equilibrium rate, even though the interest rate was slightly higher than the bank rate. It was inevitable that the curb market would reappear and even expand together with the growing economy in spite of the emergence of STFCs. However, the money lenders appreciated anew the dangers of illegal transactions in the curb market after experiencing the reforms. In an effort to undermine a loop ‘hole of the existing regulations, they developed a new, safer way of transactions deliberately taking advantage of official institutions.“ Because of this modified method of the 2‘ It is popularly known that informal money lenders frequently hold bank savings and time deposits or formal commercial bills of STFCs. With the additional funds, banks and STFCs extend loans to the particular borrowers whom the depositors agreed in advance. Without any risk of default, the money lenders then earn the savings deposit rate from the banks, and the spread between regulated and curb market rates from the borrowers. For the borrowers this arrangement is more attractive than borrowing directly from money lenders because they can often renew their loans at the bank lending rate. On the other hand, the bankers are involved in the transactions as a means of expanding deposits or of meeting the deposit quotas often assigned to them, and at the same time they usually receive a commission for such arrangements. Since the informal lenders and borrowers carry on their credit transactions through 33 Figure 2 Process of Integration of Curb Market into Official Market Official Credit Market Curb Credit Market 1' PC" rc- 178‘1 RC" \ ~ AFC (illegal) curb transactions, a relatively large portion.of the curb market appeared to be absorbed into the official institutions. In fact, there was no change in the workings of the curb market. Instead, both regulated and curb markets had grown with the "complement" relationships in the 19708. The spread between bank time deposit (or lending) and curb market rates continued to be nearly 25% points throughout the 19708. Such a large differential between the two rates suggests that the commercial bank loans and the curb market credit had been complements instead of formal financial instruments, the official institutions in this case provide an important linkage between brokers and borrowers. 34 substitutes in the borrower's portfolios.25 As noted in the previous section, most of the funds of WANMAE transactions shifted from the curb market to official markets. If the funds shifted into official markets due to an increase in the regulated interest rate as well as an. improvement of the existing statutory framework, it can be inferred that a large part of curb funds was shifted into official markets via the WANMAE transactions. Although there is no direct evidence of such a shift,26 it is quite realistic to claim that a relatively large volume of curb market funds shifted into official markets in the 19808 through emergence of WANMAE, expansion of CPs, and increased risk in the curb market, following the big financial scandal of May 1982. In spite of the sustained low bank rate, the spread between bank time deposit and curb market rates fell to 15% points in the 19808 from 25% points in the 19708. The curb market rate fell further than the bank rate. This fact suggests that the complement function of curb market loans to regulated market loans was considerably weakened because of the increased risk of curb market transactions as well as new diversified financial instruments. The money lenders gave more weight to the safety of transactions rather than to high interest rate. When new diverse 25 This complementarity relation would be, cetris-paribus, more conspicuous when the regulated interest rate is kept much lower than the market equilibrium rate and the anticipated rate of inflation is higher. However, the lack of reliable information either on the volume of credit or the interest rate in the curb market prevents us from rigorous empirical examination of the hypothesis. For the analysis of and the rationale for the complementarity relation, see Jaffee(l97l). 2°.According to the BOK's Survey ef Seviggs Marke; (1984, p. 76) the ratio of households saving through official institutions increased to 86.2% in 1984 from 55.4% in 1978. 35 instruments (e.g., WANMAE, CPs, negotiable CD8, and guaranteed bonds of medium and small cooperations) appeared and expanded as good substitutes for the curb market transactions, both borrowers and lenders shifted their financial transactions to official institutions. As a result, the curb market rate fell considerably. The consequent reduction in the spread between the two rates can be considered as evidence of a contraction of the curb market through an integration into official markets. The process of integration of curb market into official market can be easily shown in Figure 2: increased risk in the curb market shifts FC' to the left (FC"); decrease in FC' together with new diversified financial instruments shift RC' to the right (RC"); on the other hand, the diversity shifts FCd to the left (FC") and also shifts RCd to the right (RC"); and finally, since WANMAE transactions were the financial practice for regulatory avoidance, WANMAE and informal commercial bills market existed separately from0 (3.7) lnRSu - f1(lnR,), f', >0 Equations (3.6) and (3,7) represent the relationships between the representative open market rate and the rate of return on money and its substitutes respectively.35 It is assumed that there exists no interest rate regulation on the transactions deposits or the substitute assets, and 35 Of course, the representative open market rate can be considered as just an average of rates of return on various substitutes. Then the model is reduced to the preceding "two-asset" transactions model, so it is impossible to introduce explicitly new money substitutes into the “multi-asset" model of money demand. 49 that all interest rates are positively related.36 The crucial variable in the model is, of course, TC. In the original Baumol(l952) article, the term.”brokerage fee" represents all non-interest costs of borrowing or making cash withdrawals, such as (i) actual brokerage charge and (ii) implicit time costs of the transactions-- "shoe- leather” (Gordon, 1986) costs of going to banks.37 In the asset demand model, transactions (or transfer) costs are also an important aspect of asset demand. Since most near-money assets bear approximately equivalent risk characteristics, the two important features of competing assets are returns and liquidity (or the cost of transfer to an acceptable means of payment). In order to analyze the effects of F1 on M1 demand, it is convenient to distinguish two classes of F1. In addition to the technological 35 This model also can be used to examine the effects of both own- rate and substitute's rate deregulation on the elasticity of money demand with respect to the representative open market rate. While deregulation of the own-rate alone may decrease the elasticity--the same result that the previous transactions model implies, deregulation of rates on liquid substitute assets may inereaee the elasticity: alnmt/alnRt--bz+b2f'r (holding RS“ constant),|-b2(1-f'.r)l | -b,| The net effect (-b2+b2f'T-2buf'1) of both money and substitutes rates deregulation depends on the relative strength of bzf '1. and Zbuf '1. Without a knowledge of the specific functional form of equations (3.5) through (3.7), it is impossible to predict whether the interest elasticity would decrease or increase. 37 More generally, we can consider transactions costs as a function of capital and labor (the costs of factor inputs). 1Lf the exchange mechanism is regarded as a purely labor-using activity, then the costs are simply labor costs, (i.e., the real wage rate). Accordingly as the average real wage rate rises over the economy, the community will desire a greater stock of money balances (Kahn, 1973; Laidler, 1985). On a simpler level, one can assume that the real wage rate is a proxy for transactions cost. This treatment, however, ignores the other component of transactions costs. 50 advances which enable the public to economize on cash balances through reductions in transactions costs, FI takes the form of an increase in a variety of (liquid) financial assets which contribute to both medium of exchange and store of wealth functions in a varying proportion. First, consider the effects of reductions in transactions costs. The revolution in computer technology and data processing in modern financial systems brings out dramatic changes in the payment mechanism. It generally reduces both relative and absolute transfer costs and thus liquidity as well as variance of cash flows. This type of innovation causes the demand for M1 to shift down at given levels of income, prices, and interest rate as the public economizes on transactions balances.38 In equation (3.5), 1nm, decreases as lnTC, becomes smaller. The declining demand for the conventional transactions balances should primarily appear as a downward shift of the constant term captured by cllnTCt. However, we traditionally subsume transactions costs term (cllnTct) in the constant term (co) because of the absence of adequate data. This is, of course, appropriate if real transactions costs are constant. Alternatively we can accommodate a steady decline in TC with a time trend in the equation (Lieberman, 1977, 1979; Hetzel, 1984). However, so long as a model fails to account for the decline in TC, the 3" In the multi-asset stochastic money demand model that is an extension of Miller and Orr's(l966), Mibourne(1986) analyzes the theoretical effects of changes in a particular kind of asset transfer cost on M1, M2, and "Divisia" aggregate demand. He shows that the emergence of money market funds at a lower transfer cost leaves M1 unaffected because a reduction in such costs induces a switch out of savings (non-M1 components in M2) into money market funds, and that although the Divisia aggregate is never the worst, it is also never the best aggregate in any situation. 51 model may suffer misspecification and thus it can result in a systematic over-prediction of the money demand particularly during the period of rapid changes in TC. In addition to such a level effect, the decrease in transactions costs also can glee; the interest elasticity of money demand. As discussed earlier, wealth-holders of liquid assets become more sensitive to changing yield spreads at much lower transfer costs. Technological innovations and the ensuing reduction in transactions costs contribute to the coexistence of transactions and investment characteristics within the same asset, making the traditional separation between monetary and savings assets increasingly blurred. This complex phenomenon is again similar to the second case of F1 discussed below. As mentioned previously and above, the most significant aspect of F1 is that financial intermediaries provide a rapidly growing quantity of deposits and liabilities and other liquid assets with the help of financial (or deposit rate) deregulation as well as reductions of tranctions costs. Those assets may be close substitutes for checking account money and also may diminish the role of commercial banks in financial markets. The new near-monies, although not perfect substitutes for demand deposits, provide both transactions and investment services simultaneously and carry market-related interest rates at much lower transfer costs. The bulk of market-rate accounts would be non-reservable in the absence of regulation because of the cost disadvantage imposed by reserve requirements. Examples of such innovations include RPs, MMMFs, ATS, and interest paying demand deposits. The introduction of such instruments with "transaction-cum- investment" 52 services causes the demand for conventional transactions accounts to §h1§£ em as the public shifts savings balances previously lodged in the conventional M1 into the new attractive instruments. In equation (3.5), this level shift effect can be captured by the reductions in TC between conventional transactions money and.new instruments with an easier access to market rate earning assets. However, the demand for expanded definition of (transactions) money that includes all the new transaction- cum-investment assets will igerease if such instruments will attract some funds currently held in non—transactions instruments. Consequently, the conventional transactions money measure would shrink in demand, while the expanded definition would increase. This result will create a situation which needs a distinction like "MlA" and "M18" in the United States. In addition to the level effect, this type of financial innovation also raises the problem of interest and income elasticity effects. Controversies over how a proliferation of money substitutes affects the marginal relationship between interest rate and money demand are not new.39 Gurley and Shaw(l960) assert that "the growth of money substitutes increases the interest elasticity of money demand and so renders monetary policy less effective."(p. 240) .According to our model, the introduction of N new money substitutes inezeeeee (in absolute value) the interest N elasticity of conventional transactions money demand by 2 buf '1. i-n+1 39 About a half -century ago, Henry Simons, in his classic paper "Rules vs Authorities in Monetary Policy", noted that the growth of (effective) money substitutes would contribute to "financial instability", which can be interpreted to refer to the instability of money demand function. He advocated that a "program of monetary reform should seek to effect an increasingly sharp differentiation between money and private obligations.“(1936, p. 29) Ironically, the modern financial systems have been rapidly moving in the opposite direction of this suggestion. 53 Bringing to equation (3.5) the additional N numbers of money substitutes and taking the derivative with respect to 1nRt gives: n N n |alnm,/a1nR,|-|-b2(1-£',)-2 buf'1-2 b,,f',|>|-b2(1-f',)-z b,,£',|. 1-1 i-n+1 1-1 Thus the model implies that the increase in elasticity with respect to the representative open market rate can be attributed to some degree of we; mm between money and new money substitutes. This result coincides with the Gurley and Shaw hypothesis. Unfortunately, the claim ignores the important possibility that the introduction of more attractive assets causes Ml balances previously held as a store of wealth to flow out of M1 and into those assets. If it is true, the growth of money substitutes Shift§ Qom the money demand function and eliminates in good part the non-transactions balances in M1 which were once close substitutes for savings deposits and thus satisfied a wealth or asset demand for money. Then the remaining balances are largely for a transactions purpose. Conceptually, the demand for transactions money will show a closer and stable relationship with prices and income and a smaller interest sensitivity compared with the demand for savings-type deposits.‘° This hypothesis implies that most of the new financial instruments are not good (or perfect) substitutes for demand deposits but rather sub- stitutes for savings deposits or long-term investment assets. As a ‘0 This possibility was first suggested in Marty's review artic1e(l96l) of Gurley-Shaw, Money in a Theory of F1nance(l960). Since then, a voluminous empirical study on the subject has supported the hypothesis contrary to the financial view of money (Meltzer, 1963; Teigen, 1964; Cagan and Schwartz, 1975; Hafer and Hein, 1984). 54 result, a proliferation of diverse money and capital market instruments largely provided by non-bank institutions may purify the narrowly defined money by eliminating savings balances from Ml. This is exactly opposite to the conventional conjecture that it may seriously contaminate M1 with a characteristic of transaction-cum-investment services. Thus it can be concluded that the growth of money substitutes decreases, rather than increases, the interest elasticity of demand for traditional money. However, it is unclear whether the interest elasticity of demand for a comprehensive transactions money including the new instruments would rise or fall. On the one hand, the interest elasticity would decline for the same reason as in the case of deposit rate deregulation, as a larger and larger fraction of the funds in this total pays their own-rates that move more or less in tandem with short-term market rates. On the other hand, a large part of the funds lodged in such transactions instruments would easily be shifted into or out of direct holdings of money market instruments in response to fluctuations in the spread between own-rates paid on such transactions instruments and rates paid on a wide range of (liquid) financial instruments. Under such a world, Davis(l982) provides a conjecture that; "given the relatively high "investment” component that would presumably be reflected in the demand for these transactions instruments, it seems likely that this demand would be much more subject to shifts in investor sentiment relative to alternative investment instruments than is currently the case of the demand for transactions instruments. In particular, if the new transactions instruments were invested in short- term money market instruments so that the "own rate" were fixed relative to such short-term rates, the demand for these instruments might become quite sensitive to shifts in the market outlook for long-term debt and for common stocks. By the same token, the demand elasticity of transactions money with repect to long-term yields and yields on common stock 55 might be substantially greater than at present."(pp. 27-8) As a result, such a monetary aggregate would become more similar to the various financial assets held for investment motives if transactions and investment balances were to mix in an expanded transactions monetary aggregate. Its demand would be influenced largely by shifts in the public's portfolio composition when interest rate spreads and investors' wealth or sentiment change, rather than by changes in income and prices. If so, the income elasticity of demand for that monetary aggregate may also differ from that of the conventional transactions aggregate demand both because the income elasticity of investment balances may differ from the counterpart of transactions balances, and because its demand would be dominated by changes in wealth rather than in income. In summary, the central issue on the stability of money demand relationship ends in the question of whether the transactions and asset demand can be effectively separated even in the face of financial changes and of how seriously a reasonably defined transactions aggregate would be contaminated because of the existence of investment balances."1 If a particular measure of the transactions aggregate is actually distorted due to financial changes, there occurs a temporary shift in its demand function after financial changes. Since true transactions assets are unlikely to pay yields as high as other liquid assets that do not function as a part of the medium of exchange, such a separation of transactions and asset demand is likely to remain even after ID and Pl. As discussed earlier, the yields on ‘1 The discussion of redefinition of money allowing for institutional changes are presented in the next chapter. 56 transactions assets will be held down.by the costs inherent in the bank's asset transfer process for which FI has no necessary implications. The existence of relatively high reserve requirements (or voluntary reserves) on transactions accounts also holds their yields below the yields on non- reservable money market instruments. On the other hand, the public is willing to hold such assets of fixed nominal value payable on demand or in short notice in the absence of perfect synchronization of receipts and expenses even if their yields are relatively low. As long as the transactions and investment accounts continue to be effectively separated, the demand for a transactions aggregate will not be dominated by the shifts in the composition of the public's portfolio due to fluctuations in a wide range of yields or vagaries of investor's sentiment. Then its demand should be a relatively stable function of income and prices. The conclusion.drawn from analyses and arguments so far is that theory cannot tell much about the impacts of ID and F1 on the money demand relationship. Therefore, the theoretical conjectures have to rely on rigorous empirical evidence. Unfortunately, it seems unlikely that an empirical study can give any conclusive evidence on the subject, because financial systems are not in a stationary state but continually change reflecting the outcomes of all the complex, interrelated innovationary processes. 3.3. Theoretical Effects of a Change in Monetary Policy Regime As implied by the regulatory dialectic hypothesis in Chapter II, FI and regulatory responses usually interact each other. The two preceding equilibrium money demand models provide a simple explanation of the 57 effects of ID and F1, implicitly assuming that the public's money demand behavior remains unchanged in the face of alternative monetary policy regimes. However, it is well recognized that rational agents' decision rules can respond to a change in the stochastic processes of the exogenous variables facing them. The "invariance" assumption implicit in the conventional equilibrium models generally does not hold if the agents behave rationally (Lucas, 1976; Sargent, 1976; Kydland and Prescott, 1977; Barro and Gordon, 1983). In the United Kingdom, this argument has come to be called Goodhart's law: "Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes."(Goodhart, 1984, p. 96) In the context of the short-run money demand relationship, the phenomenon can be interpreted as follows. The aggregate money demand equation consists of the optimal decision rules used by wealth—holders to determine their desired real balances. The decision rules derive from maximization of the agents' objective function (e.g. , cost-minimizing in transactions model or wealth-maximizing in asset demand model), conditional on how the monetary authorities behave. Therefore, if the authorities change the monetary (or fiscal) policy regime governing nominal money stock, interest rates, and prices, rational agents' decision rules will vary and the aggregate money demand function will change accordingly. We employ a money demand model compatible with Friedman's argument"2 ‘2 Friedman(l956, 1959, 1969) has held fast to the view that the public's money demand depends on permanent, rather than current or measured, income and prices because they determine the real balances to hold in the light of their longer term position (see chapter V). 58 to illustrate how a change in policy regime can influence the public's behavior in determining its real balances. Suppose that rational agents determine their long-run equilibrium money balances on the basis of expected permanent (steady-state or natural-rate) income, expected permanent interest rate, and expected permanent prices rather than current or measured value of those variables. Then the expected (equilibrium) real money balances can be formulated in log-linear form:"3 (3.8) ln(Mt/,PP,,) - co + bllntypt - bzlntRPt It should be emphasized here that Co, b1, and b2 represent "policy- invariant” coefficients. They are derived from maximizing the agents' underlying stable preferences, which are independent of a particular policy regime. Subtracting lnP,‘ from both sides of equation (3.8) and expressing the dependent variable in terms of contemporaneous real balances gives: (3.9) 1nmt - co + bllntypt - bzlntRP, + (lntht-lnPt) Rational agents do not see directly the permanent values, but see only the current values of the relevant variables at the time that they make decisions on the amount of real balances to hold. In order to complete the model we need the hypotheses about the relationship between the unobserved permanent and the directly observed current values, and about how the agents form expectations of the permanent values in the face of imperfect information. ‘3 Keynes' ”liquidity preference" theory of money demand also implies that if the yield on the alternative asset to money is expected to change, the opportunity cost of holding money depends on the yield expected in the future (expected capital gain or loss) as well as the current yield. The money demand function then can be specified as lnmt-co+lnb1yt- bzlnRt-rbalnR',” . 59 We accomplish this by adopting Friedman's(l957) permanent income hypothesis together with a "recursive" (or sequential learning) expectations scheme that uses current and lagged values of measured variables to forecast the unobserved permanent variable. The supplementary hypotheses regarding permanent value of the variables can be specified: (3.10s) xv, - ,-,xp, + ux, (3.10b) - ,x9, + "x, (3.11) Xt - xv, + ex, (3.12s) ,-,XPc - E(XP,|0,_,) (3.12b) ,xv, R(xv,|o,_,, x,) where ,xv,-,yv,, ,RP,” and ,P9,; Xt-yt, R,, and Pt; ux,, nx,, and ext-random shocks to each dependent variable; E—expectations operator; and finally, (ltd-the agents' information set available at the beginning (t-l) of current period which includes at least the lagged values of nominal money supply, yt, Rt, and Pt. Equation (3.10s) shows that 16’, (permanent variable) is subject to a random shock UX, (the "surprise" part of the permanent value) that is assumed to be stationary, serially uncorrelated with zero mean and finite variance, and therefore that cannot be predicted from 0,-1; e.g., an increase in yPt due to the sudden productivity increase through technology innovation, or a change in RP, and PP, due to the unanticipated change in the growth rate of money supply. Equation (3.11) expresses that X, (current or measured variable) consists of the permanent part and the transitory shock part that has the same statistical properties as UXt; e.g., a change in y, (or transitory income) due to the unanticipated 60 change in money supply (level or growth rate), or a change in R, or P, due to the unanticipated one-shot change in the level of money supply. In addition, we initially assume that rational agents form expectations of X’, at the beginning of current period based on 0,-1 (equation 3.10a). Subsequently, the agents receive new information from the contemporary innovation of X, and revise the earlier forecast that is based on lagged (and hence imperfect or limited) values of information variables by utilizing the surprising information in the new observation X, (equation 3.10b). As long as the unobserved forecast error (UX,) is correlated with the newly observed forecast error (UX,+eX,) with a contemporary random disturbance (nX,) , the forecast of X’, can be improved. Equations (3.12a) and (3.12b) depict this nature of recursive expectations scheme, and provide the optimal forecast of XP, consistent with "economic and statistical theory" in a certain (or Muth, 1960) sensez“ 020x 01 - 020x + azex where H-vector of linear least squares regression coefficient conformable to 0,-1, cam-variance of UX,, cad-variance of eX,, and i-l, 2, 3 when X, denotes y,, R,, and P, respectively. The parameter 0, is referred to as a coefficient of ”signal extraction” that represents the conditional variance in X9, due to a variation in X,. The larger is this fraction, the larger is the weight placed on X, in revising E(XP,IO,-1) to form E(XP,IO,- “ For a derivation of the result, see appendix A. 61 1, X,). This makes sense because the larger 0, is, the more likely a change in X, reflects a change in XP, rather than a temporary change in X, itself. Substituting (3.13) into (3.9) gives: (3.14) lnm, - c0 + Cl + blollny, - bzozlnR, + (03-1)1nPt where cl-b1(1-01)H11nfl,-1-b2(1-02)Hzlnn,-1+(l-03)H31nn,_1 that shifts the constant term. Equation (3.14) now is a function of the observable vari- ables. It shows that income, interest rate, and price elasticity coefficients depend importantly on the parameters 0,. The larger 02,”, relative to 02“, the greater is the tendency of rational agents to regard a given unexpected variation in y,, R,, and P, as a change in yp,, RP,, and PP, to which their money demand should respond. The issue here is a problem that the value of 0, cannot be assumed unchanged across different monetary policy regimes. Under rational expectations, 0, (or a forecasting rule) depends upon the nature of the exogenous stochastic processes facing rational agents. Since a change in the authorities' monetary rules will alter the stochastic processes governing y,, R,, and P, and thereby 02,, and 02“, the parameters of the short-run money demand dynamics will vary whenever there is a change in the policy rules. If the authorities frequently change their policy attitudes in an attempt to exploit some plausible, favorable effects more fully, such a fine-tuning or discretionary (random; surprise) monetary policy will make the stochastic processes of y,, R,, and P, so unpredictable that the short- 62 run money demand function will become highly unstable.‘5 For example, the unanticipated one-shot change in the level of money supply produces temporary shocks for income, interest rates, and prices. The shock initially causes the temporary variation of these variables to be larger and in turn the coefficients of signal extraction to be smaller, but sooner or later this effect is reversed. Therefore, the short-run elasticities of money demand function will exhibit considerable instability.“ This result suggests that a stable short-run money demand function is more likely to exist when the authorities follow a strict monetary rule so that the policy-induced shocks can be minimized.‘7 ‘5 Cooley and.LeRoy(l98l) and.cordon(l984) raise another possibility, not based on the rational expectations hypothesis but in the context of identification and simultaneity issues, that changes in monetary policy regime would produce an unstable money demand function. According to their arguments, the typical policy regime has been a mixture of interest rate andumoney stabilization, and as a result elasticities of money demand cannot be identified; the coefficients in a money demand function are likely to represent a blend of money demand parameters with money supply parameters and thus shifts in the coefficient may tell more about changes in policy rules than about the responses of money demand behavior (see chapter VI). ‘6 The literature on the role of exogenous money supply in the money demand function and on "disequilibrium", ”buffer stock" or "shock absorber" theories of money (e.g. , Artis and Lewis, 1976; Laidler, 1982a) attempts to account for the instability of money demand relationship by emphasizing "temporary" off -demand-curve holdings of money following money supply shocks (see chapter VI). ‘7 Lane(l984) analyzes, in a dynamic money demand model setting, the effects of two specific monetary policy regimes on the problem of ”instrument instability”. He cautiously concludes that if the authorities attempt to smooth interest rate in the short run while gradually adjusting the interest rate to offset departures of the money stock from its target path for adhering monetary targets in the long run, that policy [R,-R,- 1+A(M,-1-M',)] would result in explosive movements of both the interest rate and the money stock, since that policy enables rational agents to anticipate future movements of interest rates and expectations of interest rate movements in turn make the demand for money unstable; however that there seems to be no reason to suppose for the interest rate to follow an unstable path if the money stock is controlled according to Friedman's 63 By contrast, suppose that the parameters of the short-run money demand relationships are invariant across alternative policy rules which the authorities might use. So long as the public is aware that policy regime has changed, it is inappropriate to assume that the public sticks to its old forecasting rules because these rules are no longer optimal for the new stochastic processes in force. The usual practice in simulating the conventional money demand models has been to assume that the stochastic processes governing these variables remain fixed even in the face of changes in monetary policy regime. But this involves the assumption that agents are "irrational" in forming forecasts. This criticism is the heart of Lucas'(l976) critique of econometric policy evaluation procedure."8 The above result, of course, relies importantly on the assumption that the public's expectations are rational and that money affects only nominal variables in the long run. If either of these two hypothesis is abandoned or relaxed, the conclusion of the short-run money demand instability posed by a change in policy regime will be different. For more rigorous examination of the money demand relationship in the rational expectations framework, we need more concrete knowledge about both the transmission mechanism of monetary policy and the correct expectations scheme. The constant growth rate rule. ‘8 However, Lucas in his later article (1978, 1986) explicitly points out how various adaptive (or irrational) expectations schemes lead to convergence to the stable rational expectations equilibrium. Since the agent can be expected to know or to have learned the consequences of different actions, his observed choices reveal stable features of his underlying preferences. Decision rules that are steady states of some adaptive process, work over a range of situations and hence are no longer revised appreciably as more experience accumulates (1986, p. 402). 64 complete structural model of the economy thus must be applied rather than a single money demand equation. CHAPTER IV FINANCIAL INNOVATIONS AND MONETARY TARGETING FI raises a number of policy issues, such as the competitive structure and market shares within the financial industry, the distribution of financial risk, and the regulation of financial institutions. However, from a monetary policy point of view, the most important issue is whether a transactions measure of money will continue to be a desirable ”intermediate target” and is how monetary policy should be conducted particularly when velocity may be changing. As discussed earlier, F1 is of great concern to the policymakers because theory and experience suggest that it is likely to make the demand for money (or income velocity of money) less stable or predictable. Greater instability of money demand may create serious problems for the continued use of monetary aggregate targets so that control of nominal GNP through money supply targeting would be inappropriate or impractical. Even though some transactions measure of money could continue to be used as the preferred intermediate target in the face of financial changes, monetary targeting would undoubtedly place a significant burden on policy makers when they have to evaluate the implications for targets of F1 and of the public's reactions to the innovations, and to adjust accordingly. It should be noted that the question of what the stable relationships between key macroeconomic variables imply for monetary policy is not a new 65 66 issue, but has been the central part of the great "rules vs discretion” debate."9 Despite the long-standing concerns, there has been little systematic research evaluating the effects of F1 on the usefulness of monetary aggregate targets. This is largely due to the difficulties in quantifying FI and its effects on financial markets. In addressing the issues, this chapter begins with developing a framework for analyzing sources of problems for monetary targeting posed by FI. Then the question of how the authorities should conduct the short- run monetary stabilization policy if they continue to use Ml as an intermediate target is discussed. In the third section alternative target variables to a txansactions monetary aggregate (Ml) are set forth and their potential as intermediate targets of monetary policy is evaluated based on the existing literature. In the final section the change in transmission mechanism of monetary policy, as the financial structure of an economy is gradually liberalized, is discussed briefly. Although that implication of financial changes is not directly related to the issue of monetary targeting, it seems particularly appropriate to the financially repressed economy. ‘9‘Viner(1962) argued.that if the relationships between key variables are unstable, then there will be no fixed rules available which will be both practical and appropriate to policy objective; it simply is impractical for policy to focus in some mechanical way on any single variable whether it.be Ml, GNP, interest rate, or even reserves themselves (p. 247). 67 4.1. A Framework for Analyzing Financial Innovations and Monetary Targeting The primary objective of monetary policy is to cause growth of some monetary aggregate equal to the predicted rate of growth of demand for that aggregate and thereby to achieve stability and predictability in its purchasing power. This goal is made operational by setting money supply growth targets equal to the predicted growth of nominal GNP less the predicted velocity (the ratio of nominal GNP to some money aggregate) growth. Of course, it is impossible from this relationship to determine the separate effects of changes in money supply in real output and prices.” Control of nominal income growth (ultimate goal) through a monetary aggregate target (intermediate target) using a reserve aggregate (operating instrument) is usually termed.as "intermediate target" or "two- stage" (control) strategy for conducting monetary policy. An.intermediate target lies between.the instruments of monetary policy that are more or less under the monetary authorities' direct control and the ultimate goal that can be influenced only indirectly by adjusting policy instruments. The authorities pursue this kind of strategy since it is more practical to achieve a goal by aiming at an intermediate target than by aiming at the goal directly. By using an intermediate and operating target, the policymakers can judge more quickly whether monetary i” A modern quantity theory of money demand (or nominal income) does not by itself tell how a change in nominal GNP is divided between price level and real income. In order to consider the separate effects, it is necessary to consider a concept of the Phillips curve (aggregate supply). Theory of the demand for money coupled with the monetarist theory of the Phillips curve (Friedman, 1968; Phelps, 1968) implies that the excess of the rate of money supply over the rate of full-employment (natural rate) output equals the rate of inflation in the long-run equilibrium. 68 policy is on the right ”track" rather than waiting until they observe the final outcome of the policy on nominal GNP. The intermediate target procedure can be expressed by considering the following two relationships.$1 First, money supply relationship is: (4.1) M, - R',m, where M measures the narrowly defined money stock (M1), R measures some reserve aggregate that can be used as an operating instrument by the central bank, and m measures the reserve aggregate multiplier. Second, money demand relationship (or equation of exchange) is: (4.2) Y, - M',V, where Y represents a measure of nominal income and V represents a measure of the income velocity of M1. Combining the two equations (4.1) and (4.2) in a growth rate form gives: (4.3) i, - £1", + v, - R', + m, + V, where dot denotes the growth rate of each variable. Thus the issue of precise control of Y, using a reserve aggregate operating procedure ultimately consists of how accurately both V, and m, can be forecasted.52 One necessary condition for the usefulness of money targeting is that 51 The framework is a slightly modified version of Rasche(1988). 52 F1 may also make currency-deposit ratio and reserve ratio and hence money multiplier 80 unpredictable that the central bank could not control monetary aggregate as intended. However, the study by Rasche and Johannes(1987) reports that the financial changes occurred in the U.S. during past decade had no significant impacts on the forecasting time series model of various reserve aggregate multipliers by ”component" approach. The issue of controlability of monetary aggregates is not discussed here since we focus primarily on the impacts of financial deregulation on the money demand relationship. 69 the money demand (or velocity) relationship must be a stable function of relatively small numbers of variables of interest--income, prices, and interest rate. The condition of stability of money demand function for the successful monetary control is quite rigorously and practically noted by Rasche(1988); "If stable relationships exist for the latter two variables, and if the forecast errors for these variables are not serially correlated, then control of nominal aggregates through a reserve operating procedure with an intermediate target variable such as M1 is feasible over medium-term horizons, even if the short-run forecast errors are so large that short-run control of nominal aggregates or even the intermediate target variable is very imprecise."(p. 454) Traditionally, the necessary stability has been considered more likely for M1 than broader aggregates because of Ml's unique role as the main part of medium of exchange.” That is, a narrow transactions aggregate has a good chance of having a stable and relatively uncomplicated demand function because it has few close substitutes. However, if the relationship changes unpredictably over time as F1 is going on, it will be difficult for monetary authorities to predict accurately what money growth rate is necessary to achieve the ultimate goal of stabilizing nominal GNP." As a result, the basic problem of money targeting policy 53 This is not the view of Friedman and Schwartz (1963, 1982). They have claimed that there is no compelling reason to regard the medium of exchange function as the "essential" function of money, and concluded that broader money inclusive time deposits is more precise empirical definition of money on the criterion of correlation with nominal income. 5‘ It should be pointed that there are relationships other than the money demand that affect the usefulness of monetary aggregate as a policy target. If the real sector becomes unstable [e.g. , Wenninger(l984) argues that financial deregulation would increase the interest elasticity of expenditure because variable (floating) loan rates would affect e1], borrowers, whereas with fixed rates loans when rates rise, only the W borrower is affected], the money-income relationship could be broken down even if the money demand were stable. Thus in focusing on the 70 depends on whether the money demand (or velocity) growth could be predicted with relatively high accuracy even in the face of the difficulties posed by F1. As explained in the previous chapter, it is difficult to evaluate correctly the theoretical impacts of PI on the demand for money even in a simple model. Many possibly offsetting changes could be occurring simultaneously. Some changes seem to be increasing the interest elasticity of demand for M1 and shifting the level of M1 demand down, while others seem to be decreasing the interest elasticity and shifting the level up. From the practical point of view, there are also a number of problems when the authorities attempt to identify a shift in the money demand function and to measure the size of the shift. Identifying shifts in the money demand is relatively easier in the case of ID or the introduction of new instruments than in the case of innovation unaccompanied by the creation of a new instrument. For the former, the authorities are able to identify and track the new developments from the very beginning. But for the latter (e.g., development of cash management techniques), it is very difficult to identify the initial shift since such innovations usually take place and diffuse gradually over a rather long period of time. In addition, there may exist a non-trivial time lag between identification of the shift and its measurement. Even when institutional stability of money demand, we examine a necessary but not sufficient condition for the successful money targeting. However, the argument relies importantly upon the real income elasticity of the demand for real balances. If the unitary income elasticity holds, then the stability of demand for real balances can be sufficient condition as well. 71 information clearly indicates that a financial innovation has taken place, a minimum number of data points (or time periods) are needed to measure the magnitude of the shift econometrically; Furthermore, if more than one change is going on at the same time, it will be even more difficult to measure the size of the shift. Consequently, the authorities may not be able to predict correctly what would happen to the demand for money at a point in time when FI takes place. While such changes in the money-income relationship are going on, the past statistical regularities do not provide a reliable means of picking target values for the conventional measure of' M1 that is consistent with the desired nominal income growth. Given these uncertainties about the implications of financial deregulation on the M1 demand, the practical question naturally arises about how the authorities should conduct monetary policy. Should the authorities abandon Ml as an intermediate target and consider other potential alternative target variables? Alternatively should they continue to use Ml as an intermediate target? And if the policymakers need some flexibility for variations in the growth of the aggregate around pre- announced targets in order to reflect institutional changes, what magnitude of flexibility should be permitted? The appropriate answer to these questions ultimately depends on how seriously FI causes the money-income relationship to deteriorate. This can be resolved only by rigorous examination of empirical evidence. FI own and does affect the moneyoincome relationship, but the problem of whether or not the predictability of monetary policy actions is actually influenced is an empirical one. Thus one direction of the response to F1 72 has been to search for a new stable money demand function because the ”true" money demand function may be stable regardless of financial changes. Merely saying that PI causes a deterioration in the money-income relationship and therefore that it is time to abandon money targeting is unconvincing argument. The substantial problems are largely associated with uncertainties in the money-income relationship during a transition phase until the new relations have settled down following FI. Certainly, this would create short-to-intermediate-run problems for manipulating M1 as a policy target. However, in the long run (long enough so that the authorities can have time to learn the new money-income relationship) there is no obvious reason to expect that the transactions money demand function would become so unstable that M1 would be inappropriate for an intermediate target. From one point of view, a decrease in the interest elasticity of demand for M1 ,with M1 being paid a market-related interest rate, would be rather desirable (Laidler, 1982b; Judd, 1983). If the demand for M1 becomes less interest sensitive and its velocity varies less with fluctuations of interest rates, then the linkage between M1 and nominal income becomes more stable. To put it in the simple IS-LM model, the LM curve becomes more vertical, so shifts in the IS curve produce smaller income disturbances than previously and shifts in the LM curve have a more predictable impact on income. Therefore, financial deregulation insulates income from various factors that otherwise would cause it to change unexpectedly. Such factors include fiscal policy, changes in inflation expectations, and instability in the public's demand for goods and services. As long as the 73 LM curve (or money demand function) remains stable and predictable, a steeper LM curve (or less interest-elastic M1 demand function) would not permanently cause problems for monetary targeting policy. Overall, it is true that the authorities could face considerable uncertainties during the transition period in which they are learning the new structural (or reduced-form) relationship of transactions money to nominal GNP and thus FI could make monetary targeting policy more difficult. This problem applies only to short-run or intermediate stabilization policy. This does not mean that FI would affect the long- run (or steady-state) properties of money. Changes in the growth rate of money supply would still have no permanent effect on the growth rate of real output and would end up changing the inflation rate. So long as the primary long-run objective of monetary policy is to constrain nominal GNP growth to a desirable (or zero) inflation rate, a role for some concept of transactions money is unavoidable. Then it seems clear that we need some idea of the definition of money even in the face of financial changes. The long-run consequences of F1 are likely ultimately to make monetary targeting policy a more reliable stabilizer, provided that the authorities are not locked forever into a growth rate rule for a particular definition of transactions monetary aggregate and thus they do not fail to monitor and adjust to the institutional changes which are bound to continue long after effects of F1 are at an end. 4.2. Adjustment of the Target Monetary Aggregate As mentioned above, it seems more appropriate and beneficial that authorities continue to target some measure of transactions money even in 74 the face of financial changes unless FI makes the money demand relationship highly unstable. If this is the case, policymakers must closely monitor developments with respect to a probable shift in M1 demand following innovations, so they are able to incorporate appropriately the corresponding change in velocity into their policy decisions. It is necessary to measure the magnitude of the shift as correctly as possible and to assess the nature of the shift, and accordingly to adjust either the target growth rate or the definition of the target money aggregate in order to take account of the shift. 4.2.1. Base Drift and Target Growth Rate Change The policymakers can deal with FI by retaining the existing,definition and by establishing a new target range that takes into account the size of a shift in the money demand. Then they must be able to make some judgement about whether the shift is permanent or temporary, and whether it is level shift or growth rate change. First, if a change in the money demand is identified as permanent, the apprOpriate policy response is a eomneneatery change in the money stock to offset the effect of velocity change on nominal GNP. But if a change is temporary, it is much better nee to reenong to the change, because a policy response may increase rather than reduce the variability of nominal GNP. For example, suppose that policymakers observe an increase in the money demand that they anticipate will reverse itself in the course of a quarter or two. If the policymakers want to neutralize the effect of the temporary change on nominal GNP, they will increase money supply to keep nominal GNP on its track and.then reduce money supply 75 later when the money demand change reverses. Since policymakers are generally uncertain about the timing and extent of such a shift, they may be too aggressive for too long, resulting in larger swings in nominal GNP than would have occurred otherwise. Such instability need not result inevitably from policy responses to temporary changes in the money demand.55 Second, the policymakers also have to distinguish between shift in the level of money demand and change in its growth rate. Figure 4 shows how policy should respond to the two different cases. In the upper part of the figure M, and V, represent the level of money stock and velocity respectively, while in the lower part M”, {4,2 and V”, 17,2 represent the growth rate of each variable implied by the slopes of each straight line in the upper part before and after the hypothetical change at to. Finally, Y, denotes a constant growth rate of nominal GNP. In the case of a permanent once-and-for-all downward shift in the level of money demand that leaves the growth rate unaffected (M,f4kz), policymakers must reduce the beee from which the growth rate of money supply is calculated, but not the target growth rate itself. This response brings about an unvarying 1, Alternatively, a permanent decrease in the growth rate of the money demand (8,94r2) calls for the corresponding permanent reduction in the grewtn rQEe at time to to offset the effect of a change in V, on Y,. If the policy fails to respond to the money demand changes 55 This line of argument is found in the literature on the "pitfalls" of stabilization policy (Baumol, 1961; Friedman, 1968; Friedman and Schwartz; 1963). Figure 4 Adjustment of Base and Growth Rate of Money Supply lnM, an, - . , ~~— ~—-——‘~——-cime to £81) Vt. I O I fr 4.: it. I i : I a . L #:5— <1 <0 1 . ' *' . l L__n__ __1 time to Downward level shift in the money demand at to with no change in the growth rate lnM, an, //::://////r””;;:ff’r i 1 I ! th vtz r £0 time 24.; Vt £1,>o if. I ...“ ——H~~- _ ‘~‘“_ 6"(0 I -..... ......- L t1“. to Decrease in the growth rate of the money demand at to 77 appropriately, the consequences will be different in the two cases. In the former, there is a temporary one-shot increase in the level of prices or real output or both. In the long run, however, the money demand returns to its previous growth rate as does the growth of nominal GNP. For the latter case, the growth rate of prices accelerates permanently, but the growth rate of real output may increase only temporarily. In general, FI can have a permanent once-and-for-all effect on the level of money demand, but perhaps only a temporary effect on its growth rate. As discussed earlier, an innovation that lowers the cost of holding M1 relative to non-M1 assets induces a portfolio shift out of non-Ml into M1 accounts permanently increasing the demand for M1 but accelerating its growth rate only temporarily. Once the portfolio realignment is completed, the growth rate of M1 demand simply may resume its previous path. However, as in the case of changes in the pattern of receipts and expenditure, factors that affect the level of money demand can influence 55 F1 can also affect its growth rate if they likewise change over time. the growth rate of money demand as income grows if it changes the income elasticity of money demand. But Fl does not need to produce a permanent effect on the level of money demand or its growth rate even if FI changes the interest elasticity of money demand. This is because the fluctuations in interest rates will average out over the course of a business cycle. 55 The structural change in money demand originates from various socio-economic factors which change slowly over time. Those include gradual development of cash management techniques, the public's banking habits, monetization of previously non-monetary sector, speed of urbanization, and so forth. For more details on the effects of structural factors on the money demand, see Lieberman(l977, 1979) and Mayor and Pearl(l984). 78 In summary, an attempt to make allowance for the impacts of F1 on the money demand in modifying base and/or target growth rate would place great burden on the policymakers. Precise assessment of how FI influences the public's money demand is required. Unfortunately, it is difficult to determine whether there has been a significant change in the money demand. It is even more difficult to differentiate between level vs growth rate or temporary vs permanent changes. Given these substantial uncertainties, policymakers should pay more attention to Brainard's(1967) finding that the optimal policy in such situations eannor fully eenieve the objective of policy.5‘7 Therefore, policymakers should not be aggressive in adjusting base and/or target growth rate but rather be more cautious than in tranquil periods. 4.2.2. Redefinition of the Monetary Aggregate The alternative and more fundamental way of adjusting the target variable is to redefine the monetary aggregate in such a way as to internalize the shifts that are occurring. This approach has the advantage of eliminating the difficulties of estimating the magnitude of an ongoing shift in the demand for the existing definition of transactions money when F1 is taking place. As mentioned above, one response to F1 has been the search for a new stable money demand equation in the face of financial changes. The usual approach for searching the true money demand function is concerned with the problem primarily in terms of misspecified independent variablesninappropriate choice of scale or opportunity cost 57 For a derivation of Brainard's result in the context of monetary targeting, see appendix B. 79 variables-- or incorrect functional form. By contrast, the discussions here are concerned with the instability of money demand function in terms of incorrect dependent variable in the money demand equation-- inappropriate aggregation of monetary aggregates. A8 FI may change the basic characteristics of the traditional M1, two different ways of searching for a new monetary aggregate have been considered. One direction is to simply redefine the aggregate to be more inclusive. When new near-money substitutes (e.g., NOW accounts, savings subject to automatic transfer, credit union share draft, andMMMMFs or RPs) are created and used as new means of payment or perfect substitutes for the narrowly defined money, all these substitutes that are spendable or automatically transferable into the conventional transactions money are logical candidates to be included in the medium of exchange definition of money."’8 However, certain market-related or fixed interest-bearing checkable deposits are the most plausible candidates for inclusion in a 5° Traditionally, there are three different ways of defining money; (1) money as a medium of exchange (ii) money as a liquid store of value or ”what money really does" (Gurley and Shaw, 1960) (iii) money as a ”scientific construct” (Friedman and Schwartz, 1963). All three approaches seem to agree that the conventional M1 as a medium of exchange is a part of money. But they disagree on the matter of whether other financial assets are so close to currency and spendable deposits and thus of whether one or more of these other things should also defined as money. At the heart of the controversy over what and.what not to include as money is the substitution relationship between money, conventionally defined, and other things. This old issue of substitutability between money and near-money has recently become a heated debate again as FI may have blurred the distinction between transactions and savings assets. Garcia and Pak(l979), Wenninger and Sivesind(1979) argued that the money redefined to include some of these new means of payment was found to have a fairly stable relation with nominal income and interest rate during 19708. 80 new monetary aggregatex"9 The rationale for the redefinition in this way is that deposits with similar characteristics, i.e., effective r e W for demand deposits, should be added together regardless of which type of institution, commercial banks or non-bank financial institutions, offers them; and that the checkable deposits should be included in M1 whether or not they pay interest (Simpson, 1979, 1980). As discussed earlier, any aggregate that contains interest-bearing checkable deposits is likely to have both transactions and savings characteristics, and therefore its growth might come at the expense of non-transactions deposits. Particularly in the absence of "universal" reserve requirements (i.e. ,imposition of required reserves on transactions instruments in the new aggregate regardless of issuers), the new aggregate comprised of both interest-bearing and non-interest-bearing components seems to be similar to a broader monetary aggregate like current M2. With an increased importance of the investment motive, the demand for that aggregate would be more subject to the vagaries of shifting market sentiment and become more unstable than the conventional M1 demand. However, in spite of financial changes it is highly likely that the pure transactions accounts and the savings accounts still can be effectively separated for the reasons discussed earlier. Thus the deliberately redefined transactions aggregate which includes the new interest-earning checkable deposits would be able to reestablish a stable transactions 59 In 1980 the Federal Reserve redefined M1 (the so-called MlB) to include NOW and ATS accounts ("other checkable deposits"). It also created a new artificial monetary aggregate (the so-called "shift-adjusted M1B") that adds back the fraction of any new instrument representing the shift out of the previous definition of M1. 81 money demand function so that the new aggregate can be used as a useful monetary target. Another potential direction of redefining the monetary aggregate is to develop new money measure as a weighted everage of the different basic monetary' components.5° The 'various liquid. financial. assets are ‘not perfectly substitutable among each other as money, but neither are they completely unrelated. This imperfect substitutability or differences in the "moneyness' of assets raises the question of whether the conventional simple sum aggregation (i.e., assignment of an equal weight of unity to each component) is an appropriate procedure for capturing the contribution of various assets to their monetary function. Some Federal Reserve Board staff economists (Barnett, 1980, 1982; Barnett, Offenbacher and Spindt, 1981; Porter and Offenbacher ,1982) have advanced "Divisia aggregates" for the “consistent" measure of money, which assigns a weight to each monetary component based on its "user (opportunity) costs" to represent the marginal monetary services yielded by each component. The user costs are proportional to the differences between the yield on a ”bench-market” asset--1ike human capital which yields no liquidity, and thus is held solely for its pecuniary yield-~and.the components' own yields. The share of each component's user cost in total user cost is used to weight that 5° Gurley(1960) was one of the first to promote this approach. He suggested that money supply be defined as a weighted sum of currency, demand deposits, and other substitutes with weights being assigned on the basis of the ”degree” of their substitutability. Under this definition, weights of unity are assigned to currency, demand deposits, and those perfect substitutes. Zero weights are given to assets which are completely unrelated to currency and demand deposit. And.weights between zero and one are given to assets which are imperfect substitutes for currency and demand deposits. 82 component's growth rate and the sum of these weighted growth rates equals the growth rate of the total Divisia index.‘51 The motivation for this application of Divisia aggregation to monetary assets is to obtain an empirically tractable means of separating the monetary (transactions) from the non-monetary (pure store of wealth) function, and to aggregate only over the monetary services. Combining liquid assets with very different user costs into a single simple sum aggregate will likely ensure that the effects of interest rate differentials show up in the behavior of that aggregate. By contrast, the index number aggregate appropriately compensates for these effects so the resulting aggregate presumably exhibits a more stable relationship with respect to an index of all the relevant interest rates than the simple summation of the assets. However, Divisia aggregates do not sufficiently deal with instability of money demand originating from reductions in the real transactions costs. The existing Divisia index procedure takes account of the interest rate differentials, but does not take transactions costs into account explicitly. Judd and Scadding's(l982a) examination of the empirical ‘1 The formula for the Divisia quantity index in discrete time approximation can be written as: n lt‘Qt. ’ Ith-l'LZI wit(1nQit ' anit- 1) where Q,-Divisia monetary quantity index for assets i-l. . .N and 1 szQn. Pit-lQit-l w“..- —< + —> 2 ZPnQu. 2P1t-1Qit-1 with P1, and Q1, denoting the user cost and dollar volume of the ith asset respectively. For more detailed treatments, see Theil(197l) , Barnett(l980), and Porter and Offerbacher(l982). 83 evidence concludes that the Divisia approach makes little empirical difference for the stability of the narrow aggregates (M1 and M2) in that it also faces all the problems 'plaguing the standard. money' demand function. But it does produce a more stable demand function for the broader aggregates (M3 and total liquidity). From the practical standpoint, the usefulness of the Divisia approach is also limited. The Divisia aggregate may not be an easy measure to understand, let alone control. In the absence of direct and reliable information on the opportunity costs of monetary assets, the user costs are very difficult to estimate. The question of how to redefine money is ultimately an empirical one. The composite of monetary items which bears the statistically most stable relation over time to income and other important macroeconomic variables should be considered money for the monetary targeting policy purpose. 4.3. Overview of Alternative Intermediate Targets Some of difficulties posed by F1 for the interpretation of M1 and for the conduct of monetary policy can be reduced by redefining M1 or by adjusting the target range as discussed above. However, given the unpredictable timing and size of the shifts in M1 demand, even the best adjustments would not get monetary policy back on the right track. If this is the case, it is no longer appropriate for the monetary authorities to continue using M1 as the preferred intermediate target. The authorities should then consider some alternative intermediate targets which would reduce the disruptive effects that new FI has on observed growth rates of M1. 84 Among the potential candidates for an intermediate target are interest rates (nominal or real), narrower monetary aggregates (the monetary base or a reserve aggregate), broader monetary aggregates (M2 or M3), certain non-monetary measures (debt or credit), and nominal GNP itself. Moreover, some Keynsians claim that in an uncertain world with rapid FI, it is more important that policy makers supplement information on monetary aggregates with all the information available from various sources, such as interest rates, near-monies, other financial assets, and direct indicators of current and future developments in the economy (Porter et a1, 1979; Pierce, 1982). This view seems to be consistent with Samuelson's(l970) ”look at (and react to) everything" dictum. According to the targets-and-instruments framework or optimal control theory,62 it is almost a theorem that policy can do better with a collection of variables rather than with just one. Any variable other than their ultimate target that the authorities cannot control directly should be used as an "information variable” (Kareken et a1, 1973); it may provide information on the appropriate settings of 52 Optimal control theory assumes that the authorities attempt to solve a stochastic optimization problem either for one period or for the entire future. They try to maximize some objective function (generally in some way related to the deviations of GNP from target level or path) subject to the constraints implied by the structure or "law of motion” of the economy that is usually assumed to have known er determinierie parameters. However, rational expectations school (Lucas, 1976; Kydland and Prescott, 1977) argues that it is inappropriate to analyze policy as a ”game against nature" and to prescribe the optimal way of manipulating a system whose structure is given. This entirely different approach is associated with the adoption of a money rule; the money supply itself is often proposed as the objective of monetary rule only for a long-run.anti- inflation monetary strategy. Lane(l985) surveys the literature on the targets-and-instruments of monetary policy providing various rationales for the use of intermediate target. 85 instruments in pursuit of the targets, but should not be regarded as a ”surrogate" goal of policy. Therefore, the intermediate target procedure which focuses exclusively on money, not only misuses the information provided by the variable used as an intermediate target, but also wastes the information embodied in other information variables. In this light, ”discretion" may be described as meaning that policy is formulated taking account of a varying body of information depending on what information becomes available, and that policy reacts to this information in a way that cannot be pre-specified, since it will vary over time as more knowledge about the structure of the economy becomes available. We can think of this type of approach as mnlr1n1e_rnrger§ alternative in contrast with money supply target only. At first glance, multiple targets strategy seems to be appealing. It is impractical, however, for the simple reason that multiple targets are usually incompatible in nature so they reduce the advantage of employing an intermediate target in the conduct of monetary policy. For example, money market models suggest that the authorities can almost always hit a monetary aggregate target or an interest rate target, but almost never hit both simultaneously. If a choice among a variety of alternative targets thus has to be made, at least two necessary characteristics of an intermediate target must be considered to evaluate the usefulness of those possible candidates relative to M1:63 ‘” In addition, it is frequently cited that a good intermediate target should be "timely and precisely” measured.and also convey some information about the "future" movements of the goal variable. But neither such data availability nor leading indicator characteristics seems the necessary condition for choosing an intermediate target although both characteristics can serve as sufficient conditions. 86 (i) The measure must be closely related to the economic activity so that changes in it account for a significant part of changes in the ultimate objective of policy. This criterion implies, in turn, that the intermediate target must exhibit a stable demand function of important variables of interest of policy makers. (ii) The measure must be subject to close and direct central bank control. This criterion also suggests a good intermediate target be not unduly influenced by changes in the goal variable of monetary policy. A first possible target is, of course, the interest rate. However, apart from the problem of determining what level of the interest rate is consistent with achievement of the desirable objective, there is little benefit from using an interest rate as the intermediate target particularly in an inflationary economy. As the past experience has proved, interest rates are poor targets for the purpose of controlling inflation because "cause and effect” runs more from inflation to interest rates than the other way around. Another potential target is the monetary base target advocated by Brunner(undated) and Friedman(l984) .“ Of the various monetary aggregates, the monetary base (high-powered money) is probably a less biased indicator than any other. It is more closely tied to, and more rapidly affected.by the instruments of monetary policy. However, it does not stand to reason that the demand for the monetary base (or base-income velocity) would be stable in the face of financial changes, while the demand for money would suffer considerable instabilities. Since the demand for the monetary base ultimately depends on money demand, the demand for base is equally likely to be unstable if the money demand is subject to instability. 5‘ Brunner has reported on the behavior of the base-income relationship for the Shadow Open Market Committee. Friedman's (1984) latest preference is to hold the base constant as an ultimate goal. 87 A third candidate is a broader monetary aggregate particularly M2. The proponents of M2 claim that it contains much more and relevant information about the movements of future GNP (Tinsley et a1, 1980; Roley, 1982).":5 This assertion is largely based on the possibility that PI is to some extent internalized by broader aggregates. Relatively small shifts in M2 compared to M1 are expected when the liquid funds released from M1 for investment in interest earning assets are shifted into assets included in the broader monetary aggregate. This case is true only if FI occurs within the banking sector especially between the components of M1 and.M2. It is not a general feature of F1 because a significant FI usually takes place outside commercial banks (e.g., non-monetary financial institutions that provide various money substitutes not included in M2). As noted earlier, the public's demand for transactions balances is theoretically more stable than the demand for savings balances. Moreover, M2 also suffers from definitional inconsistencies similar to M1. Some highly illiquid deposits are included (e.g., long-term deposits for some special purposes) whereas highly liquid accounts are excluded (e.g., large CD8 and money market funds). The deemphasis of M1 in favor of M2 as an intermediate target cannot be assessed fully without a comparative empirical analysis both in terms of the stability of the demand for M1, M2 and their controlability. A voluminous empirical research reports that the demand for M1 appears to have been surprisingly robust in the face of F1 and ID of the early 19808, ‘5 In the mid-1982, the unexpected sharp decline in M1 velocity led the Fed to place "less than the usual" emphasis on M1 and to give more attention to M2 as an indicator of monetary policy. 88 whereas the interest elasticity of M2, particularly non-Ml components, appears to have been more substantially affected (Hafer, 1981; Batten and Thornton, 1983; Judd and.Motley, 1984). This result suggests thatle will continue to be a useful guide to policy in the foreseeable future even though one certainly cannot rule out the possibility that Ml will be affected by F1 in the future. A fourth potential candidate is some broad debt (or credit) aggregate advocated by B.Friedman(l98l, 1982, 1983), Morris(l982), and Kopcke and Dockser(l982). This choice is based on their assertion that debt bears as close and stable relationship to economic activity as any conventional monetary aggregate, and that ultimately debt can ‘be controlled as precisely as the monetary aggregate. In addition, B.Friedman claims that policy makers should consider more information including observations on both ”liability and asset" sides of the balance sheet. The arguments have been criticized both on the theoretical and empirical grounds. First, there is no firm theoretical consensus concerning the relationship between debt (or credit) and economic activity, nor the mechanism by which debt could be controlled using existing policy instruments in the absence of direct credit controls. It is also likely that given the incentives for evading credit controls, even direct credit controls will not work for very long or in a predictable fashion. In addition, difficulties arise in.defining,a debt measure such as what range of the aggregate to include or whether to measure at face value or at market value. More importantly, there exist empirical findings opposite to the case for debt as an intermediate target (Porter and Offenbacher, 1983; Hafer, 1985a). In general the various tests show more evidence of 89 feedback from GNP to debt than from GNP to M1, and evidence that relative to debt, Ml exerts a greater influence on GNP. Furthermore, the evidence indicates that debt growth has no additional explanatory significance once the effects of M1 growth on GNP growth are accounted for. This result directly contradicts with the prevalent assertion that a multiple-target (the more the better) strategy would enhance the performance of an economy. The final and most controversial alternative is whether monetary authorities should target nominal GNP directly instead of using a monetary aggregate as an intermediate policy target. Many critics of money supply targeting have proposed nominal GNP as an alternative intermediate target, particularly in response to the recently alleged instability of money- income velocity (Tobin, 1980, 1983; Hall, 1983; Gordon, 1985; Taylor, 1985). As briefly mentioned in the context of multiple targets, these critics regard the appropriate role of money supply as just one of several information variables which should be incorporated in the optimal (combination) policy- -nomina1 GNP targeting. If the ultimate goal of the authorities is not to minimize the probability that the money supply will depart from the target range, but to influence the level of economic activity and the inflation rate, the optimal money policy will naturally be different from a policy that is desirable under a monetary targeting policy. Thus the critics of money supply tagets have argued that it is "suboptimal" for the monetary authorities to use money supply as an intermediate target. That is, the strict adherence to a pre-announced path of money supply would result in not only making inappropriate use of 90 information that the money supply contains, but discarding the information which other important variables provide. Specifically, the usual arguments for nominal GNP target are based on the following assertions.66 Nominal GNP targeting engenerieelly calls for an offsetting adjustment in money supply when velocity changes due to shocks to aggregate demand side, such as shifts in money demand as a result of PI. Thus policymakers should exercise complete discretion to accommodate velocity surprises with money supply changes so that real GNP is stabilized. On the other hand, it calls for an automatic reduction in real GNP when shocks originate from aggregate supply side, such as abrupt changes in oil price or (labor) productivity. This should stabilize inflation, as the resulting slack will put downward pressure on the price level. In any case, the target path selected for nominal GNP should not be independent of the current and expected future state of the economy. It can. be reset annually taking into account all the sources of information.available including price and wage developments, unemployment and real output, and so on. Having explained the arguments for nominal GNP targeting, we can evaluate it simply in terms of its relationship to ultimate objective although such a naive assessment is unlikely to resolve the real issue discussed later. Nominal GNP can be intermediate (not ultimate) target ‘5 In order to go beyond the naive rationale for GNP targeting, Taylor(l985) analyzes, in a dynamic setting, the "propagation" effect of GNP targeting on a business cycle in addition to the immediate ”impact" effect. His findings confirm that GNP targeting of feedback responding to lagged inflation and lagged changes in inflation would not much improve the record of discretionary policy. He also points out that the discussions of GNP rules usually fail to link the specification of the rule with the behavior of the monetary authorities. 91 variable in the sense that policymakers do not care about nominal GNP, per 8e; they really care about the mix between real output and prices. It is then obvious that nominal GNP is more closely related to the ultimate goal of policy than any other possible intermediate target variable. However, nominal GNP is only partly determined by monetary policy because many other factors, such as fiscal policy and various exogenous shocks, also have an influence on it. These other factors may cause nominal GNP to fall even though monetary policy is expansionary. Moreover, it is almost impossible to distinguish clearly whether fluctuations in nominal GNP are due to velocity shocks or due to price and productivity shocks. A8 a matter of fact, we may think of nominal GNP target as "velocity- adjusted" monetary aggregate target because ultimately it must be translated into some monetary aggregate that the authorities can control.67 Thus there appears to be less difference between a nominal GNP target and a monetary aggregate target than appears at first glance. However, the twa targeting procedures are fundamentally different on the issue of whether policy makers should consistently follow stable announced "rules” or whether they should have and exercise ”discretion" in successive decisions . ea ‘57 If we consider nominal income targeting as a money supply rule with feedback, then the equivalence of constant money growth rule and nominal GNP targeting depends on the erocnaetic nrenertiee of income velocity of money. Suppose that the feedback monetary policy obeys AlnM,-c+BE,- 1[Aan,], 45850, and that the velocity follows a random walk (i.e., Aan,-p+e, and thus E,-1[Aan,-p) . Then the monetary policy rule for nominal GNP targeting becomes AlnM,-c+Bp that is a constant. 58 After all, the real issue ends in the long-standing great debate on "rules vs discretion" for which a rigorous discussion is beyond the scope of this study. 92 Consider a discrepancy between predicted and actual velocity. If the authorities operate by announcing monetary aggregate target, then they are more or less locked into this target range. In contrast if they announce nominal GNP target, they can change their implicit target for the monetary aggregate much more easily. The central issue in the debate is therefore the extent of flexibility that the authorities should exercise in the conduct of monetary policy. Although it is far from clear that the answer is ”the more the better”, we can cautiously argue against the greater flexibility that is provided by nominal GNP targeting. First, by allowing the authorities to fine-tune in order to adjust for every random fluctuation in the velocity, greater flexibility has often had destabilizing effects in the past. By implication, the economy would surely have been more stable if a rule for the money supply growth had been followed (Friedman and Schwartz, 1963) . Secondly, it may provide the authorities with the plausible excuse or rationale for accelerating money supply growth whenever they want. The policymakers might appeal to instabilities and shifts in the money demand function and thus revert to the traditional policy of stabilizing interest rates by accommodating the procyclical elements in the money demand. Thirdly, in setting a nominal GNP target the monetary authorities are likely to be exposed to greater political pressures than in setting a monetary target. Finally, more forceful criticism of nominal GNP target (or optimal monetary policy) can be found in the rational expectations literature. Under rational expectations, the main conclusion implies that policy should be predictable; optimal feedback policy is no better than any simple pre-specified policy rule; and unsystematic (and thus 93 unanticipated) policy will, on balance, increase the variance of output (Sargent and Wallace, 1975; Barro, 1976a) . In addition, the Lucas' critique(l976) and.the "time-inconsistency" problem (Kydland.and.Prescott, 1977; Barro and Gordon, 1983) suggest that optimal policy may be very difficult to design and also be an ill-defined problem. It is therefore preferable for the authorities to follow simple pre-announced policy rules which at least give the public a firm basis on which its expectations of the future are formed, rather than to attempt to carry out an optimal policy which is ill-defined. In conclusion, the case for alternative targets are relatively weak theoretically as well as empirically. We have accumulated considerable knowledge bearing on the transactions monetary aggregate relative to other potential target variables, and thus the abandonment of monetary targeting policy is very costly. Furthermore, monetary targeting is a necessary strategy for the ultimate goal of stabilizing prices as long as we accept the proposition that ”inflation is essentially a monetary phenomenon." 4.4. Changing Transmission Mechanism of Monetary Policy The channels through which monetary policy operates depends on the financial structure of an economy. Financial liberalization, by removing interest rate regulationLand.direct credit controls, changes the structure of financial markets. Consequently, it influences the relative importance of different channels for monetary policy. In a financially repressed economy, with interest rate restrictions and domestic credit controls, the effects of' monetary policy ‘work primarily through changes in the credit availability rather than 94 adjustments of interest rates. Consider the action by the central bank to tighten monetary policy either through rationing of its discount window credit or its guidance on the growth of bank lending to the non-financial sector. The resulting financial "crowding out" squeezes aggregate demand and dampens inflation. But adjustments of interest rate ceilings in the regulated institutions cannot influence the attitude of the public and result in little effect on changes in aggregate spending, so long as the regulated interest rates are far below the free-market rate. By contrast as the government reduces or removes its direct controls over interest rate and credit allocation in the regulated market, market forces become more important in the allocation of credit so that the role for interest rates in transmitting monetary policy effects is increased. The growth of less regulated.or unregulated financial institutions and the elimination of interest ceilings on bank deposits and loans expand the range of financial instruments affected by the variations in market interest rates. As a result, when interest rates rise in response to contractionary monetary policy, the public is induced to hold more financial assets. Then aggregate demand especially for fixed investment is depressed. Monetary policy can also have adverse short-run supply-side effects to the extent that rising interest rates increase the cost of working capital. Put in another way, changes in market interest rates are transmitted more rapidly and pervasively to all sectors of the economy as more financial instruments carry market-related interest rates. CHAPTER V EMPIRICAL EVIDENCE ON THE STABILITY OF THE SHORT-RUN MONEY DEMAND FUNCTION IN KOREA The theoretical conjectures about the impacts of a changing financial environment on the money demand relationship are exhaustively examined in chapter III. In that investigation the most striking impression is that theory alone can not discriminate among the numerous hypotheses about the impacts of financial changes on a monetary economy without systematic and rigorous empirical research. Unfortunately, it is difficult to evaluate the effects of F1 on the money demand function and the usefulness of the monetary aggregate target econometrically. PI is not yet complete but still ongoing. This is especially true of Korea, which is in the transition stage of structural adjustment from a developing to a developed economy. In general, several changes take place simultaneously so it is even more difficult to quantify and separate the financial changes and their effects on the money demand (or velocity) behavior. Under these circumstances it is the height of arrogance to suggest that there are conclusive findings associated with the empirical study of this issue. However, it is worth while to investigate whether the money demand function in Korea has been stable throughout the rapid and dramatic recent financial changes, and to attempt to identify the nature and causes of the observed statistical instabilities. The empirical study allows us to 95 96 assess the previously discussed various theoretical hypotheses statistically, based on the experiences of F1 to the present. By ruling out the plausible ”lip" services that are not supported by empirical evidence, the study contributes to narrowing the real problems that the monetary authorities face during financial deregulation. In this chapter we investigate the stability of the money demand relationship for Korea during the past two decades--a period of substantial financial market changes-- in the formal econometric framework by utilizing time series data on money, prices, real income, and interest rates. little has been done to specifically investigate the issue of money demand stability for Korea. This is in a sharp contrast with an explosive outpouring of research on velocity or money demand stability for the United States, that followed the so-called "missing money" phenomenon. The study examines the stability of quarterly money demand specifications for the period l973/I-1989/II. Since the use of annual data severely restricts the available degrees of freedom, tests of an annual money demand function cannot be conducted rigorously. The choice of the initial sample point for estimating money demand specifications is motivated by two considerations. First, it marks roughly the incipient stage of the short-term finance companies (STFCs) which are the harbinger of F18 that occurred thereafter. Second and more importantly, the average balances of monetary aggregates as well as the official interest rates on short-term bills are not available prior to late 1972. This chapter is organized as follows. In section.1 some specification problems related to the short-run dynamics of money demand function are reviewed briefly. In section 2 the appropriate short-run money demand 97 function is specified, which considers some special aspects of Korean economy. The various hypotheses on the money demand specifications are also tested. We primarily focus on MlB aggregate, which is not the official definition of money but most analogous to the transactions money like MlB in the U.S. M1 (conventional narrow money) and.M2 (broad money; official target aggregate) are examined when.appropriate. The comparative study proves to be useful especially in association with the stability tests of money demand function throughout financial changes. In section 3 a series of stability tests for these monetary aggregates are performed. We investigate whether the money demand relationship has been stable over the recent financial changes, and try to identify the extent, nature, and causes of any statistical instabilities. The various hypotheses concerning the behavior of money demand in the face of financial changes are evaluated in section 4. 5.1. Overview of Specifications of the Short-Run Money Demand Function In the transactions view of the demand for real money balances, money is held primarily because of the lack of synchronization between receipts and expenditure as well as the existence of positive transactions costs. Many of the empirical short-run money demand specifications are founded on such a transactions theory of the demand for money. The underlying hypothesis is that there exists an aggregate long-run (desired) equilibrium for real money balances (m,: typically divided by the GNP 98 deflator)69 as a function of a real income (transactions) measure (y,: typically real GNP) and one or more nominal interest rates (R,: typically short-term rate) reflecting the opportunity cost of holding non-interest- bearing cash balances rather than other possible substitutes for money. This relationship is commonly specified in double-log linear terms:70 (5.1) lnm”, - “o + allny, - (:2an Equation (5.1) often has been estimated directly using annual data. Since the equation represents the long-run equilibrium in which full adjustment between actual and desired real money balances is assumed to be completed within one year, no particular partial adjustment process is specified. 6° The usual assumption on the relationship between equilibrium nominal balances and equilibrium real balances can be expressed as lnM*,-lnm*,+1nP,. It restricts the equilibrium demand for nominal money balances to homogeneity of degree one in the price level or the zero price elasticity of the demand for real money balances, while the short-run demand for nominal balances may not satisfy such a homogeneity constraint. All the theoretical models of the demand for real money balances imply the zero price elasticity and the overwhelming portion of empirical evidence confirms the hypothesis as well. 7° In fitting the money demand equation statistically, the logarithms of variables are usually taken rather than their original values. The device has an advantage of reducing the possibility of heteroskedasticity problem when the error term needs to be added to equation (5.1) in converting into tochastic function, and of giving a convenient linear equation to fit. With the advent of high and volatile short-term nominal interest rates, ”semi-log” specification that includes levels of rates rather than their logarithms (i.e. , lnm,-ao+a11ny,-a2R,) has been popular in an attempt to improve the post sample forecasting ability of earlier study. By estimating a W coefficient (82) instead of a s i t coefficient ((22) at all levels of rates in double-log specification, the semi-log specification allows interest rate elasticities to vary positively with fluctuations in the level of rates (az-az/R,). That is, the semi-log specification implicitly assumes that the M rather than W in interest rates is more relevant to the demand for money. However, the slope coefficient is not free from units of measure whereas the elasticity coefficient is free from units of measure. For theoretical considerations in favor of semi-log specification, see Friedman and Schwartz(l982, pp. 265-66). 99 However, when equation (5.1) is estimated with quarterly data, a more flexible specification is needed to characterize the short-run money market ”stock disequilibrium” that may exist. The observed aggregate real money balances (m,) and the desired equilibrium holdings (m*,) may not be equal in the contemporaneous period since transactions (portfolio adjustment) costs can prevent immediate adjustment of actual balances to their desired level. Thus a supplementary "stock adjustment” hypothesis is required to reduce the empirical problem to the observed variables. The most commonly used adjustment mechanism-~a ”real stock adjustment" hypothesis (RSAH)-- can be formulated: (5.2) lnm,-lnm,-1 - A(lnm‘,-lnm,-1) where OSASI represents the coefficient of adjustment--the speed at which actual money holdings adjust to the gap between last period's stock and current desired level. Substituting equation (5.1) into equation (5.2) gives: (5.3) lnm, - A(ao+a11ny,-a21nR,) + (l-A)lnm,-1 The equation then represents the "Goldfeld specification” or the 71 “standard" quarterly money demand function. One important implication 71 The stock adjustment lag approach has been criticized by Friedman(l959, 1969) on the two grounds; first, changes in the relevant income variable (permanent income) cannot produce discrepancies between the actual stock of assets and the stock of assets desired in the long run and second, changes in the desired share of money in total assets, owing to changes in interest rates, are achievable with little or no lag. A specification similar to equation (5.3) then can be generated if one assumes that the desired demand for money depends upon expected or permanent values of the explanatory variables rather than their current or measured values and the desired level is always achieved, and that expectations are formed adaptively (see chapter III). Thus, the dynamics of the adjustment process could be due to W rather than transactions costs. That is, the gradual adjustment in holdings of money to observed changes in its determinants reflects an W £211 response of demand to the expected or permanent values of the 100 of this specification is that an increase in the price level will induce an immediate increase in nominal money holdings to equate the real value of last period's nominal money holdings to the currently desired level. The RSAH has been criticized because of its asymmetric assumption that while the adjustment of money holdings will occur partially and gradually in response to changes in income or interest rates, such adjustment will occur fully and instantaneously without a lag in response to price level changes. Goldfe1d(1973, 1976), White(l978, 1981), and Laidler(1980) have suggested an alternative adjustment mechanism--a "nominal stock adjustment" hypothesis (NSAH). The hypothesis can be written: (5.4) lnM,-lnM,-1 - A(lnM*,-lnM,-1) where M is nominal money balances, that is, M,-m,P,. Combining equations (5.1) and (5.4) and solving for real money balances as the dependent variable gives: (5.5) lnm, - 1(ao-I-a11ny,-lnR,) + (l-A)1n(M,-1/P,) The only difference in specifications between money demand equations (5.3) and (5.5) is the form of the lagged variable. Under RSAH lagged nominal money balances are deflated by lagged prices, whereas under NSAH they are deflated by current prices. However, implicit in NSAH is the assumption that real money'balances respond negatively to the observed inflation rate with the same geometric distribution lag as for changes in income or interest rates. This is easily shown by rewriting equation (5.5) in the independent variables combined with a gradual response of the expected values themselves to the observable current values. 101 form: 72 (5.6) lnm, - Alnm', + (1-A)lnm.,-1 - (1-A)ln(P,/P,-1) By contrast, the real stock adjustment model (5.3) implies that given m', there is no independent effect of the observed inflation rate. In periods of relatively stable prices, such as in the 19508 to the early 19708 in the United States, the restriction in RSAH is not a major source of difficulty in empirical studies. However, in less developed countries which are experiencing highly variable and substantial inflation, this kind of restriction can be the source of considerable parameter instability. It can easily introduce a large upward or downward bias in the estimated value of the speed of adjustment coefficient, depending on the size and sign of the covariances among the independent variables. Specifically in order to test which, if either, of two hypotheses is consistent with data, equations (5.3) and (5.5) can be combined to form the nonnested composite model in terms of real balances: (5.7) lnm, - Alnm',+(l-A)lnm,-1+B,ln(P,/P,-1)+lenP, - A(ao+a11ny,-a21nR,)+(l - A) lnm,-1+B,,ln(P,/P,-1)+lenP, - E°+Bllny,-lenR,+831nm,-1+B,1n(P,/P,-1)+lenP, 72 Taking Friedman's idea for money demand into consideration, the relationship between ”equilibrium" nominal balances and ”equilibrium” real balances could be stated more appropriately as lnM*,-1nm',+lnP‘,, where P', is some measure of the expected or permanent price level. If this relationship, rather than the conventionally assumed relationship (lnM*,-1nm*,+1nP,), is substituted into NSAH(5.4), the resulting short-run demand for real balance is 1nm,-lnm',+(l-.\)ln(M,-1/P,)-A(1r,-1r',), where x,-ln(P,/P,-1) is the observed inflation rate, 1r,-1n(P',/P,-1) is the anticipated inflation rate, and thus (1r,-1r',) represents the unanticipated inflation rate. For an elaborate interpretation of nominal stock adjustment model, see Rasche(1986, pp. 21-22). 102 The linear restrictions can be imposed upon 8, of equation (5.7) implied by the alternative stock adjustment hypothesis and by the linear homogeneity hypothesis in real income or price level. If RSAH is true, then 8,-0. If NSAH is true, then B3+B,-O. In addition, homogeneity of degree zero in price level and unitary long-run income elasticity imply 85-0 and 81+83-1 respectively. Combinations of these restrictions lead to numerous specifications of quarterly money demand function such as the models of Hamburger(l977, 1982) and Goldfeld(l973, 1976).73 As indicated above, both RSAH and.NSAH money demand models are highly restrictive. These models constrain distributed lags to decay geometrically and require that the distributed lag pattern be identical for each independent variable. A more general form of the demand for money which includes RSAH and NSAH as special cases is given: (5.8) lnm, - 00 + 01(L)1ny, + 02(L)lnR, + 03(L)lnP, where 01(L), 02(L), and 03(L) are polynomials in the lag operator L that is defined by LX,-X,-1. The model allows the distributed lags to be different for different independent variables and to take on any shape. In this regard, Rasche(1986) comes to the conclusion that: "The current state of money demand theory reveals little if anything in the way of prior restriction on the shape of distributed lags, thus the true priors that can be derived without reference to the data are quite diffuse. Under such circumstances, cautions [Schmidt and Waud(1973)] about the dangers of specification error because of erroneous restrictions on estimated coefficients throw doubt on the credibility on many of the existing results."(p. 25) 73 Empirical results for the United States (e. g., Mibourne, 1983; Spencer, 1985; and Hwang, 1985) show that NSAH dominates RSAH, and that the unitary long-run income elasticity is generally not supported by the data contrary to Hamburger. 103 By contrast, the RSAH model imposes restrictions; 01(L)-31{§§1-A)1L1, 02(L)-32A1€§l-A)1L1, and 03(L)-0, while the NSAH model imposes the same restrictions in 01(1.) and 02(L) but is less restrictive with regard to the price level; 03(L)-(A-1)+ %_gl-A)1+1L1. Indeed, Mehra(l978), Spencer(1985), and Rasche (1986), suggest that the bulk of polynomial distributed lag estimates which impose a common geometric decay or relatively low order (frequently quadratic or cubic) polynomial restrictions on distributed lag patterns, are not supported by the data. Another completely different criticism of the equilibrium specification associated with the stock adjustment models has been raised by Carr and Darby(198l), White(l981), Judd and Scadding(1981, 1982c), and Laidler(1982a). They argue that it is inappropriate to regard any observed real balances as the point which lies on a short-run money demand curve. The conventional equilibrium money demand specifications implicitly assume that aggregate nominal money supply responds passively to exogenous changes in the demand for money in the face of changes in each argument in the money demand function. But in a world where the aggregate nominal money stock is exogenously determined (i.e. , completely controlled by the monetary authorities), the public must temporarily accept changes in its money holdings via a passive, ”shock absorber" or "butter stock” reaction. In terms of the conventional equilibrium version, this implies that the public is forced temporarily to hold 104 disequilibrium amounts of money that are off its short-run demand curve." Consider the case when.undesired temporary holdings of newly acquired money are provided by money "dropped from an airplane” frequently called "helicopter money". Just as wealth holders are assumed to adjust real money balances to the new desired level only gradually in response to changes in income or interest rates so they would dispose of such a surplus windfall acquisition of money only gradually. This kind of adjustment lag is not captured in the conventional money demand models. The proponents of temporary disequilibrium money provide for the lag by adding a fraction of the current exogenous or unexpected changes in money (Carr and Darby) or changes in the (bank) credit availability (Judd and Scadding) to traditional equilibrium money demand models. The shock- absorber version of nominal stock adjustment model then can be written: (5.9) lnm, - Alma“, + (1-A)1n(u,-,/P,) + 61n(sc,/Bc,_,) where BC represents some measure of outstanding (bank) credit or nominal money supply innovation. However, the additional variable in real money balances regressions can be considered a 12duggd;§g;m variable that proxies more complicated behavioral relationships or transmission mechanisms of money, rather than measuring the impact effect of money supply shock on the demand for money. Consequently, it becomes an intractable problem to identify the demand curve in such a model. Furthermore, White(l981) argues that: ”A number of neglected considerations justify the expectation that only a relatively small portion of actual money supply 7" White(1981) calls this short-run disequilibrium ”flow disequilibrium", while he calls the divergence of the short-run desired level (indicated by the short-run demand curve) from the long-run desired level "stock disequilibrium". 105 changes would not be accounted for by equilibrium demand functions: it seems unlikely that open market changes in interest rates would cause disequilibrium; any endogenous changes in interest rates induced by other changes in money would act on the high long-run interest elasticity of demand; and what undesired money still survived would probably be pushed endogenously out of circulation.“(p. 535) Thus he concludes that the proposed supplementary behavior relationship to take a temporary disequilibrium money into account seems theoretically much weaker than is generally perceived. The final comment on the conventional money demand specification is related to "identification and simultaneity” issues raised by Laidler(1980), Cooley and LeRoy(1981), Gordon (1984), and Hetzel(l984). This argument assumes identification of an equilibrium money demand function in the form of (5.1), but questions the standard specifications used to identify the short-run money demand equation, that implicitly assumes aggregate nominal money supply responds passively to any exogenous change in the demand for money. The critics contend that the right hand side (RHS) variables in money demand regression, interest rate and real income, are not exogenous, and that aggregate nominal money supply is set by the monetary authorities in a way that makes money respond less than completely to the arguments of the money demand function. The standard stock adjustment models then can be considered a W, rather than a function representing the public's demand for real money balances. In the absence of firm knowledge of what constitutes the "true" structural model, the common practice of estimation of money demand regression equations by ordinary least squares (0L8) is subject to the problem of simultaneous equations bias. The 0L8 estimates can not be relied upon to yield unbiased 106 estimates of the structural parameters of the aggregate money demand function. Some remarks are warranted with regard to these criticisms. First, empirical estimates of the nominal adjustment specifications traditionally define the dependent variables to be real money balances as given by (5.3), not nominal money balances as given by (5.4). Such real balances can be viewed as W, that is, determined by the price level, interest rates and real income.” Second, in a less developed country a typical monetary policy regime can be represented by the "real bills" doctrine. The central bank always stands ready to rediscount "eligible" real bills (related to real business as opposed to speculative transactions) through the discount window, pegging interest rate at low levels in the short run as well as in the long run. The government also tends to administer prices through price and wage controls (incomes policy). In this case, it is quite reasonable that even nominal money supply is endogenously Wings! whereas official interest rates and measured prices (though distorted) are W with regard to the demand for real money 75 Empirically Mehra(l978) found that the real money balances are not statistically exogenous with respect to interest rates and real GNP. In a similar spirit, Rasche(1986, p. 24) further elaborates that with the assumption of M't-m'tP't rather than M't-m'tPt, aggregate expected real balances are assumed to be adjusted to equilibrium real balances, expected change in the price level, and previously held aggregate real balances: 1n(M,/P',)-Alnm*,+(1-,\)lnm,-l- (1-;\)1n(P't/Pt_1) As long as the expected price level is not totally predetermined, ln(Ht/P't) is clearly an aggregate which is controlled by decisions of private economic units regardless of the monetary control regime. 107 balances. Third, as noted by Hetzel(l984), the estimated parameters in the conventional money demand regression equations may be subject to simultaneous equations bias, but they still depend in an essential way on the public's money demand. Stability overtime of an estimated money demand regression and the ability of such a regression to generate good out-of-sample predictions continue to be empirical evidence consistent with the hypothesis of a stable money demand function. Finally, the bulk of empirical studies (e.g., Goldfeld, 1973; Lieberman, 1977; and Rasche, 1986) reports that the simultaneous equations bias that results from estimating money demand relationships in a single equation framework is quite small. As suggested from the above review of the existing literature, most specification issues of the short-run money demand function remain unresolved. Further research is needed to identify the dynamics of a portfolio adjustment model more appropriately. 108 5.2. Empirical Estimates of the Short-Run Money Demand Function in Korea The issue of which scale and opportunity cost variables must be included in the money demand function is still the source of ongoing dispute. As seen earlier, the transactions model of money demand employs current real income as a proxy for transactions. This practice is quite standard even though GNP (or NNP) is an incomplete measure of transactions.76 However, the problem of choosing the appropriate opportunity cost variable is the most controversial issue both on the theoretical and empirical grounds. In the usual empirical studies, the direct cost and own-rate of return on money (e.g., "expected losses", "storage charges", and interest earned or "non-pecuniary service") are implicitly assumed zero usually because of inadequate data. The major indirect cost of holding money is the income forgone from the assets that could have been held instead of money. Accordingly, the opportunity cost depends on the alternative considered and on the holding period for which the substitute is regarded as sacrificed. Given numerous possible alternative assets, it is not surprising that the existing research, using one (or a small number of) observable yield(s), has shown diverse and confusing results.77 7‘ For example, Modigliani, Rasche and Cooper(l970), Goldfeld(l973, 1976), Lieberman(l977, 1980), and Laidler(1980) recognize that real GNP may be a poor proxy for transactions. They attempted to incorporate some measure of financial transactions, real final sales, bank.debits to demand deposits, and wealth variables or permanent income, but the studies in general concluded that little is gained by incorporating these variables. Moreover, it is quite difficult to obtain the appropriate wealth variable or permanent income. 77 The study of Heller and Kahn(1979) is an exception. Following Friedman's suggestion that "the whole term structure" of interest rates affects the demand for money and there is no a priori reason to regard a short or long rate as ”the" alternative cost of holding cash balances, 109 The studies have explored (i) short-term rates like the treasury bills, CPs, and savings deposits rates (ii) long-tenm rates on government or corporate bonds and (iii) some measure of anticipated inflation rate as a proxy for nominal yield on physical assets. Consider the case of Korean economy that has been characterized by financial repression, underdeveloped money and capital markets, pervasive price controls, and substantial inflation” ‘The financial assets primarily consist of short-term maturities of sixty-to-ninety day, except for a relatively small volume of corporate and government bonds that are usually transacted through the cartel agreements among the limited financial institutions. Since both short and long rates are highly controlled and tend to move together, there is no apparent distinction between the two rates.. At the same time, official interest rates and bank rates have slow moving trends for long periods with the exception of abrupt changes in interest regulations so nominal interest rates rarely reflect inflationary expectations. Hence it is safe to say that a plausible term structure of interest rates has not existed. Under these circumstances it is unlikely that we are able to find one (or a small set of) financial asset(s) that can be regarded as the closest substitute for money. It is inevitable that physical assets, such as they approximate whole term structure of interest rates by the quadratic function in maturity (i.e., lnR1-80+81t1+82t21, i-l...n) and use three parameters of the quadratic function in standard money demand functions. They conclude that "this approximation performed favorably relative to standard specifications of the money-demand function that utilized only one interest rate as the opportunity-cost variable as well as the ones that introduce several interest rates. Furthermore, the function...using this particular approximation appeared to be stable during a period when standard function ...display significant shifts in parameters."(p. 127) 110 land, buildings, and consumer capital, must be considered as alternatives to holding money although this treatment may be more or less inconsistent with the transactions view of money. Most money demand studies of developing countries and countries undergoing high inflation (e. g., Cagan, 1956; Frenkel, 1977; and Kahn, 1977) have generally used a weighted average of past rates of price changes as a proxy for the anticipated nominal yield on real assets. However, the use of the rate of changes in prices proves unsatisfactory since it fails to measure the real yield with nominal interest rates held constant. It is also well-known in Korea that the behavior of the prices recorded in official indexes has been distorted to a great extent by the effect of government's price controls.7 The recorded price indexes tend to understate the true price level and to overstate the level of real GNP as well as aggregate real money balances. On these grounds, Friedman and Schwartz(l982) argue for the use of the rate of change of nominal income rather than of prices as a proxy for the nominal yield on physical assets: "The rate of change of nominal income is the sum of the rate of change of prices and the rate of change of output, and the rate of change of output is an estimate, though a downward biased estimate, of the real yield. Hence the rate of change of nominal income can be regarded as a better proxy than the rate of change of prices alone for the total nominal yield on physical assets.”(p. 276) Our approach to the specification of money demand function begins with 7’ According to BOK's Sugey of Savings HQIKQE (1989, p. 153), it is revealed that more than 90% of the public does not believe the announced official price indexes (CPI and WPI) reflect the changes in prices appropriately. The survey also shows that the public judges actual inflation rates by increases in market prices of houses, rental cost of housing, and actual cost of living--the factors that are not well captured by the officially measured price indexes. 111 the nonnested composite model by using quarterly data for the full sample period l973/I-1989/II.’9 Instead of the double-log specification suggested by equation (5.7), the model estimated takes a semi-log specification in interest rates because, as seen later, the latter tends to perform better than the former. It also incorporates the rate of change of nominal GNP as an additional explanatory variable. The model then can be formulated: (5. 10) 1m-%+s,1ny,+szm+s,i,+s,1nm,-,+s,a,+351np,+e, where Tg-ln(Y;/qu) denotes the rate of change of nominal income, ag-ln(Pu/Pb1) denotes the rate of change of prices, and at is the random disturbance term. .As discussed in the preceding section, equation (5.10) enables us to test the statistical acceptability of two competing adjustment processes as well as the linear homogeneity hypothesis in price or real income. In the equation the dependent variable is the real MlB (the closest concept of transactions balances) using the GNP deflator (1980-100) as the price index to deflate nominal M18. The explanatory variables are real GNP, the GNP deflator, the yields on corporate bonds,80 the rate of change of nominal income, and the lagged dependent variable. As usual with model like equation (5.10), an initial estimate of the equation by 0LS indicated the presence of significant serial correlation 7°.A preliminary specification search started with the unconstrained distributed lags model suggested by equation (5.8). When estimated freely with several combinations of a plausible lag length or estimated with the imposition of relative low order polynomial restrictions in distributed lag patterns, the attempts failed to find any statistical regularity consistent over time. 3° The rates are most continually and consistently available data among various official rates. Although the corporate bond rates can be considered to be relatively market-determined, the yields are not much different from the rates on the short-term financial instruments such as commercial bills of sixty-to-ninety day for the reason mentioned earlier. 112 among the residuals. Consequently, the Cochrane-Orcutt (CORC) iterative procedure (the technique commonly implemented to estimate money demand when quarterly data are used) is employed for the correction of first- order serial correlation in the error process.81 The regression results for the unrestricted (nonnested composite) model and the restricted. models implied 'by ‘various hypotheses are presented in Appendix C1. In a search for appropriate specifications of the money demand function, the likelihood ratio (IR) tests are performed first. Table 5.1 reports the LR test statistics of the restrictions specified in the first and third columns of the table.82 The test uniformly rejects the restriction of RSAH. The probability of making an error in rejecting the restriction (type I error) is practically zero. The rejection of RSAH does not depend on the presence ’1 Econometric theory indicates that the coexistence of a lagged dependent variable and serially correlated disturbance terms leads to coefficient estimates that are inefficient but unbiased (see Theil, 1971). The CORC estimates thus may iterate to a local rather than a global minimum of the sum-of-squared residuals. A number of recent studies (Lieberman, 1980; Hafer and Hein, 1980; and Blinder, 1986) have used more sophisticated serial correlation correction procedures, such as maximum likelihood(ML), Hi1dreth—Lu(HILU) grid search, and Hatanaka's residual Adjusted Aitken technique, to obtain "efficient and consistent” estimates. However, Offenbacher's (1981) comparative study of alternative estimators on a ”standard" money demand equation shows that for various sample periods and specifications, the CORC, ML, and.HILU estimates are virtually identical, and that the CORC and ML estimates are more reasonable than the results from Hatanaka's procedures in terms of both econometric theory and intuition about the length of the money demand adjustment lag. ’2 LR test statistics are calculated by -2(lnLr-lnLu) that is asymptotically distributed xgq,‘where lnLr represents the log-likelihood value of restricted specification, lnLu represents the log-likelihood value of unrestricted specification, and q indicates the number of restriction. It should be noted that the log-likelihood values are not strictly comparable because they correspond to different autocorrelation coefficients across alternative specifications. 113 Table 5.1 Likelihood Ratio Test Statistics Test of x2 Conditional x2 Statistics Test of Statistics B‘fB5-O 0.74 85 -0/ Egtfis -0 0.02 85 -0 43 .42** 81+B. -1/ 8,485 -0 l . 86 £3 -0 0 . 01 814-8, -l/B,+85-0 , 86-0 6 . 80** 814-8., -1 2 . 32 8.4-85 -0, 86-0 0.76 85 -0/ 85-0 0.08 85 -0 , 85 -0 43 . 50** 81+B, -l/ 85 -0 3 . 96* Bfi-Bs -0 , 86 -0 , 81+8r0 7 . 56 81+B, -l/85-0 , 36-0 10 . 46** BS -0 , £6 -0 , £1+£‘-0 53 . 96** **Significant at the 1% level. *Significant at the 5% level. of the a.priori restriction on price or income elasticities. 0n.the other hand, there is no evidence to reject NASH regardless of the a priori restriction of price or income elasticities. The size of the type I error is also very large. In addition, the tests uniformly fail to reject the zero price elasticity restriction. The long-run unitary income elasticity is not rejected at the 5% level of significance when no adjustment restrictions are imposed. However, under the accepted restrictions of both NASH and zero price elasticity, unitary income elasticity is rejected at the 1% level. Since the constrained estimates are more efficient if the constraints are true, it is reasonable to conclude that the long-run income elasticity is not unity. The hypothesis of the long-run unitary income elasticity is also rejected.under the restrictions of RSAH and zero price elasticity but with a much lower marginal significance level than under NSAH. The above LR test results confirm the previous argument that the inappropriate restriction in RSAH can be the source of considerable 114 parameter instability for the countries suffering from highly volatile and substantial inflation. Indeed, the standard error of regression (SEE) decreases by 28% under the specification of NSAH and.zero price elasticity compared to the specification of RSAH and zero price elasticity (see Appendix Cl). Since both NSAH and zero price elasticity hypotheses are not statistically rejected by the sample data, the subsequent statistical tests are based. on. the following level and. first-difference specifications. First, the level equation is specified: (5. 11) 1nm,-s,+s,1ny,+sza,+s,§,+s,1n(M,-,/p,)+e, where ct is assumed to follow the first-order autocorrelation, i.e., \/"’et-pct-1+nt, where p is a constant and at is a white noise term with classical properties. Taking first-difference of equation (5.11) gives: (5. 12) Alnmt-B' ,A1ny,+s' ,An,+s' filly-8' ,A1n(n,_,/p,)+q, Although the log level equation (5.11) is more general than the log first- difference (rate of change) equation (5.12), estimation of the latter not only avoids an important econometric problem related to the estimation of the former, but also provides some useful information on the nature of structural change. The difference between the two alternative specifications lies in the a priori assumption about the error structure. The two specifications are empirically equivalent if the autocorrelation coefficient is restricted to unity in equation (5.11). As seen in footnote 81, the presence of a lagged dependent variable and serially correlated errors renders the CORC estimates inconsistent and inefficient. If the disturbances in equation (5.12) are serially uncorrelated, estimation using a simple OLS technique 115 will avoid the unresolved econometric trouble associated with the serial correlation correction procedure in equation (5.11).83 The first-difference specification‘also can.serve to locate the likely points of an intercept shift or to provide evidence as to whether the underlying slope coefficients have changed. For example, Hafer and Hein(l980, 1981) argue that "a once-and-for-all intercept shift" in equation (5.11) will appear as a ”one-time increment in the disturbance pattern” of equation (5.12), and that if ”the marginal relationships" embodied in equation (5.11) have changed, similar changes will be exhibited in the slope coefficients in equation (5.12).8‘ We can expect, of course, that the coefficient of determination of the rate of change equation will be much lower than that of the level equation, especially given the strong secular trend in our variables. This is because the random disturbances effect plays a much larger role in the first-difference specification compared ‘with the systematic component effect. The regression results for both level (5.11) and first-difference (5.12) specifications are presented in Table 5.2. First, consider the results for the level specification. The overall explanatory power is quite high; the Durbin-Watson statistic indicates that the problem of first-order serial correlation in the residuals is removed; moreover, each 33 Taking first-difference in general has been suggested as a means of converting non-stationary stochastic process into stationary one and thus reducing the possibility of a ”spurious" regression result. See Granger andflNewbold(1974), Williams (1978), and.Plosser and.Schwert(l978). 8‘ For more complete explanations of this derivation, see Hafer and Hein(l98l) and for a critics of this argument, see 0ffenbacher(l98l). 116 Table 5.2 Quarterly MlB Demand Functions (1973/I-1989/II) Level‘ First-Differenceb Dependent Variable lnmt Alnmt Constant -1.396(6.20)c lnyt 0.259(7.54) Alnyt 0.290(9.77)c 3t -0.422(2.77) AK, -0.545(2.l6) Yt -0.l70(8.43) AYt -0.l76(10.27) 1n(Mtq/Pt) 0.787(28.46) Aln(Mbq/Pt) 0.544(7.50) dRZ 0.997 0.744 SEExlO 0.297 0.321 D-W 2.130 2.304 rho 0.417(3.61) ‘Level equation is estimated using the CORC procedure for the correction of serial correlation. bFirst-difference equation is estimated using simple 0L8. No constant term appears since the estimated coefficient of first-difference of a time trend appeared virtually zero (see also ”time trend", pp. 128-29). cThe numbers in parentheses are absolute value of t statistics. dRz is the adjusted coefficient of determination; SEE is the standard error of the estimated equation; D-W is the Durbin-Watson statistic; rho is the estimate of the serial correlation coefficient. variable's estimated coefficient is signed correctly and significantly different from zero at the 1% level. The estimated speed of adjustment (1-8,) is about 21% per quarter.85 While the quarterly speed of adjustment is faster than the 0-10% estimates that some researchers have reported for other countries, it seems still "implausibly slow", judged from the intuition that money market adjusts relatively quickly or the usual assumption of the long-run equilibrium moneywdemand literature that money market_clears well within one_year. As seen shortly, this is largely due to the basic econometric problem generated by serial correlation in the residuals. 85 Goodfriend(l985), from a critical point of view, interprets that the estimated coefficient of lagged dependent variable largely reflects the effects of ”measurement error" rather than speed of adjustment. 117 The estimated long-run income elasticity [bl/(1490] is 1.22. The income elasticity greater than.one is statistically significant, i.e., the earlier LR test rejects the long-run unitary income elasticity (Lia-[91) at the 1% level as does the conventional F ratio test at the 5% level. The result is consistent with Friedman's finding that "money is a '1uxury' rather than a 'necessity'”.86 Although the result may simply reflect that both 81 and 8, are biased upward due to measurement errors and to the basic estimation. problem, it also can 'be attributed to the lack of an appropriate substitute for money and to the precautionary money demand related.to great‘uncertainty'about the overall socio-economic environment. The estimated long-run interest elasticity evaluated at the mean value of interest rater-[éz/(l-bo] multiplied by the average rate-- is 0.37 in absolute value. However, 82 is much less precisely estimated with its absolute t value the smallest of all the estimated coefficients. As seen in the next section on stability tests, the range between lower and upper limits of 82 is so large that the interest rate parameter tends to exhibit instability over time. On the other hand, the estimated interest elasticity seems to be relatively large compared to the value implied by the transactions money demand theory. This presumably reflects an influence of the downward biases in interest rates and prices. The estimated long-run elasticity with respect to the rate of change of nominal income--[8&/(l-8,)] multiplied by the mean value of annualized 1,;- is about 0.14 in absolute value. The point estimate of the elasticity for the rate of change of nominal income is much smaller than 5° See Friedman(l959) and Friedman and Schwartz(1963, 1982). 118 for the interest rate. However, the t value for 83 is three times higher than for 82. Two alternative interpretations are possible. Suppose Tb represents a good proxy for the "total nominal yield" on physical assets. The physical (real) assets are as a poorer substitute for money than liquid financial (nominal) assets. Then any change in money holdings at the margin is largely as a substitute for financial assets. The implication is consistent with the transactions view of money demand or the standard Keynsian liquidity preference theory. An alternative interpretation is that the interest rate is a better measure of the relevant nominal yield on financial assets than the rate of change of nominal GNP is of the relevant nominal yield on physical assets. The latter is an indirect measure of the yield, while the former is a direct measure in the contemporaneous money market. If both nominal yields on financial and physical assets tend to move together as in the case for short and long rates, the use of either rate serves as a proxy for the other, avoiding the problem of multicollinearity between the two rates. However, this is not true. The simple correlation coefficient between R, and {It is just 0.014. Therefore, the inclusion of T, introduces an effect independent of the nominal interest rate. Next, consider the results for the first-difference specification.87 87 In comparing the results for level and first-difference regressions, one should be cautioned against using the reported R? or SEE as a basis to judge goodness-of-fit across equations. Granger and Newbold(l974), for example "Spurious Regressions", show that when the dependent and independent variables follow a random walk, a non-zero R2 will be expected even if no relationship between those variables actually exists. When the equation is estimated in first-difference form, the variables no longer are nonstationary and thus R? is expected to be zero. In addition, it is hard to compare directly the variability of rates of 119 The adjusted coefficient of determination is, as expected, much lower than that of the level specification. There is no evidence to reject the hypothesis of serially independent residuals, i.e., the Durbin-Watson statistic reveals little problem with any first-order serial correlation in the error terms. Each variable's estimated coefficient has the expected sign and achieves statistical significance at the 1% level with the exception of the interest rate coefficient (significant at the 5% level). The estimated short-run elasticities are quite similar to those found in the level form. However, it is surprising that the estimated coefficient on the lagged dependent variable is much smaller and thus the speed of partial adjustment becomes substantially faster. Specifically, the quarterly speed of adjustment increases from 21% in the level version to 46% in the rate of change version. The difference indicates that the estimated long-run elasticity of each independent variable is half as large in the rate of change version compared to the level version, even though the corresponding short-run elasticity is very close in both specifications. Thus, the result for the first-difference specification seems to support the notion of economies to scale in money holdings. This directly conflicts with the long-run income elasticity of 1.22 in the level specification. This conflict may seem somewhat puzzling. The puzzle can be solved by recognizing that the usual empirical change with that of levels because each SEE is in different units respectively based on the different units of dependent variables across regressions. For rates of change, there is little point in expressing variations relative to the mean since its mean value may be zero or negative. 120 stock-adjustment models are subject to a fundamental "identification" problem. In practice, the puzzle is considered as a statistical artifact. The regression of level equation yields estimates of relatively low serial correlation and implausibly slow adjustment coefficient, whereas the regression of first-difference equation produces estimate of relatively fast adjustment coefficient with the a priori restriction of autocorrelation coefficient to unity. Thus the result may reflect more than one local minimum in the sum-of-squared residuals function, although we have little ability to pin down a global minimum of the short-run money demand function. It is virtually impossible to separate the speed of adjustment from autocorrelation and to identify the speed of adjustment correctly . 88 At this point, it may be argued that the basic money demand models are incorrectly specified since they do not take appropriate account of various important factors which may influence the behavior of money demand. In the remaining part of this section, the specification problems that afflict the money demand function are investigated further. Unless 3° Econometric theory has not yet developed to deal with such a basic problem inherent in dynamic models. Griliches(l967) pointed out that any estimation technique will have trouble identifying between a model with high serial correlation and fast adjustment and one with low serial correlation and low adjustment. Recently, Blinder(l986) shows that most stock-adjustment models have "two local minimum” in the sum of squared residuals function, and that the usual serial correlation correction technique, including the CORC procedure, typically picks out a "wrong” local rather than the global minimum. Granger and Newbold(l974) criticize further that any equation subject to serial correlation is "misspecified" so the coefficient estimates are unreliable. Finally, Rasche(1986) reaches the conclusion that "given the problem... , it is virtually impossible to identify the dynamics of a portfolio adjustment model in the absence of some a priori specification of the autocorrelation process of the error structure.”(p. 27) 121 otherwise mentioned, the subsequent tests are applied to the first- difference equation, M1 and M2 demand functions,89 and the truncated sample l973/I-l982/IV regression. WW As mentioned in footnote 70, the constant interest elasticity specification is more appropriate than the constant semi-log slope specification if relative (percent) rather than absolute (percent point) changes in interest rates are what matters for the demand for money. To examine this point, we re-estimated the money demand function by including the logarithms of interest rates instead of the interest rates themselves. The results for both level and first-difference specifications are given in Table 5.3. All the corresponding parameter estimates including the summary statistics are virtually identical to those in Table 5.2 except for the coefficients of the interest rate and the intercept. In the double-log specification, the short-run interest elasticities are 0.086 (level) and 0.105 (first-difference) in absolute‘valuew The corresponding values are 0.080 and 0.102 evaluated at the average rate in the semi-log specification. As a result, there is little difference between the two - 33931082195. ’9 Of course, one may argue that it is inappropriate to specify the same demand functions for both narrow and broad money because the broad money demand function would not be identical to the narrow money demand function. However, it is very rare in practice to specify the broad money demand function differently from the narrow money demand function. This is largely due to the practical difficulties in obtaining appropriate data, such as wealth variable or permanent income and own-rate of return, that are exactly the same problem which plagues the narrow money demand function as well. 122 Table 5.3 Double-Log Specification of Quarterly MlB Demand Functions (l973/I-l989/II)‘ Level First-difference Dependent Variable lnmt Alnmt Constant -l.214(5.ll) lnyt 0.260(7.50) Alnyt 0.291(9.68) lnRt -0.086(2.68) AlnRt -0.105(1.96) Yt -0.l70(8.38) AYt -0.l75(10.18) ln(Mtq/Pt) 0.782(27.18) Aln(Mtq/Pt) 0.540(7.38) R2 0.997 0.741 SEExlO 0.298 0.324 D-W 2.128 2.295 rho 0.417(3.6l) 'All remarks are the same as Table 5.2. However, as far as forecasting ability is concerned, the semi-log specification appears to be somewhat better than the double—log specification. When the money demand of the post-l982/IV is simulated from the model of the pre-l982/IV sample, the former improves the "static" root-mean-square-error (RMSE) by 7.5% for the level form and 4.9% for the rate of change form compared to the latter.90 This result suggests that the semi-log specification in interest rates is more appropriate than the doubleelog specification. .,~ 5 ent va ab e mat The basic level equation was re-estimated using the instrumental variable (IV) technique to examine whether it is properly identified as the demand equation or it is subject to a serious simultaneous equations 9° The forecasting performance is more fully discussed in the next section. 123 bias. As mentioned in the preceding section, the simultaneous equations bias causes the estimated equation to represent some unspecified hybrid of the demand and supply relations. The difficult problem here is to construct the appropriate instrumental variables which should not be correlated with the disturbance term of the money demand equation but should be highly correlated with variables that appear on the RHS of the money demand equation. The choice of such variables is almost impossible without agreement about the true structural model of the entire economy. Following the suggestion by Fair(1970) , the IVs employed in the first- stage include yt, Rt. (Mt-1/Pt) , Ttneach lagged once; the curb market rate, the stock of bank credit, and the monetary base --each current and lagged one period; '1" and xv-each four quarters lagged to capture seasonality; and finally, the constant term and a time trend."1 The IV estimation does not change any estimate of the parameters in a meaningful way (see Appendix C2). The IV estimates are virtually the same as those of the basic model. This may simply reflect the difficulty in constructing appropriate instrumental variables. 0n the other hand, it may suggest that the simultaneous equations bias is not likely to cause a serious problem in the estimation of the short-run money demand function within a single equation framework. However, some cautions about these conclusions are needed. There is a question about the appropriateness of the IVs used here, and also the sample size is not so sufficiently large 91 In an attempt to get more efficient and consistent coefficient estimates, many instruments were experimented with several plausible combinations of lagged values adding other exogenous variables. The equation with the lowest standard error is reported as appendix C2. But other results were also not much different from the reported result. 124 as to appeal to the asymptotic properties of the IV estimator. e iv o tu co v s The effects of including alternative costs of holding money were investigated. First, when the curb market rate (RCURB) is included instead of the corporate bond rate,92 no significant change is found in the estimates except for the interest rate parameter (see Appendix C3.a). The estimated coefficient of the curb market rate is much smaller in absolute value than that of the corporate bond rate. The smaller curb rate sensitivity simply reflects the effect of the spread between the two rates. The higher average curb rate produces the corresponding smaller semi-log slope coefficient when the elasticities with respect to the two rates are equal. It should be stressed again that the curb rate is an indirect measure (and hence unreliable), and that it largely reflects a risk premium rather than the opportunity cost of holding money or a direct influence of monetary policy (see Chapter 11). When both corporate bond and curb market rates are included, the t statistics for the two coefficients are far below any acceptable level of significance although they are correctly signed (see Appendix C3.b). This result suggests that the alternative rate has no additional explanatory power once either rate is accounted for. It may also reflect that the two rates are highly collinear and that the sample size is too small to measure the effects of the two rates with any precision. ’2 When the government bond rate was used or the corporate bond rate was replaced with the CP rate available after l983/I, the results were similar to the basic model. 125 In addition, tests were performed to examine whether the own-rates of return on money are an important factor in explaining the behavior of demand for real balances. Because the official bank deposit rates have been strictly regulated at very low levels, incorporating such own-rates of return into the money demand function is not likely to have any significant effect. The experiments with the bank deposit rates (3-6 months or one year) on households demand or savings deposits (RSD) conform to the intuition (see Appendix C4.a and C4.b). As a result, the conjecture that small investors who do not have sufficient funds to invest in money market assets may be sensitive to the bank deposit rates is not supported by the data. Usually, these people take advantage of the privately organized rotating credit association (KYE) as an effective way to hold small amounts of savings. This study followed.K1ein's(1974) device for computing an own-rate of return on money by assuming that currency yields a zero nominal yield and that banks are forced to pass on to their depositors the competitive market interest rate including both direct and indirect payments. The own-rate on money is calculated as a weightgg_§zg;ggg of zero (return on currency and bank reserves) and the market interest (corporate bond) rate (return on the rest of deposits) with the weights being H/MlB (ratio of the monetary base to MlB) and 1-(H/M1B) respectively. That is, RT-0(H/MlB)+R{1-(H/MlB)}.93 The result of including this approximation.of the own-rate (RT), in addition to the corporate bond rate, shows that although the measure of the own-rate takes on a positive sign as expected, 93 For detailed explanations and statistical validity of Klein's device, see Friedman and Schwartz(l982, pp. 260-71). 126 it remains statistically insignificant at any reasonable level (see Appendix C5.a). 0n the other hand, the absolute value of coefficient on the corporate bond rate is somewhat increased compared to the case of the corporate bond rate alone.“ We cannot tell whether the result primarily reflects the “spurious correlation” rather than the "real effects of economic forces" without better measures of the own rate. Theory suggests that the relevant opportunity cost of holding extra money balances is the spread between the return.on.close substitute assets and the own rate. Thus we included the spread in yields (R-RT) in the money demand function. The result shows that the point estimate of the spread is again the same as the value of coefficient for the corporate bond rate when RT is included separately (see Appendix C5.b). This finding argues somewhat in favor of the real effects for the increase in the interest sensitivity of money demand rather than entirely a spurious statistical correlation. Nonetheless, the effect of substituting the spread. for the corporate 'bond rate alone is relatively’ small: the difference in the two slope coefficients is within one standard error of the estimated coefficient; and the fit of the regression is not improved at all. 9" Carson and Frew(l980) point out that the improvement in Klein's fit may be spurious, reflecting errors of measurement in M common to m and B/M. Such common errors of measurement tend to introduce "spurious elements” into the computed coefficients of R and RT, contributing to a higher absolute value of the coefficient of R. They also note that the ”economic forces” [the differential rate (R-RT); the relevant alternative cost of holding money] would work in the same direction as the spurious correlation. 127 a v t e t on So far as the anticipated or the unanticipated rate of inflation goes, there exists little room for either rate to enter the basic models individually. The nominal stock—adjustment mechanism implicitly reflects the current rate of inflation (officially measured or distorted rather than actual rate) through the lagged dependent variable term (Mt-l/Pt) . In addition, the rate of change of nominal income (1}) can be regarded to take on the anticipated rate of inflation, serving as a proxy for the nominal yield on physical assets. Consequently, the effect of unanticipated inflation is indirectly captured in the models. The results of experiments for some measure of anticipated or unanticipated inflation rate shows no significant effects in the basic model (see Appendix C6.a and C6.b).95 The finding that such a measure of inflation rate is significant in a real stock adjustment type model is evidence of misspecification since the basic nominal stock adjustment version outperforms the alternative model. Disequilibrium The experiments were performed to test whether an additional variable is needed for a temporary disequilibrium or buffer stock money. Following 95 In order to obtain an appropriate measure of the expected rate of inflation («'t), a number of ARIMA models (relatively low order) of the inflation rate (It) were experimented. But the results were unsatisfactory producing insignificant estimated coefficients. So the four quarters lagged official inflation rate (R,,) was considered as the expected inflation rate for the current quarter. The simple correlation between at and at-.. is 0.84. This high correlation may be primarily due‘to the seasonality inherent in the Korean economy along with the tendency of accommodating (passive) monetary policy to that seasonality. 128 Judd and Scadding, we added the rate of change of the bank credit (BC) or the total domestic credit (TDC) although these variables introduce the risk of estimating a reduced-form equation rather than the money demand function. The results show no significant role of such variables in the basic model, refuting the argument that observed real balances are off the short-run money demand curve (see Appendix C7.a and C7.b) .96 W Finally, we introduced a time trend variable (TIME) as an additional explanatory variable to examine whether the slow moving structural or institutional factors, including improvements in cash management techniques, have a secular trend influence on the demand for real balances. The time trend variable, of course, cannot distinguish an individual effect of a specific event from the overall effect of a number of possibly offsetting factors, all of which are not readily measurable. It only provides a crude net effect of the long-run trend. The time trend _ coeffigigngwis afound.,..to--be- -insignificant_,_(see ”Appendix” ”98.5),“ One exception is observed fornglevelequatipon, In this case, the coefficient ?£,£h§.-.-§i!!§1jrend _is significant at the 1% level for the full sample regression (though marginally significant for the truncated pre-1982/IV sample regression). It has a negative value of 0.30% to 0.35% at a quarterly rate or about 1.5% at an annual rate (see Appendix C8.b and 9° For M1, the coefficient on the rate of change of the bank credit was found marginally significant for the full-sample period. But it was not robust to changes in the sample periods (see Appendix C7.c and C7.d). 129 C8.c). The result largely reflects the effect that the demand for currency has gradually declined with the rapid and continual expansion of bank branches as well as with the automatic depositing of salaries through overall-households-deposits account. However, the magnitude of the coefficient is so small, relative to that of the intercept or the SEE, that the coefficient for the first-difference of TIME turns out to be insignificant in the M1 rate of change equation (see Appendix C8.d)."7 ___’~ ’7 Since the first-difference of a time trend variable is just a column of ones, the estimated constant in the first-difference equation represents the coefficient of the first-difference of TIME. 130 5.3. Stability Test of the Short-Run Money Demand Function The full-sample estimates of the quarterly money demand function in level and first-differences suggest that our basic models are quite satisfactory, judging by standard significance tests and hypothesized values of structural parameters. However, the question of whether or not the money demand function has been stable over time is another issue. Although the term stability (or instability) generally refers to the statistical finding that the parameter estimates remain constant (or change) across differing economic environment, it is not so precisely defined as to distinguish the different types of instabilities that can plague time-series estimates. Consequently, the available stability tests are difficult to interpret.98 The departures from constancy of parameters may show up in different ways and the various tests may not be equally powerful in detecting a particular kind of instability encountered. Here we employ several stability tests to address the issue of money demand stability from different perspectives. 5.3.1. Cusum-Squares Test The "cusum-squares" test formulated by Brown, et.a1.(l975) is applied first. The test basically examines whether the squared one-period-ahead prediction errors from a set of "recursive" regressions cumulate at an approximately constant rate. If at a point in time the calculated value of the cusum-squares of the recursive residuals is greater than the 95 For an elaborate explanation of the concept of stability and several different types of instabilities, see Boughton (1981). 131 Table 5.4 Cusum-Squares Test Results Level First-Difference MlB 0.100 (0.106) 0.092 (0.058) Ml 0.172*(0.l45) 0.193*(0.169) M2 0.136 (0.103) 0.156 (0.148) #The numbers in parentheses represent backward cusum-squares statistics. *Significant at the 10% level: the critical values are 0.172 and 0.199 (for the 5% level). critical value at a. given level of significance, then the null hypothesis that the estimated. relationship is stable can. be rejected. at that significance level.99 The test statistics also can be plotted against time along with parallel sets of significance lines which provide the statistical "boundaries" used to indicate the timing of a possible structural shift. One advantageous feature of the cusum-squares test is ‘” The cusum-squares statistic is calculated by the formula: 1' ‘1' s,- 2 wZ, / z w, , r-k+1...T t-k+1 t-k+1 where k is the number of regressors including the constant, T is the last sample point and wgt represents the squared one-period-ahead prediction error for point t based on the regression truncated at point t-l. The test requires to run recursive regressions beginning with initial (terminal) k observations for the forward (backward) test and adding one observation each time until the end (beginning) of the sample is reached. Under the Ho of a stable function (i.e., a joint hypothesis of structural stability and stability of the error process), E(Sr)-(r-k)/(T-k). By the formula, Sr-O when r-k, and Sr-l when r-T. The test then consists of a comparison between the absolute value of Sr-E(Sr) and the predetermined value for a given significance level. An alternative test using the recursive residuals is the "cusum” test: wr-(l/a)2 wt, where a is the standard error of regression. Under the Ho, E(w})-0. Since it is known that the cusum test is less powerful relative to the cusum-squares test, the test is not pursued here. For more details, see Brown, Durbin and Evans(1975), and Johnston(l984), and for a critical evaluation of the power of these tests, see Garbade(1977). Finally, the application of the cusum-squares test to the money demand function is found in Heller and Kahn(l979), Boughton(l979, 1981), and Hafer and Hein(l979). 132 that unlike the conventional Chow(l960) test, the technique does not require prior specification of the likely break point. The results for the cusum-squares tests are presented in table 5.4.100 The test statistics for MlB uniformly reject the hypothesis of structural change for the MlB demand relationship. This is also true for M2, even though the test statistics exhibit a much larger marginal significance than those for MlB. By contrast, the forward cusum squares of M1 in both level and first-difference models reject the hypothesis of no structural change at the 10% level of significance, though the backward cusum squares fail to reject the null hypothesis. However, this is not the whole story underlying the cusum squares test. The implications become more apparent upon a close examination of the time series of the test statistics as plotted in Figure 5.1-5.3. The forward cusum squares statistics are plotted against time for each monetary aggregate. In addition to the plot of 8,, each figure plots the mean values of Sr and three confidence lines which are drawn parallel to 10° Since the cusum-squares test is derived on the assumption that errors are serially independent and the variance of errors are equal, the statistics for level equations are calculated from the recursive regressions of the transformed data x,-x,-bx,_,. That is, the autocorrelation coefficient for the level equation is assumed to be constant throughout the whole sample period with the value of the full- sample estimate. Alternatively, the test statistics were calculated by estimating the serial coefficient in each alternative sample period separately using the CORC procedure. The resulting values of test statistics were almost the same as those reported. 133 Figure 5.1 M18 Forward Cusum-Squares a) Level %’/ :0“ IJS *lZS Lee 74 75 '75 '77 '79 '79 '59 '91 '92 '93 '94 '95 ’95 '97 '99" 025‘ b) First-Difference 1 4 J v’7 '74 '75 '75 '77 '79 '79 '99"91 '92 '93 '94 ‘95 '95 '97 '99 Figure 5.2 M1 Forward Cusum-Squares b) First-Difference a) Level I." usJ l2§ L99 l754 lSB 815 8.054 '0. lb 74 '75 '75 '77 '79 '79 '99 '94 '92 '93 '94f'95 '95 '97 '59{' ' 74 '75 '75 '77 '79 '79 '99 '94 '97 '93 '94 '95 '95 97 99 a Figure 5.3 M2 Forward Cusum-Squares a) Level 74 '75 '75 '77 '79 '79 '99 '91 '92 '93 '94 '55 '95"97 '99" -..25 I ' 1 f1 WI I I V U 'I 5 5 ‘ 74 '75 75 77 79 79 '95 91 92 93 94 '95 95 97 99 b) First-Difference ____ 134 the mean values line for given levels of significance. When the plot of Sr crosses one of these boundaries, the hypothesis of stability is rejected at the corresponding significance level. The figures thus reveal a varied picture of the timing of a possible structural breakdown. As observed in Figure 5.1, the S, plots for both specifications of the MlB demand function never intersect the statistical boundary. Furthermore, all the 8,. plots W except for 1975/1 (level) probably assciated with "oil shock". In Figure 5.2, the Sr plot crosses the 10% confidence boundary in 1981/II (level) and 1981/1-1981/11 and l982/I-1982/II (first-difference). In Figure 5.3, the 8, plots do not cross the 10% confidence band. More importantly, the characteristic feature of Figure 5.2 and 5.3 compared to Figure 5.1 is that both M1 and M2 cusum-squares test statistics WW mg.1°1 This feature suggests that M1 and M2 demand functions may be unstable, even though their test statistics do not reject the stability hypothesis. 5.3.2. Chow and LR test The most commonly used test of structural change is the standard ”analysis-of-covariance" (ANOCOVA) test and the "predictive” test, both of which were suggested by Chow(l960).102 Alternatively, the likelihood 1‘” The plots of backward cusum squares also showed similar patterns of the drifts. 102 The test statistics are calculated by the formula: (RSSr-RSSu)/Tz predictive test statistic - ~F(Tz, Tl-k) RSSu/(Tl-k) 135 ratio (LR) test is asymptotically equivalent to the ANOCOVA test.103 These tests are appropriate for the instability hypothesis that "a function could be stable up to some point, experience a shift in one or more parameters at that point, and then be stable again thereafter". (Boughton, 1981, p. 582) Unfortunately, these tests rely on an a priori hypothesis of the location of the break point. The timing of a possible structural shift can be chosen with regard to historical events such as a substantial change in the financial systems. Alternatively the tests can be applied by locating a likely date of the shift on a more objective and statistical basis. First of all, the cusum-squares test results may serve to detect a possible breakpoint as discussed previously. Alternatively, the time series on Quandt's(l960) log-likelihood ratio can be computed. The point where this ratio takes a (global) minimum value can be considered as the (RSSr-RSS'u)/k ANOCOVA test statistic - ~F(k, T-k) RSS" where RSS, is the residual sum of squares from all T observations regression, RSSu from first T1 observations, and RSS"“ is the sum of residual sum of squares from first T1 observations and last T2 observations, and k is the number of regressors. Although the predictive test is useful especially when the ANOCOVA test is not applicable with the last sample being undersized (Tzk. Moreover, Wilson(1978) shows that even in the case of T2>k, the predictive test has a better power against the Ho than the ANOCOVA test. “’3 The LR test statistic is calculated by the formula: m— -2(1nLr-lnLu)-x2q where lnLr is the log-likelihood value from the full-sample regression and lnLu is the sum of the log-likelihood value from first T1 observations and last T2 observations regressions. 136 Figure 5.4 Quandt's Log-Likelihood Ratio (Forward) a) Level " 15 ‘ I l I ‘le 1 l I I I L4 ‘25 74 '75 '75 '77"79 '79 ‘99 '91 '92 '93 '94 '95 '95 '97 b) First-Difference 22.9.? ‘. ... ’3: G D “1&81 -12'5 I I I I I I 'l 'l . I . 74 75 75 77 79 75 '99 91 92 93 94 95 95 97 MIA ooooooonl ....... M2 137 Figure_5.5 Residuals of MlB First-Difference Regression IJII W ~ ‘ 1"" I ..ou' -0Jfl5 ankfinnmmuunumsnma Figure 5.6 Residuals of M1 First-Difference Regression 2:1 . A .- "W U1, 1mm -ll5 l *0“ I l I I V I ' I '1' I I l I I U nununmnmuuuusunma Figure 5.7 Residuals of M2 First-Difference Regression I I I N .on‘ .0“) IM A l' 525; I ~ I mtflh '13! j I nhfihnfihhhhhhilhhh #Dotted and dashed lines represent i1 SEE and :2 SEE of the respective regressions. 138 most likely break point in the estimated relationship (see Figure 5.4).““ Finally, we examine the residuals of first-difference equations for large residual ”outliers” (exceeding two SEE) that may indicate the timing of a possible structural shift in the constant of a levels equation (see Figure 5.5-5.7). On the basis of the above information, we identify and select the most likely timing of a structural change as late l9825m5 Historically the period coincides with the financially epoch making point, as mentioned in the preceding section on financial changes in Korea. After the big financial scandal of May 1982, the illegal WANMAE market began to flourish and at the same time with more rapid expansion of the STFCs, there was an unprecedented increase in.the size of the official commercial bills market around this period. The period also roughly marks the turning point from a high inflationary to a low inflationary economy. Once a possible breakpoint is determined, the conventional F test and ‘LR test are performed using the split at 1982/IV to assess the statistical um Quandt's log-likelihood ratio is calculated by the formula: Qr-(lnLr-lnLu) where lnLr is the log-likelihood value of full-sample regression and lnLu is the sum of log-likelihood value of first r observations and last (T-r) observations regressions. However, the significance test cannot be conducted since the distribution of Qr is unknown under the Ho. nu Quandt's log-likelihood ratio statistics indicate early 1987 as the most likely break point by reaching global minimum values uniformly (except for M2 first-difference equation) around that period. If the split at 1987/I or 1987/11 is used for the standard stability test procedure, the effective test: is impossible 'because of' insufficient degrees of freedom left in the last observations. Anyway, both ANOCOVA and predictive tests using the split at 1987/I or 1987/II failed to reject the stability hypothesis for MlB at any acceptable significance level. 139 significance of the change. The outcomes are reported in Table 5.5.106 For MlB, the test statistics reject the hypothesis of structural change while LR test (level and first-difference) and ANOCOVA test (first- difference) show marginal significance at the 5% level. By contrast, the structural stability hypothesis for M1 and.M2 is rejected at the 1% level of significance except for the predictive testsf107 Thus the evidence presented in 'Table 5.5 suggests that the MlB demand. function is reasonably stable over time even though both the M1 and M2 functions experienced.some sort of (statistical) instability’beginning in late 1982. Table 5.5 Chow and LR Test Results Level First-Difference Predictive ANOCOVA LR Predictive ANOCOVA LR MlB 1.31 1.98 11.42* 1.23 2.83* 11.86* Ml 1.30 4.04** 21.44** 1.49 4.44** 17.68** M2 1.69 4.53** 22.46** 1.71 5.11** l9.92** d.o.f(26,35) (5,56) (5) (26,36) (4,58) (4) **Significant at the 1% level. *Significant at the 5% level. nm The presence of serial correlation in the level equations again casts doubt on appropriateness of the test procedures. Here the statistics are calculated by estimating the autocorrelation coefficient in each alternative sample period separately using the CORC procedure. First-order autocorrelation coefficients are quite close, for example, 0.417 (full-sample) and 0.421 (first sub-sample) in the case of MlB. An alternative "seemingly unrelated estimation" technique constrained the serial correlation coefficient to be constant in each of the respective sample periods. The results for the alternative procedures were almost the same as those reported in Table 5.6. 137 Contrary to the suggestion by Wilson(1978), the predictive test results appear to be less powerful against the Ho than the ANOCOVA test. When the double-log specifications are employed, the resulting predictive test statistics reject the stability hypothesis at the 5% level for M2 level equation as well as for M1 and M2 first-difference equations. 140 This finding is consistent with the cusum-squares test results reported previously. 5.3.3. Recursive Regression and Simulation The preceding stability tests are concerned with whether the overall relationships between real money balances, real income, and rates of return on financial and physical assets, have been stable throughout the whole sample period. None of the tests provides information on which individual coefficients are unstable. It is also important to examine how accurately the truncated-sample regressions are able to forecast recent out-of-sample money demand. To be more specific in an investigation of the temporal stability of the money demand relationships, we begin estimation with the truncated- sample of the pre-break period l973/I-l982/IV and increase the sample period in increments of four quarters. The results for MlB level and first-difference specifications are presented in Table 5.6 and 5.7 (see Appendix D1-D4 for M1 and M2). In addition to the estimated coefficients and their absolute t statistics, the tables include the relevant summary statistics and the forecasting performances for the four quarters immediately following the end of the truncated-sample period. For both level and first-difference.M1B equations, the point estimates of real income, rate of change of nominal GNP, and lagged dependent variable are fairly consistent with the addition of the observations 141 Table 5.6 Recursive Regression and Simulation Results (Level) Constant 1973/I -1. -1982/IV(6. -1993/1v-1. .45)(7. (6 -l984/IV-l -1985/IV—1 -1986/IV-1. (6. -1987/IV-1 —1988/IV-1. (6. -1999/11-1. .20)(7. (6 1n 1nYt 5%. Y5 558 0.274 -0.668 -0.187 0.812 21)(6. 555 0. .498 0. (6. l6)(7. .448 0. (6. 12)(7. 396 0. 09)(7. .424 0. (6. 37)(7. 400 0. 24)(7. 396 0. 96)(3.80) (8.40)(l9.79) 290 -0.521 -0.190 0.761 73)(2.94) (8.89)(19.08) 294 -0.481 -0.l90 0.733 78)(2.49) (8.83)(17.39) 277 -0.455 -0.184 0.759 53)(2.67) (8.64)(21.52) 265 -0.457 -0.178 0.774 45)(2.78) (8.60)(23.60) 264 -0.467 -0.l77 0.784 67)(2.84) (8.82)(25.58) 258 -0.420 -0.172 0.792 43)(2.75) (8.45)(27.68) 259 -0.422 -0.l70 0.787 54)(2.77) (8.43)(28.46) 0.991 0.993 0.994 0.995 0.996 0.997 0.997 0.997 (1113,49,) 17.2 3739:1110 0.281 0.286 0.300 0.297 0.293 0.290 0.294 0.297 ‘4 Quarters D-W rho RMSExIO 2.319 0.421 0.489 (2.78) 2.198 0.509 0.452 (3.45) 2.157 0.556 0.291 (3.84) 2.195 0.473 0.259 (3.61) 2.171 0.460 0.282 (3.66) 2.177 0.472 0.354 (3.92) 2.166 0.415 0.382b (3.54) 2.130 0.417 (3.61) #The numbers in parentheses are the absolute value of t statistics. Iv ‘RMSE - {(1/4)§J(1nm,‘-lnii1t')2}1’2 where lnmt and 111111, denote the actual um and simulated value respertively. bTwo quarters RMSE. 142 Table 5.7 Recursive Regression and Simulation Results (First-Difference) . Aln ‘4 Quarters Alnyt AR, AY, (M115,-,/P,) R2 SEExlO D-W RMSExlO 1973/I 0.313 -0.777 -0.197 0.608 0.823 0.307 2.514 0.288 -1982/IV (9.61) (3.07)(10.59) (7.23) -1983/IV 0.318 -0.750 -0.197 0.600 0.824 0.304 2.488 0.378 (10.12)(3.00)(10.91) (7.35) -1984/IV 0.321 -0.723 -0.197 0.552 0.812 0.311 2.354 0.379 (10.19)(2.84)(10.83) (6.89) -1985/IV 0.308 -0.706 -0.191 0.552 0.793 0.316 2.406 0.322 (9.88) (2.74)(10.63) (7.00) ~1986/IV 0.297 -0.699 -0.185 0.570 0.780 0.316 2.399 0.279 (9.70) (2.72)(10.49) (7.44) -l987/IV 0.292 -0.696 -0.l83 0.593 0.772 0.313 2.437 0.416 (9.81) (2.74)(10.6l) (7.95) -l988/IV 0.291 -0.618 -0.179 0.561 0.747 0.320 2.367 0.317 (9.75) (2.41)(10.36) (7.67) -1989/II 0.290 -0.545 -0.176 0.544 0.744 0.321 2.304 (9.77) (2.16)(10.28) (7.50) #No constant term appears because the estimated coefficient on the first-difference of a time trend was virtually zero. ##Other remarks are all the same as Table 5.6. 143 beyond 1982/IV.108 The estimated coefficients on these variables all change by less than. one standard. error with. the exception. of the coefficient on the lagged dependent variable for the sample period ending in 1984/IV (level).109 Furthermore, the variance of the estimated residuals as well as the estimated first-order autocorrelation coefficients with the exception of the sample periods ending in 1983/IV and 1984/IV does not change significantly. This indicates that the error process is also stable and that there is no serious heteroskedasticity problem. If the money demand function was unstable or seriously misspecified, the estimated relationship would show a substantial degree of heteroskedasticity in the estimated standard errors as well as a marked change in the serial correlation coefficients.”0 The degree of variation in the error structure and the static RMSE is calculated from the recursive regression results in Table 5.6-5.7 and Appendix.Dl-D4, and.presented in‘Table 5.8. The coefficients of variation show that the MlB error structure is stable relative to that of either M1 um One unfortunate feature of the first-difference specification is that there still appears to remain a little serial correlation problem as indicated by the Durbin-Watson statistics, especially for the first two regressions. However, there is no definite statistical evidence to reject the hypothesis of serially independent error terms. um As observed in Appendix D1-D4, this consistency does not hold for M1 and M2 level and first-difference equations. 11° For example, it is widely recognized that the money demand function for the U.S. shows a significant increase in the standard error of the regression as wll as a substantial change in the serial correlation of the error structure, once the sample period includes the mid 19703 to the early 19805. Current research of the money demand function focuses on the issue of why the error structure of the money demand relationship changed. 144 Table 5.8 Stability of Error Structure Level First-Difference MlB M1 M2 M18 M1 M2 Mean of SEExlO Standard Deviation 0.292 0.348 0.136 0.314 0.370 0.150 of SEExlO 0.006 0.009 0.006 0.006 0.013 0.007 Coefficient of Variation 0.022 0.025 0.042 0.019 0.036 0.048 Mean of rho 0.466 0.657 0.546 Standard Deviation of rho 0.049 0.037 0.046 Coefficient of Variation 0.106 0.056 0.084 Mean of RMSExIO 0.358 0.378 0.166 0.347 0.420 0.173 Standard Deviation of RMSExlO 0.088 0.137 0.060 0.052 0.179 0.066 Coefficient of Variation 0.246 0.363 0.365 0.149 0.426 0.384 #Coefficient of ‘variation is calculated from dividing the standard deviation of each variable by its mean value. The values may not coincide with each other due to rounding. or M2 although the MlB serial correlation coefficient fluctuates a little more. For example, for the first-difference equations, the values of the coefficients of variation for the M18 are less than a half of the corresponding values for either M1 or M2. If the degree of variation of error structure for alternative monetary aggregates is evaluated simply in terms of the size of the SEE or the RMSE, as many researchers do, an incorrect conclusion could easily be reached" .As seen in'Table 5.8, these statistics are smallest for M2. As far as the coefficient of the interest rate is concerned, the results tell the entirely different story in Table 5.6 and 5.7. The estimated.coefficients on the interest rate are sometimes significant only 145 at the 5% level, i.e., marginally significant relative to the coefficients on the other RHS variables. Moreover, the estimated short-run interest sensitivity declines gradually throughout the sample period although the changes do not fluctuate by more than two standard errors. The empirical result seems to support the hypothesis that financial innovations and partial (or gradual) interest rate deregulation 'have decreased the interest elasticity of money demand. Thus, our result is consistent with the Cagan and Schwartz(1975) hypothesis and contradicts the assertion of Gurley and Shaw(l960). As observed in.Appendix D1-D4, this phenomenon is also true of M1 and M2. But the estimated coefficient of the interest rate in the M2 equation is much smaller in absolute value as well as less significant than the same coefficient in either the M1 or the MlB equation. The smaller interest sensitivity for broad money relative to narrow money is exactly opposite to the traditional conclusion of monetary theory. This conflict can be easily resolved once we realize the existence of peculiar institutional factors in Korea. The public never regards savings and time deposits in banks as an effective means of savings, because interest rates on these deposits are regulated far below the free market rate, and also 'because the actual (not officially recorded) inflation rate is high. The greater part of non-MlB components of M2 thus consists of very heterogeneous accounts for the special purposes, rather than the pure savings and time deposits as a vehicle of portfolio investment. Corporations and the public are frequently forced to deposit a proportion of their credit when getting loans from commercial banks. The customers also tend to hold time deposits for the privilege 146 or priority of obtaining bank loans in the near future. These kinds of compensating balances are inevitably practiced under the distorted financial system where there is always an excess demand for the bank credit. In addition, the public holds time deposits at a very low interest rate (e.g., a 2-3% annual rate) as a "subscription deposit" for buying apartments because only people with such deposits can apply for a newly- built apartment. These apartments are a good investment objective in an environment of high inflation, high population density, and very limited land. Although the above-mentioned is not the entire list of special deposits, most of non-M18 components of M2 are intrinsically insensitive to the changes in interest rates. Instead, those accounts depend largely upon the availability of credit or the fluctuations in the market prices of housing and real estate. To verify this argument statistically, a regression for the non-M18 components of M2 (NETM2) is estimated for the high inflationary period l973/I-l982/IV. The regression results for both level and first- difference equations are: ln(NETM2/P)t--0.035 +0.034lnyt +0.077R, -0.0313°rt +0.943(NETM2,,-1/Pt) (-0.29) (1.49) (1.04) (-2.21) (36.97) RF-0.994 SEE-0.0168 D-ws2.13 (5.13) Aln(NETM2/P)t-0.053Alnyt(+0.026ARt -0.035ATt-+0.832Aln(NETM2vq/Pt) (1.91) (0.16) {-2.26) (9.49) R?-0.946 SEE=0.0213 D-W=2.32 rho--0.43(-2.53) (5.14) As observed, the estimated coefficients on real income and interest rate 147 are not significantly different from zero in either specification; the estimated coefficients of interest rate are even positively signed though they are virtually zero; the values of coefficients for lagged dependent variable are not significantly smaller than one at the 1% level although they are different from one at the 5% level, so the estimated speed of adjustment is practically zero. Finally, there is no evidence of serial correlation in the error terms for the level equation while there exists a significant negative serial correlation for the first-difference equation. All these findings cannot be reconciled with traditional monetary theory without appealing to the particular conditions above-mentioned. Therefore, the instability of M2 demand funtion is largely because of those distorted institutional factors specific to Korea. Rather than undergoing a single shift or a series of shifts, the M2 demand function appears to be inherently unstable throughout the entire sample period. This is not consistent with the conclusion of many of the money demand studies for Korea that report stability of M2 compared to narrow measures of money and the focus of the monetary authorities onM2.111 The ration- ale for focusing primarily on M2 should be sought elsewhere. It is more plausible to claim that the monetary authorities target M2 since they intend to control bank credit and M2 is more closely related to the bank “4 As the main reason for the comparative stability of M2 relative to M1, a numerous papers by BOK's research staffs frequently cite the predictability of M2 in terms of the simple size of SEE or RMSE of M2 demand function without rigorous stability tests for alternative monetary aggregates. As discussed earlier, it proves false to judge the relative stability by the absolute values of these statistics rather than their coefficients of variation. 148 Table 5.9 Post-Sample Static Forecast Errors of MlB Demanda Actual ln(MlB/P)t Level First-Difference 1983/1 4.231 -0.061 -0.055 2 4.254 -0.038 0.009 3 4.298 -0.031 0.003 4 4.396 -0.059 -0.015 1984/1 4.320 -0.038 ~0.022 2 4.294 -0.031 -0.002 3 4.336 0.023 0.046 4 4.398 -0.072 -0.056 1985/l 4.403 0.046 0.065 2 4.435 0.029 0.034 3 4.466 0.015 -0.005 4 4.566 -0.014 -0.019 1986/l 4.582 0.011 0.020 2 4.639 0.047 0.056 3 4.690 0.013 -0.008 4 4.759 -0.014 -0.023 1987/1 4.782 0.012 0.023 2 4.836 0.033 0.041 3 4.863 0.011 -0.008 4 4.945 0.043 0.028 1988/1 4.968 -0.032 -0.033 2 4.972 0.011 0.024 3 5.032 0.056 0.051 4 5.069 -0.027 -0.052 1989/1 5.051 -0.054 -0.041 2 5.010 0.003 0.033 ‘Actual less simulated. 149 credit than either M1 or MlB. Next we consider the forecasting performances of the truncated sample regressions over the period beyond 1982/1V1 Table 5.9 presents the static forecasting errors for the four quarters immediately following the end period of the truncated-sample regressions in Table 5.6 and 5.7. While many economic studies rely on.dynamic forecasting to analyze the stability of strucural relationship, recently static forecasting is the preferred method for evaluating temporal stability of the short-run money demand relationships.112 The difference between static and dynamic forecasting is that the dynamic forecast uses previously forecasted values of the lagged dependent variable instead of the actual value of the variable. However, the dynamic forecasting procedure tends to provide "incorrect and misleading" information on the pattern and the extent of a shift in the short-run money demand relationship, even though it is appropriate for the “3 Statistically both forecasting procedures will yield unbiased forecasts (i.e., zero expected value of the forecast error) under the Ho of stability. However, the variance of dynamic forecast error in general is larger than that of static forecast error, so the dynamic forecasting procedure can be considered inefficient compared to the static forecasting procedure. For example, suppose that the money demand function experiences a once-and-for-all (permanent) intercept shift of size 6 at a point in time. Under the Ho of stability, the static forecast error will be 6 that will persist regardless of the time for which forecast is made. By contrast, the dynamic forecast will deviate from the actual value both because the intercept shift is not built into the forecast and because the lagged dependent variable is inaccurately forecasted for the intervening periods. Since the dynamic forecast error is a weighted average of current and past static forecast errors, dynamic forecast errors will accumulate over time. With the information on the pattern of dynamic forecast errors only, the researcher could probably not only misjudge the once-and-for-all (or several discrete) intercept shift in.the relationship as a continuous shift but also regard the forecast errors as ”ever-increasing" rather than a one-time intercept shift. For a statistical proof and the empirical evidence of this claim, see Hein(l980). 150 long-range forecast of money demand. As observed in Table 5.6 and 5.7, the static RMSEs are very close to thecorresponding in-sample standard.errors with.the two exceptions of RMSE in the level equation of 5.0% and 4.5% over 1983 and 1984 respectively. Furthermore, except for simulations over the year 1983 and 1987 the static forecast errors presented in Table 5.9 are not consistently one-sided. If a fundamental shift in the behavioral relationships has occurred (i.e., one of the coefficients Bo, 81, £2, 83, or 8,, has systematically changed), a level of real balances that is consistently below (or above) the simulated level should be observed following the downward (or upward) shift. However, little evidence of such consistently one-sided forecast errors is exhibited in Table 5.9. Thus the simulation results do not indicate a shift in the behavioral relationships of money demand. Nevertheless, a question still remains. If the demand for real balances did not shift around late 1982 or early 1983, why are real money balances consistently low in 1983 and high in 1987 relative to the respective predicted levels? One can interpret the behavior of real balances in 1983 and 1987 as evidence of a temporary "random shock”, not a downward or upward shift of the money demand function,113 since the negative or positive forecast errors are not significant compared to the standard errors of the relevant regressions. 113 According to the proponents of the disequilibrium money, changes in the supply of money can dominate the short-run movements in monetary aggregates. While there is no consensus about the mechanism of money supply shock, the proponents claim that deviations of actual real balances from the simulated values of a money demand function may be evidence of "supply-side” shocks rather than demand shifts as many suggest. However, this argument does not hold here. In fact, the growth rates of nominal MlB in 1983 and 1987 were not unusual from its historical ranges. 151 In sum, empirical evidence from the recursive regressions as well as the post-sample forecasting performances supports the conclusion that MlB demand function in spite robust even with extension of the sample period beyond 1982, although there is some suggestion of a gradual decrease in the interest sensitivity. However, it can be argued that the recursive regression approach does not really provide statistical evidence of the stability between the periods before and after 1982, since the recursive regressions incorporate the recent money demand behavior by updating sequentially the post-1982 data” .As an alternative, stability tests using the dummy variable techniques that split the whole sample into the two sub-sample at 1982/IV are investigated. 5.3.4. Dummy Variables Test The predictive Chow test asks whether there is at least one observation among last T2 observations whose mean value is inconsistent with the estimated relationship from first T1 observations. It does not point out which observations deviate most seriously from the pre-shift model. A useful dummy variable technique that gives the predictive errors relating to the timing of a possible structural change is suggested by Dufour(1980, 1982). The procedure estimates the full-sample regression, including dummy variables that take on the value of one or zero for each quarter beyond the selected break point. The estimated coefficients of the dummy variables then indicate the post-sample static forecast errors because the estimated coefficients on the explanatory variables are based on the pre-break data points. More importantly, the t statistics for the dummy variables indicate which forecast errors deviate significantly from 152 the pre-break model. By using the estimated coefficients on the dummy variables for the post-break period, we can examine the nature, pattern, and magnitude of the forecast errors or the structural shift.“‘ In order to conduct the Dufour test, we estimate both level and first- difference specifications adding individual dummy variables for each period 1983/1-1939/11:ns . 1989/II 111(14/17),-15o +s,1ny, +929, +9337, +9,1n(M,-,/P,) +2 5,13,, +5, (5.15) 983/1 s-l . 1989/11 A1n(M/P),-8'1Alny, +9359, +B'3AY, +B',A1n(M,-1/P,) +2 15',n,,+r,, (5.16) 8'1983/I where D,,-1 for t-s, 0 otherwise. The regression results for MlB are reported in Table 5.10 (see Appendix D5 and D6 for M1 and M2). For the level equation, all the estimated coefficients on the dummy variables are negative. As mentioned earlier, this indicates that the levels of actual real balances are consistently below the simulated levels, i.e., the static forecasts continuously overpredict the money demand throughout the post-1982/IV.11'5 Moreover, the Theil bias coefficient (UM) indicates that about 80% of the forecast error is attributable to bias (systematic one- sided prediction errors) although the RMSE is not more than two and half 11‘ For more details, see Dufour(1982), and application of the technique to stability test of the money demand function is found in Hafer(1985b). “5 The basic qualitative implications discussed below do not change when the constant interest elasticity specifications are estimated. 116The forecast is static with regard to the structural part (lagged dependent variable), but it is dynamic in association with the estimated autocorrelation coefficient from the pre-l982/IV model. 153 Table 5.10 Dufour Test Results Level First-Difference Dependent Variable lnm, Alnm, Constant -l.558(6.21) 1ny, 0.274(6.97) Alny, 0.313(9.61) R, -0.668(3.79) AR, -0.777(3.07) Y, -0.187(8.41) AY, -0.197(10.59) 1n(M,q/P,) 0.812(19.74) Aln(M,q/P,) 0.608(7.23) 1983/1 -0.061(1.89) ~0.055(l.75) 2 -0.064(l.66) 0.009(0.28) 3 -0.058(1.46) 0.003(0.10) 4 -0.083(2.03)* -0.015(0.47) 1984/1 -0.105(2.53)* -0.023(0.75) 2 -0.109(2.60)* -0.002(0.06) 3 -0.052(1.28) 0.047(1.52) 4 -0.122(2.94)* -0.055(1.74) 1985/1 -0.054(1.30) 0.065(2.07)* 2 -0.038(0.90) 0.031(0.97) 3 -0.041(0.95) -0.005(0.17) 4 -0.075(1.70) -0.023(0.71) 1986/1 -0.066(l.44) 0.017(0.55) 2 -0.026(0.55) 0.054(l.7l) 3 -0.041(0.87) -0.011(0.34) 4 -0.077(1.61) -0.028(0.89) 1987/1 -0.065(1.31) 0.022(0.70) 2 -0.038(0.76) 0.041(1.30) 3 -0.050(0.98) -0.010(0.32) 4 -0.024(0.48) 0.026(0.83) 1988/1 -0.082(1.54) -0.035(1.10) 2 -0.060(0.14) 0.031(0.98) 3 -0.006(0.11) 0.050(1.62) 4 -0.069(1.29) -0.056(1.78) 1989/1 -0.127(2.28)* —0.046(l.47) 2 -0.082(1.52) 0.043(1.34) R2 0.998 0.767 SEExlO 0.281 0.307 D-W 2.322 2.517 rho 0.421(2.79) ‘Mean ErrorxlO -0.635 0.029 Mean Absolute Error x10 0.635 0.303 RMSE x10 0.706 0.360 Mean Square Error x10 0.050 0.013 bUM x10 0.041 0.000 US x10 0.000 0.000 UC x10 0.009 0.012 154 #Absolute value of t statistics appear in parentheses. ##No constant term appears in the first-difference equation since the estimated. coefficient of first-difference of a time trend. was not significantly different from zero. *Signnificant at the 5% level. 1999/2 1999/ ‘Mean Error-(l/26) E 8,, Mean Absolute Error-(1/26) E '28,] 9-1993/1 9-1993/1 1999/2z 1999/2 RMSE-{(1/26) 2 8 )1’2, and Mean Square Error-(l/26) 2 £2, 9-1993/1 9-1993/1 bUM-(R-F)z where R-mean of realized ln(M1A/P),, F-mean of its predicted value based on the structural coefficient estimates in the upper part of the table. US-(SR-SF)2 where 81,-standard error of realized, and SF—standard error of predicted. UC-2(1-pRF)SRSp where pRF-correlation coefficient between realized and predicted value. §The relevant statistics may not coincide with each other due to rounding. times the in-sample standard error.117 This result seems to suggest the desired real money balances have decreased, given levels of real income and rates of return on financial and physical assets. In another word the money demand shifts downward permanently. If this is true, the Dufour test result is in sharp contradiction to the previous cusum-squares and F test results. However, the most significant departures from the pre-l982/IV model of money demand are gentered in late 1983 and 1984, a period closely related to an aftermath of the big financial scandalf118 This feature suggests that the money 117 For detailed explanations of Theil's "inequality coefficient" , see Thei1(1966, 1971) and Pindyck and Lubinfe1d(1981). “3 The nature of forecast errors for M2 (level) is quite different: the t statistics for 10 out of 26 dummy variables exceed the 5 % level although the forecast errors exhibit the same pattern. as MlB *with consistently one-sided negative values. This evidence supports the view that the disturbances in credit markets or financial innovations have had more direct and greater effects on the broad.monetary aggregates than the narrow transactions balances. The result also can be interpreted as indirect evidence that the money lenders who had time deposits-type at _. 155 demand function is simply subject to larger random shocks during the forecasted period 1983-1984 than the average shock during the estimation period, rather than a systematic downward shift. When the result for the first-difference equation is examined further, the random shock interpretation appears more likely than the downward shift. Unlike the level equation, the forecast errors in the first- difference equation show only a one time significant departure from the pre-l982/IV model during the past 26 quarters. Furthermore, the forecast errors alternate in sign and are of approximately equal magnitudes. The Theil decomposition coefficients also suggest that the forecast errors are due to unsysmatic random fluctuations.119 The forecast errors thus can be regarded as temporary, suggesting that the MlB demand function (level) was subject to relatively large random shocks in late 1983 and 1984 due to turbulent credit markets. Therefore, it can be concluded that the Dufour test result for the static forecast errors provides little evidence to indicate a fundamental and systematic shift in the behavioral relationship of MlB demand throughout l973/I-1989/II. Interestingly enough, none of the stability tests suggests that the MlB demand function has been absolutely stable over time. Instead those tests indicate that the MlB demand function has been more stable relative accounts abruptly withdrew their deposits as the curb market suddenly contracted considerably due to the increased risk in the curb market transactions following the big scandal of May 1982. “9 Again, this is not the case for M2 and M1: not only 4 (M2) and 5(Ml) of the coefficients' t values are significant at the 5% level but also the significant forecast errors are spread out over time. The evidence confirms the previous findings that M2 and M1 have been more generally unstable. 156 to alternative monetary aggregates, even though the MlB demand function has experienced some instability. Occasionally the MlB demand function is only marginally stable, suggesting that something in the relationship may have changed over time. On the one hand, the recursive regression implies that the interest sensitivity has gradually declined. On the other hand, the simulation errors over the post-l982/IV indicate that the intercept of the level equation may have shifted. To investigate further the stability of each coefficient over the 1982/IV break, we re-estimated both specifications for the full—sample period by forming interaction terms with the RHS variables through the use 20 of dummy variables:1 1:1(14/1>),-15o +6002 +(15,+a,02)1ny, +(132 +02D2)R,+ (93+a302)i, +9,+a,02)1n(1»1,_,/1>,) +5, (5.17) Aln(M/P),-(B'1+a'1D2)Alny, +(B'z-I-a'zD2)AR, +(9',+a',02)211'r, +(9',+a',02)51n(11,_,/P,)+q, (5.19) where D2-1 for l983/I-1989/II and.zero elsewhere. The results for MlB are presented in Table 5.11 (see Appendix D7 and D8 for M1 and M2). First, consider the level equation. None of the coefficients on the interaction terms is statistically significant; there is little improvement to the model in terms of R? and SEE; and both F ratio and LR statistics fail to reject the stability hypothesis at the 1% level. Thus 12° This method yields an identical conclusion concerning Ho with the standard ANOCOVA test when the latter procedure does not account for changes in the error structure. But the method has the advantage of automatically producing indications on which individual coefficient may differ across the two sub-sample periods. Alternatively, it is possible to get exactly the same result if we introduce interaction terms with the RHS variables through the use of two separate dummy variables--D1-1 for the pre-break period, 0 otherwise and D2-1 for the post-break period, 0 otherwise. 157 Table 5.11 Interactions Test Results Level First-Difference‘ Dependent Variable lnm, Alnm, Constant -1.505(5.82) 1ny, 0.265(6.49) Alny, 0.313(9.70) R, -0.655(3.91) AR, -0.777(3.10) Y, -0.l82(7.83) AY, -0.l97(10.69) ln(M,1/P,) 0.820(20.60) Aln(M,q/P,) 0.608(7.30) D2 0.524(0.91) D21ny, -0.09l(1.08) D2Alny, -0.098(1.50) D2R, 0.341(0.41) D2AR, 1.757(1.81) D2Y, 0.071(l.43) D2AY, 0.086(2.19)* D21n(M,q/P, 0.051(0.74) D2A1n(M,q/P,) -0.125(0.82) R2 0 997 0.771 SEExlO 0.290 0.304 D-W 2.202 2.336 rho 0.340(2.48) F(5.56)-l.62 XS-B . 92 F(4.58)-2.83* x,-11. 79* #The numbers in parentheses represent absolute value of t statistics. ‘No constant term appears because the estimated coefficient of the first- difference of a time trend was virtually zero. *Significant at the 5% level. the result does not support the weak implications of the recursive regressions and the post-1982 forecast errors. it reaffirms Rather, theinterpretation that the changes in the interest sensitivity and several significant forecast errors are largely attributable to the random shocks instead of a systematic phenomenon. Nonetheless, we re-estimated the regression by forming the interaction terms only with the intercept and the interest rate. The result is given below: ln(M/P),--1.362 +0.2421ny, -0.630R, -0.166Y,-+0.834(M,q/P,) (~6.16) (7.04) (-3.94) (-9.19) (27.01) -0.11392 +0.343029, (-0.95) (0.43) (5.19) 158 R?-0.998 SEE-0.029 D-W-2.13 rho-0.32(2.49) As seen above, the result does not support a downward shift of the intercept or a decrease in the slope coefficient of interest rate. Neither interaction term is significant at any acceptable level, nor is there an improvement to the fit of the equation. Next, consider the first-difference equation. The coefficients on interest rate and rate of change of nominal GNP interactions are found to be marginally significant; the regression is somewhat improved in terms of the decrease of 5.4% in SEE as well as the increase of 3.4% in R2; in addition, F ratio and LR statistics reject the hypothesis of no structural change at the 5% level although they do not reject at the 1% level. The result indicates that both parameters of interest rate and rate of change of nominal GNP declined in absolute value over the post-1982/IV. In particular, the coefficient of interest rate may shift from negative to positive across the two sub-sample periods. This result is consistent with the gradual decrease in the interest sensitivity as reported by the recursive regressions. The regression was run again with only the coefficients on interest rate and rate of change of nominal GNP allowed to take on different values in the two separate sub-periods. The result is given below: Aln(M/P)t-0.288Alny, -0.719AR, -0.182AY,1+0.561A(lnM,4/P,) (10.03) (-2.93) (-10.76) (7.95) +1.836D2AR, +0.0220259, (5.20) (1.91) (1.12) Rz-o.760 SEE-0.031 D-W—2.36 Now the estimated coefficient of the interaction term with the rate of 159 change of nominal GNP is no longer significantly different from zero, while the interest rate coefficient still appears changed significantly at the 5% level from negative to positive. As a result, it is apparent that much of the marginal instability of the demand for real balances is accounted for by the unstable interest rate coefficient. This conclusion is not surprising given that almost all official interest rates in Korea are distorted to a great extent. It does not stand to reason that any particular interest rate accurately reflects the opportunity cost of holding money. Under these circumstances, it is likely that the unstable coefficient of interest rates will be the main cause of any observed money demand instability. It is interesting that in. spite of' the unstable interest rate coefficient, our money demand function has been relatively stable in terms of its out-of-sample forecasting ability. It is the result of the fact that the money demand function exhibits consistent relationships with the RHS variables excluding interest rates. With relatively stable prices over the post-1982, the fluctuations of interest rates have been much less volatile than the pre-l982 period. As a result, the changes in interest rates have had little effect on the money demand. 5.4. Evaluation of Alternative Hypotheses Throughout the preceding empirical section of this chapter, the primary focus has been on specification of the short-run money demand function and on examination of the stability of the money demand function across the 1982/IV, a period that marks the beginning of a rapidly changing financial environment as well as the turning point of a low 160 inflation and interest rate regime. In some respects, the empirical results are not fully satisfactory. There may exist other specifications of the short-run money demand function that are more stable over time. With the multiplicity of stability tests presented, it is hard to draw a clear conclusion as to whether our money demand function has been stable throughout the whole sample period. What is more important, financial innovation continues reflecting the overall outcomes of interrelated innovationary processes, so it is almost impossible to evaluate a particular hypothesis concerning the money demand relationship correctly. However, some contributions can ‘be made toward furthering our understanding of the money demand relationship and of monetary targeting if various hypotheses are assessed carefully based on the empirical evidence presented. In association with the theoretical conjectures discussed, the above empirical results can be interpreted as follows. Hypothesis 1: The growth of money substitutes (or interest rate deregulation on non-bank financial institutions), such as CMA, RPs, and other money market funds, increases the interest elasticity of money demand. This assertion can be rejected with relatively high confidence because in general the empirical results indicate that the interest sensitivity of money demand has gradually declined since short-term financial markets began to flourish. At the same time, the static forecasts for the levels of real balances uniformly overpredict MlB real balances over the post- 1982. Thus it seems safe to claim that a proliferation of such short-term financial instruments have purified, rather than contaminated, the transactions money by progressively eliminating savings or non- 161 transactions balances previously lodged in the conventional M1. Hypothesis 2: The introduction of interest-bearing demand deposits such as overall-households-deposits and free-savings-deposits decreases the interest elasticity of transactions balances; in addition, the public is induced to shift asset demand funds into these interest-bearing demand deposits. The first view is likely to be supported because the tendency towards a decreasing interest sensitivity of money demand becomes more apparent after late 1985, shortly after free-savings-deposits were introduced. However, the second conjecture is not supported. There is no evidence of an upward shift of the MlB demand function after the introduction of free- savings-deposits in April 1985. Contrary to the hypothesis, the most likely downward shift point is identified as around late 1982 shortly after overall-households-deposits went into effect. If non-transactions funds really shifted into these interest-bearing demand deposits, an increase, rather than a decrease, in the interest rate sensitivity should be observed. Finally, the stability of M18 relative to M1 suggests that the removal (or relaxing) of the interest rate ceilings on transactions deposits contributes to economizing on the conventional M1 primarily by inducing substitution between curency or traditional demand deposits and new interest-bearing demand deposits, and hence that deposit rate deregulation is unlikely to cause MlB to become a transactions-cum- investment vehicle to any considerable extent. Hypothesis 3: The rapid spread of telecommunications and computer technology in the financial systems not only changes the marginal relationships of the demand for money but also shifts the transactions 162 money down, particularly the precautionary money demand, through reductions of transactions costs as well as uncertainty about cash flows. The empirical results do not support the conjecture of an alteration in the marginal relationships underlying between real balances and its determinants. No systematic changes in the parameters of MlB demand function are found with the single exception of interest rate coefficient. If the conjecture really holds, an increasing, rather than decreasing, interest rate slope coefficient should be observed as the public becomes more sensitive to the spreads in interest rates. However, for the conjecture of the level shift, there is an indication that the demand for the conventional transactions money has declined. The time trend is negative at an annual rate of approximately 1.5% in the M1 demand function. Nonetheless, we can not separate how much of the secular decrease in the M1 demand is attributable to innovations in cash management. It should be emphasized that the improvements in cash management technique are the consequence of all the interrelated financial innovations. Interestingly enough, there is no apparent evidence that the demand for MlB has a negative secular trend. Hypothesis 4: The interest elasticity of money demand has decreased in the 19803 as nominal interest rates have fallen along with low inflation rates. The semi—log specification in interest rates appears to outperform the double-log specification in terms of the forecasting ability for the post- 1982 period. Thus it is reasonable to say that the interest elasticity of money demand varies positively with the changes in interest rates. The interest elasticity decreases when the sample period is extended to 163 include the low interest rate period after 1982. However, the decrease of the interest elasticity is not entirely due to lower nominal rates, because the semi-log slope coefficient of interest rates does not remain constant, but decreases as well over the post-1982. Part of the decrease in the interest elasticity comes from the decrease in the slope coefficient and.must be ascribed to other explanations associated with the financial repression as well as disturbances in the credit market conditions. Hypothesis 5: Financial innovation has less impact on the demand for M2 than for transactions balances since the broader monetary aggregate itself tends to internalize the effects of financial innovation. We can easily reject this argument because the M2 demand function is highly unstable. In particular, the instability of non-MlB components of M2 suggests that the M2 demand.will not be stable in the future unless the abnormal, peculiar institutional characteristics specific to Korea are eliminated" Exactly opposite to this hypothesis, it can be inferred.that, with a single exception of the interest-bearing demand deposits, all the innovations or credit market disturbances have greater impacts on M2 rather than on MlB. Therefore, the assertion should be relegated to be no more than an excuse for control of bank credit or accelerating money growth whenever the monetary authorities want. Hypothesis 6: The target monetary aggregate should be redefined more broadly to include "immediately available funds" (IAFs) such as CMA, RPs, and other money market funds or to be as a weighted sum of various (liquid) financial assets over their transactions services only. The empirical result shows that the demand for the conventional Ml 164 measure appears unstable in a changing financial environment. Clearly, financial innovation has a considerable effect on the conventional definition.of'moneyu However, our redefined transactions measure of money including overall-households~deposits and free-savings-deposits exhibits a reasonably stable relationship with small number of its determinants in spite of the significant financial changes during the last decade. In this line, these new interest-bearing demand deposits can be regarded as perfect substitutes for the conventional money. It is uncertain if the monetary aggregates including IAFs or a fraction of various liquid financial assets will exhibit a more stable relationship than the MlB. Whether or not such aggregates should be the target. monetary aggregate is an. empirical. matter that ‘needs to 'be investigated further. So far as our institution goes, however, such IAFs or other liquid assets may be more proper substitutes for investment (idle or non-transactions) funds rather than pure transactions balances. There is no guarantee that those monetary aggregates will continuously exhibit the stable relationship in the future even if they fit well over some sub- periods of past historical data. Hypothesis 7: Instabilities of the money demand can be attributed to the change in monetary policy regime in early 1981, the so-called shift from direct domestic credit control to indirect monetary control regime. Before evaluating this hypothesis, there is great doubt that the attitude of financial and monetary policy actually changed as announced. In practice, official announcements do not always mean a real change of the policy regime, but sometimes are nothing more than bureaucratic lip services. Presumably, no one believes that there was a fundamental shift 165 in monetary control regime. If the monetary authorities really placed more emphasis on an appropriate transactions aggregate rather than the bank credit (or total domestic credit), then the money demand would be more stable in.the post-1981 than in the pre-l981 of direct credit control regime. The opposite is true in terms of the regression residuals. In short, the monetary policy regime can be characterized as discretionary ”yo-yo swings" (Friedman, 1982) without any reasonable rules. It is surprising that despite such a capricious financial and monetary policy regime, our money demand function shows relatively consistent estimated coefficients. Furthermore, changes in the bank credit or the total domestic credit are found not to cause any significant disequilibrium buffer stock money. In this respect, one may claim that the money demand relationship has been stable because there has been no real change in monetary policy regime and accordingly rational agents have not revised their decision rules appreciably. The argument is not inconsistent with the rational expectationists' view. Hypothesis 8: In the face of a changing financial environment, the money demand function (or velocity) is so unstable or unpredictable that control of nominal GNP through monetary aggregate targeting is infeasible or unsuccessful. This frequently alleged conjecture about instability is not supported by the experience over the past two decades. The stability tests presented here in general fail to reject the stability hypothesis of the MlB demand relationship. Except for the coefficient of interest rates, there are very few changes in the parameters for the MlB demand function when the sample period is extended over the turbulent post-1982 period. 166 The static RMSEs in the forecasting experiments are also well within two times the in-sample standard error. In particular, the forecast errors in the first-difference specification suggest that the post-1982 errors are transitory in nature, and that they are primarily due to the distorted, turbulent credit market conditions rather than a systematic shift in the demand for transactions money. This random-walk nature of the forecast error implies that the average forecast accuracy can be improved as forecast horizon is lengthened to two or more quarters.121 Thus the empirical evidence indicates that although monetary policy relying on quarter-to-quarter forecasts of money demand growth may not fare well, the long-run relationship embodied in the money demand function can be exploited successfully by emphasizing more long-run monetary growth and long-term goals of stable income growth and price stability. There is little empirical support for the tenuous assertion that attempts to control inflation through restrictive monetary policy will be unsuccessful since the money demand relationship is unstable. In conclusion, the hypothesis of structural change in the demand for transactions money is either not significant or only marginally significant. The marginal instabilities experienced in the 19803 are largely due to the unstable interest sensitivity for which financial repression, particularly the low interest rate policy and direct credit 121 This conclusion is also reached by the numerous time series literature of velocity including Gould and Nelson(1974), Nelson and Plosser(1982), and Hein and Veuglers(l983). The general conclusion is that velocity follows a random walk with a drift, i.e. , velocity growth fluctuates randomly about a fixed mean. 167 controls, is accountable. To the extent that financial innovations are responsible for such.a marginal instability, the ultimate precursor of the instability is the distorted, chaotic credit market conditions which in turn originate from extemporaneous financial and monetary policy. In other words, it is legitimate to question whether the money demand would have been subject to the instabilities experienced had the financial and monetary policy followed a stable rule from the longer-run perspectives. While significant financial changes occurred during the last decade and in turn had great impacts on M1 as well as M2, there is little evidence that those innovations caused the underlying structural relationship of the MlB demand to change systematically so that monetary targeting policy would become ineffective or unsuccessful in achieving the long-run economic stability. Provided that the target money aggregate is appropriately redefined in response to institutional changes, the monetary targeting can continue to be the most effective macroeconomic policy and should be used for the stable economic growth as well. CHAPTER VI SUMMARY AND CONCLUSIONS There is no doubt that we lack sufficient understanding of financial innovation to properly deal with the subject in a monetary economy. It should be borne in mind that the phenomenon is not yet complete and the discussions here are confined.to the innovation.that has occurred to date. It is virtually impossible to correctly identify the dynamics of a portfolio adjustment model. On the other hand, the true money demand function may be more stable than the model presented here. Financial changes can and do affect the existing money-income relationship, but the question of whether or not the relationship is significantly influnced is an empirical matter that needs to be investigated rigorously. Once the interest-bearing transactions deposits are added to the conventional money, the MlB demand function in Korea except for the interest sensitivity is reasonably stable over the past 64 quarters. In a financially repressed economy, it is not surprising that the interest rate coefficient in the demand for real money balances is main cause of the observed marginal instability of the money demand function. Nevertheless, there is little empirical evidence to support various popular assertions about the money demand behavior and.monetary targeting in association with the recent financial changes. Prior to undertaking this thesis, I thought it likely that the money demand function in Korea 168 169 would be highly unstable because of the substantial (and volatile) inflation and interest rates for which discretionary money supply or bank credit innovations are ultimately responsible. Interestingly enough, dhis preconception does not appear to be supported by the empirical tests. The arguments and analyses in the text can be summarized as follows. Chapter 11: Since a combination of various socio-economic conditions seems necessary for financial innovation to take place, no simple hypothesis is sufficient to explain the complexity of innovation- generating influences and processes. Further research on the subject should attempt to integrate the existing individual hypotheses systematically into a more general framework. Financial innovation and deregulation in Korea can be viewed as a process of regulatory avoidance through which the free market responds to the low interest rate policy and the direct credit controls over regulated financial institutions in an effort to overcome credit bottlenecks. This experience suggests that the scope of government regulatory powers can be considerably’constrainedfiby'the free market forces or'profit opportunities of financial firms when there exist the conflicts between economic and political power. Unlike many developed countries, high and volatile interest rates do not seem an essential part of innovations as far as the Korean experience is concerned. Chapter III: It is difficult to evaluate correctly the theoretical effects of financial innovation on the money demand relationship even in a simple model. Some changes can shift the level of money demand up or decrease the interest elasticity of money demand, while others can do the opposite. Moreover, the short-run money demand model implies that the 170 more discretionary the monetary policy regime is, the more difficult it is to identify the parameters of the short-run money demand function. However, the legal deregulation of deposit rates should not affect the money demand function importantly. Even under deposit rate ceilings, banks have paid market-related implicit interest on transactions deposits through various compensatory arrangements. The legal deregulation simply converts the previously-paid implicit returns into explicit interest returns. The most significant aspect of all major financial innovation is that the traditional distinction between monetary and financial assets becomes increasingly fuzzy. Thus the central issue on the stability of money demand relationship is reduced to the question of whether transactions and asset demand balances can be effectively separated in the face of financial innovation. However, it is realistic to expect that such a separation continues to exist even after financial innovation and deposit rate deregulation. True transactions assets are unlikely to pay yields as high as other (liquid) investment assets not only because of the existence of extra costs (or risk) inherent in the bank's asset transformation process but also because of the public's willingness to hold such assets of fixed nominal value payable on demand or in short notice in the absence of perfect synchronization of receipts and expenditures. Chapter IV: The substantial uncertainties about money-income relationship in the face of financial innovation and deregulation are primarily applied to a transition period in which the new relationship becomes established. This does not mean that financial innovation affects 171 the long-run (or steady-state) properties of money demand. Given the difficulties encountered in an attempt to allow for the impacts of financial innovation on the money demand in the conduct of short-run stabilization monetary policy, policymakers should not be aggressive for the adjustment of base and/or target growth rate of money supply, but be more prudent than as usual. Again, this does not imply that the authorities must stick to a fixed. growth rate rule forever for a particular definition of monetary aggregates. Rather they should closely monitor the developments of institutional changes or probable shifts in money demand so that the target monetary aggregate can be redefined as timely and as appropriately as possible. The rationale for various alternative intermediate target variables seems to be much weaker theoretically as well as empirically than the critics of money supply targeting generally claim. As long as we accept the proposition that "Inflation is always and everywhere a monetary phenomenon.”, and as long as the long-run objective of monetary policy is to constrain nominal GNP to a desirable inflation rate, a role for some measure of transactions monetary aggregate is unavoidable and control of the monetary aggregate is the necessary strategy for bringing about the desired behavior of nominal GNP. In addition, when changes in market interest rates are transmitted more rapidly and more pervasively to all sectors of the economy after financial liberalization, large short-run fluctuations in interest rates should be prevented. Chapter V: It is admitted that most issues in the empirical money demand research, particularly those related to problems of short-run money demand specification, are not yet resolved. Any prior restriction on the 172 dynamics of a portfolio adjustment model, without a sufficient reference to the data-based. time series characteristics, casts doubt on the credibility of estimated results. In spite of such problems inherent in conventional money demand regressions, the estimated aggregate money demand.re1ationship still depends importantly on the public's money demand behavior. Stability overtime of an estimated.money demand regression and satisfactory post-sample forecasting ability can.be regarded as empirical evidence consistent with the hypothesis of'a stable money demand function. For the empirical investigations of the stability of money demand function in the changing financial environment of Korea, the chosen sample period is 1973/I-l989/II. The aggregate long-run equilibrium demand for real money balances (nominal money balances deflated.by the GNP deflator; 1980-100) is specified as a function of real income, corporate bond rate, and rate of change of nominal income. Because of insufficient degrees of freedom, an annual money demand function cannot be examined rigorously. A preliminary specification search for the quarterly'money demand function started with either unconstrained distributed lags models or with relatively low order polynomial restrictions on the distributed lag patterns. These attempts failed to produce any satisfactory results. Through various specification search tests, the appropriate specification of the quarterly MlB demand function is found to have the following characteristics. 1) The NSAH model is much better than the RSAH model. 2) The zero price elasticity of real money balances is supported. 3) The long-run unitary income elasticity is rejected. 4) The semi-log specification in interest rates outperforms the double- 173 log form particularly in the out-of-sample forecasting ability. 5) Estimation under the single equation framework is not likely to cause a serious simultaneous equations bias problem. 6) Other alternative opportunity costs variables have no additional explanatory significance once the corporate bond rate is accounted for. 7) Measures of the anticipated or the unanticipated rate of inflation have no separate effect on the real balances independent of the lagged dependent variable and the rate of change of nominal income. 8) There is no significant role for the buffer stock or temporary disequilibrium money. 9) No secular time tend is found. Both log level and log first-difference forms of the semi-log specification are estimated and subject to stability tests. The major results for the estimated relationships are as follows: 1) The rate of change of nominal income is interpreted as a proxy for the total nominal yield on physical assets, compensating more or less for the effect of a downward biased price level. It captures an effect independent of nominal interest rates. 2) The coefficient for the rate of change of nominal income is much smaller in absolute value than that of the corporate bond rate. This finding suggests that any change in money holdings at the margin is primarily as a substitute for financial assets. It is consistent with the implication of the transactions money demand theory. 3) The short-run elasticities of the respective independent variables are virtually the same in both levels and differenced specifications. But the coefficient of the lagged dependent variable in the level form is 174 much larger than that in the first-difference form. The puzzle can be viewed as a statistical artifact inherent in the usual empirical stock- adjustment models. With econometric techniques to date, it is almost impossible to separate the speed of adjustment from serial correlation and to identify the rate of portfolio adjustment correctly. A series of stability tests are conducted, since the shifts of money demand function may show up in different ways and various stability test techniques may not be equally powerful in identifying a particular kind of instability encountered. 1) The time series plots of cusum-squares test statistics exhibit different patterns among alternative aggregates. The statistics for MlB remain very close to the corresponding mean values, while the statistics for M1 and M2 drift considerably overtime. This feature indicates that the M18 demand function is quite stable even though the M1 and M2 demand functions may suffer instability. 2) The most likely timing of a structural change is identified and selected around late 1982 or early 1983, judging from the statistical inferences from cusum-squares test statistics, Quandt's log-likelihood ratio, and residual patterns of the first-difference equations. 3) The conventional F tests for a breakpoint at 1982/IV fail to reject the stability hypothesis of the MlB demand relationship but the tests reject the null hypothesis for M1 and.M2 at the 1% level of significance. 4) The recursive (or truncated sample) regression results show that the estimated coefficients of the RHS variables in the MlB demand function with the exception of the interest rate coefficient are quite consistent as the sample period is extended beyond 1982/IV. However, the interest 175 sensitivity turns out to decline gradually. This result is in favor of Cagan and Schwartz(l975) hypothesis and against that of Gurely and Shaw (1960). The same phenomenon is found in both M1 and M2. However, the interest sensitivity of M2 is much smaller as well as less significant. The smaller interest sensitivity for broad money relative to narrow money is attributable to the non-M18 components of M2 that consist of accounts for special purposes rather than pure savings balances. 5) The static forecast error patterns over the post-l982/IV also suggest that the MlB demand function has been more stable than either M1 or M2 demand function. In particular, the forecast errors for the MlB first-difference specification alternate in sign and are of almost equal magnitudes. The Theil decomposition coefficients indicate that the forecast errors are largely due to unsystematic random shocks from the turbulent credit market conditions rather than a systematic shift of the underlying behavioral relationship of the MlB demand function. 6) Finally, the analysis of covariance provides evidence that the parameters of the MlB demand function are not subject to a shift across 1982/IV except for the parameter of interest rate. As observed above, various stability test methods do not yield the same results. It is not easy to reach the unambiguous conclusion that our money demand function is stable throughout the whole sample period. However, it is certain that the MlB demand function is more stable than the demand function for alternative aggregates. The observed marginal instability is largely due to the unstable interest rate coefficient which is ultimately attributable to financial repression. While significant financial innovations occurred during the last decade and had great 176 impacts on M1 as well as M2, there is little evidence that those innovations caused the behavioral relationship underlying the MlB demand function to change fundamentally. Therefore, a rationale for various popular assertions about the money demand function and.monetary targeting in association with the recent financial changes should. be sought elsewhere. Finally, more progress will be made if future empirical research efforts are devoted in the following direction: the examination of time- series properties of velocity will allow us to determine if it is possible to find a set of efficient restrictions on a money demand specification that would be consistent with the properties of velocity; the approach to the estimation.of money demand equations from monthly and annual (as time passes by) data also enables us to pay more attention to the time-series characteristics of sample information and to rely less on implausible prior restrictions on the distributed lag patterns; finally, it is worth while to investigate a wealth (or permanent and temporary income) effect as transactions assets become more contributing to investment purposes than in the past. APPENDICES APPENDIX A Rational agents face ”signal extraction" problem in an environment in which they have imperfect information about the unobserved permanent variables XP, on which their decision-making bases. By using a linear regression recursively (or sequentially), the agents can forecast XP, in an optimal way. If X9, is projected against 0,-1 at the beginning of current period (3.1091) xv, - p[xp,|n,-,] + ux, where E(Ux,o,-,)-0, E(UX,)—0, E(UX2,)-azu, and E(UX,UX,)-0 for tfls. Subsequently, X, becomes available as of time t in addition to 0,-1 and is assumed related to XP, according to (3.11) X, - XP, + 6X, where E(ex,n,-,)-0, E(6X,)-0, 9(ex2,)-a2,x, and E(ex,ex,)-0 for res. Substituting (3.10s) into (3.11) gives (3.15) x, - P[XP,|0,-1] + ux, + ex, where again E(UX,+6X,)-0 and E[(UX,+6X,)0,-1]-0 so that E(UX,+cXt)z- azuwfix. Since by construction both UX, and 6X, are orthogonal to 0,-1, equation (3.15) is a linear regression equation with disturbance UX,+6X,. Thus we have (3.16) P[X,|0,-1] - P[xp,|o,_,] Now if XP, is projected against X, in addition to 0,-1 (3.10b) XP, - P[XP,|0,-,, x,] + ax, where 77X, possesses all the usual characteristics of random disturbance. Let P[XP,|n,-,, x,]-m,_,+9x,. Then 177 178 Project both sides of (3.17) on 0,-1 to obtain (3.19) P[XP,|0,-,) - mm + 99[x,|0,-,] because of P[Hfl,-1|0,-1]-Hn,-1 and P[nX,I0,-1]-0. Subtracting (3.18) from (3.17) gives (3.19) XP, - P[xp,|17,-,] - 9(x, - P[x,|o,-,]) + .711, The least squares orthogonality condition of 71X, to 0,-1 implies that “X,- P[X,Ifl,-1]) must be the projection of (XP,-P[XP,I0,-1]) against (X,-P[X,l0,- 1]). Thus (3.19) can be rewritten as (3.20) x9, - P[xv,|o,-,] + p[(xp,-1>[xp,|o,_,])|(x,-P[x,|n,_,])] + fix, The above equation shows that, as an observation X, becomes available, the forecast of 10’, can be improved by adding to the imperfect forecast (P[XP,I0,-1]) the projection of "unobserved" forecast error (XP,-P[XP,|0,-1]) on the "observed" forecast error (X,-P[X,|0,-1]), i.e., so long as these forecast errors are correlated, the new observation X, carries a useful information for estimating Xp,. From equations (3.10a), (3.15), and (3.16), XP,-P[XP,|0,-1]-UX,, and X,-P[X,I0,-1]-UX,+6X,. The least squares regression coefficient is then given by E[UX,(UX,+£X,)] 020x (3.21) 0 - - E(UX,+6X,)2 azux + 02:75 By virtue of the orthogonality conditions on ”X, equation (3.19) can be rewritten as ,XP, - P[xv,|0,_,] + 901, - P[x,|o,-,]) - 9x, + (1-9)P[XP,|0,-,] ('.'P[XP,IO,_1]-P[X,IO,_1] ) - 9x, + (140510,.1 (-.- P[XP,|O,-1]-HO,-1 ) which is equivalent to the equation (3.13) in the text. 179 APPENDIX B Suppose that policy makers specify the St. Louis Fed-type of reduced- form equation: (4.4) Y, - vM, + U, where Y-nominal GNP (goal variable), M-money supply (intermediate target), U-various exogenous variables other than M that affect Y, and v-response coefficient of Y to M (money multiplier). In the face of financial changes, policy makers are very uncertain about money-income relationship (multiplicative uncertainty). That is, they estimate. V’ by fitting equation (4.4) to historical sample data, but they are aware that the actual value of v may substantially different from its expected value (v) . They are also uncertain about U, (additive uncertainty) although they forecast U, as precisely as possible. In addition, we assume that policy makers choose the optimal policy on the basis of minimizing mean squares of error (MSE) of the actual Y, around the target Y",: (4.5) MSE - E(Y, - 37",)2 - E(Y, - 9:39,)2 + E(EY, - 17",)2 - azy-+ (bias)2 where EY,-expected value of Y,, and nay-variance of Y,. In a deterministic world or "certainty equivalence” (uncertainty associated only with U,), policy makers set EY,HYKH i.e., they act relying on the expected value as if they were certain the expected value would actually occur. Then MSE,-E(vM,+U,-Y",)z. First order condition (FOC) for minimizing MSE, thus gives the optimal policy under such a deterministic world: 180 A Y's ' Ur. (4.6) E - (7 However, in the presence of multiplicative uncertainty as well as additive uncertainty, the policy action itself affect Y, because of a randomness of v. Then, (4.7) 1159:,- E(vM,+U,-vM,-U,)2 + E(vM,+U,-Y",)2 - ava,2 + 0112 + 2pavauM, + (1"IM,-+-I"J,-Y",)2 where azv-variance of v, 02" - variance of U,, and p-correlation coefficient between v and U,. FOC for minimizing MSEu then gives the optimal policy under both multiplicative and additive uncertainties: v(Y",-U,) -pa,,aU (4.8) 11* - (724-02,, Compared with M in (4.6), M' in (4.7) suggests that policy makers should make use of more information (p, av, and on) than just expected values of v and U,. Assuming p-O gives: (9*,-f1,)/9 (4.9) M' - 1+(a',,/v)2 The optimal policy in (4.9) implies that policy makers should partially fill the expected "gap" between.YT, and U, as long as they are uncertain about v (i.e., aflffl). In particular, when the money demand relationship may exhibit greater uncertainties during transition periods following financial changes, the money-income relationship (v) is more likely subject to uncertain variations. Therefore, policy makers should not try to accommodate money supply for every fluctuation in Y,iIIthe face of money demand instabilities during the transition phase. 181 Appendix C1 Specification of Quarterly Money Demand Function (1973/I-1989/II) Dependint Variable - 1n (MlB/p), Nonne s ted B,+85-0 85-0 Made]. 3‘4'35-0 35-0 36-0 31+3‘-1 36-0 36-0 Constant -l.457 -l.398 -2.199 -l.456 -l.375 -1.396 -2.205 (6.16) (6.15) (7.38) (6.20) (6.04) (6.20) (7.61) 1ny, 0.268 0.260 0.382 0.268 0.263 0.259 0.383 (7.51) (7.48) (8.55) (7.56) (20.98) (7.54) (8.78) R, -0.446 -0.428 -0.725 -0.444 -0.577 -0.422 -0.708 (2.67) (2.60) (3.76) (2.84) (3.34) (2.77) (3.99) Y, -0.168 -0.l70 -0.l6l -0.l68 -0.165 -0.170 -0.161 (8.11) (8.36) (5.42) (8.17) (8.21) (8.43) (5.48) lnm,, 0.782 0.784 0.705 0.782 0.737 0.787 0.712 (20.45) (20.54) (16,51) (28.01) (20.98) (28.46) (21.02) x, -0.710 -0.784 -0.710 -0.697 -O.787 (7.42) (20.54) (7.53) (7.57) (28.46) 1nP, 0.001 0.002 0.007 0.038 (0.02) (0.09) (0.23) (2.13) R2 0.997 0.997 0.995 0.997 0.998 0.997 0.995 SEExlO 0.299 0.299 0.413 0.297 0.303 0.297 0.410 D-W 2.124 2.134 2.082 2.210 2.184 2.130 2.073 rho 0.434 0.423 0.305 0.432 0.544 0.417 0.295 (3.45) (3.35) (2.26) (3.75) (4.83) (3.61) (2.38) LR 142.06 141.69 120.35 142.06 140.90 141.68 120.31 # The numbers in parentheses are absolute value of t statistics; R? is the coefficient of determination corrected for degrees of freedom; SEE is the standard error of the estimated equation; D-W is the Durbin-Watson statistics; rho is the CORC estimate of first-order autocorrelation coefficient; LR is the value of the log-likelihood function. 182 (continued) ~l.026 ~l.833 -l.351 -2.080 (6.30) (9.53) (6.01) (7.74) 0.231 0.401 0.259 0.382 (23.64) (15.63) (21.34) (14.91) -0.623 -1.l24 -0.558 -0.988 (3.56) (3.74) (3.35) (4.51) -0,154 -0.l81 -0.l67 -0.166 (8.25) (6.96) (8.65) (5.93) 0.769 0.599 0.741 0.618 (23.64) (15.63) (21.34) (14.91) -0.769 -0.741 (23.64) (21.34) 0.037 0.071 (2.09) (3.18) 0.987 0.974 0.988 0.976 0.310 0.440 0.301 0.422 2.199 2.427 2.188 2.233 0.558 0.758 0.535 0.512 (4.82) (8.59) (4.74) (4.53) 138.28 115.08 140.76 118.37 183 Appendix C2 11mm --1.291 + 0.2451ny, - 0.4599, + 0.166Y,-+ 0.7951n(M,,/P,) (-5.60) (6.90) (-2.65) (-9.02) (27.26) R?-0.997 SEEx10-0.298 D-W32.123 rho-0.414(3.62) Appendix C3.a 11m, - -1.253 +0.2481ny, -0.238RCURB, -0.1617,-+0.7901n(M,,/P,) (-5.49) (7.34) (-2.99) (-9.05) (27.39) R?-0.997 SEEx10-0.295 D-W52.064 rho-0.370(3.10) Appendix C3.b lnm, --1.304 +0.256lny, 022214, 0151900129, -0.166Y, +0.7761n(M,-1/P,, (-5.59) (7.44) (-1.06) (-1.31) (-9.14) (26.99) R?-0.997 SEEx10-0.295 D-W-2.093 rho-0.389(3.23) Appendix C4.a lxmu- -1.399 +0.2631ny, -0.197R, -0.365RSD, -o.1729,.+0.7791n(M,,/P,) (-6.26) (7 67) (-0.91) (-1.39) (-9.53) (27.96) R?-0.997 SEEx10—0.294 D-W-2.119 rho-0.395(3.26) Appendix C4.b 11m, - -1.443 +0.2431ny, -0.127(R,-RSD,) -0.162Y, +0.8191n(M,q/P,) (-6.04) (6.94) (-0.52) (-7.79) (31.45) 9?-0.997 SEEx10-0.314 D-W-2.150 rho-0.464(4.15) Appendix C5.a 11m, - -1.363 +0.2631ny, -0.526R, +0.306RT, -0.171Y,-+0.7681n(M,q/P,) (6.00) (7.67) (-2.79) (1.01) (-9.54) (23.35) R?-0.997 SEEx10-0.297 D-W52.131 rho=0.446(3.69) 184 Appendix C5.b 1mm, - -1.347 +0.262lny, -0.526(R,-RT,) -0.l70Y,1+0.7621n(M,1/P,) (-5.99) (7.72) (-2.64) (-8.63) (22.90) R?-0.997 SEEx10-0.296 0-w-2.131 rho—0.473(3.92) Appendix C6.a 11m, - -1.395 +0.2581ny, -0.4199, -0.1717, -0.013«,, +0.787ln(M,1/P,) (-5.91) (7.15) (-2.70) (-8.26) (-0.15) (27.99) R?-0.997 SEEx10-0.299 D-Wh2.l31 rho-0.416(3.57) Appendix C6.b lxmn - -1.39o +0.259lny, -0.426R, -0.1709,.+0.133(«,-«,,) +0.7971n(M,,/P,) (-6.20) (7.59) (-2.75) (-9.52) (1.24) (29.41) R?-0.997 SEEx10-0.295 0-ws2.129 rho-0.436(3.77) Appendix C7.a lnm, - -1.533 +0.2771ny, 04939., 019437, +0.7801n(M,-1/P,) (-6.07) (7.49) (-2.96) (-7.96) (27.58) +0. 294ln(BC,/BC,-1) (1.13) R5-0.997 SEEx10-0.296 D-W-2.149 rho-0.440(3.79) Appendix C7.b 1mm, - -1.495 +0.26llny, -0.665R, -0.1909, +0.9241n(M,,/P,) (-6.04) (6.67) (-3.79) (-9.20) (19.95) +0.1351n(TDC,/TDC,1) (1.66) R?-0.991 SEEx10-0.273 0-w-2.269 rho-0.451(3.05) Sample Period-1973/I-l982/IV 195 Apendix C7.c lnml, - -1.563 +0.4021ny, -0.686R, -0.239Y,~+0.4501n(M1,1/P,) (-5.95) (10.17) (-2.96) (-9.97) (7.24) +0.5321n(Bc,/Bc,,) (1.79) R?-0.988 SEEx10-0.351 D-W-l.841 rho-0.631(5.38) Appendix C7.d lnml,1- -1.696 +0.3691ny, -0.957R, -0.241Y,-+0.5961n(M1,1/P,) (-4.55) (8.90) (-3.31) (-9.87) (6.40) +0.4351n(BC,/BC,1) (1.40) R?-0.959 SEEx10-0.334 D-W—1.867 rho-0.745(5.84) Sample Period-1973/I-1982/IV Appendix C8.a 11m, - -1.219 +0.002TIME + 0.264lny, -o.4719, -0.170Y,-+0.7161n(M,1/P,) (-3.90) (0.91) (7.79) (-2.63) (-8.61) (9.10) R?-0.997 SEExlO-0.298 D-W52.163 rho-0.520(3.36) Appendix C8.b lnml, - -1.974 00035711113 + 0.3841ny, -0.576R, 02271}, +0.643ln(Ml,-1/P,) (-5.30) (-2.50) (9.39) (-3.47) (-9.17) (7.59) R?-0.988 SEEx10-0.349 D-W-l.902 rho-0.337(2.09) Appendix C8.c lnml, - -1.941 -0.003TIME + 0.3511ny, -0.867R, -0.231Y,-+0.7081n(Ml,4/P,) (-4.41) (-1.69) (7.90) (-3.52) (-9.09) (5.94) 9?-0.957, SEEx10-0.336, 0-w—1.951, rho-0.520(2.59) Sample Period-1973/I-l982/IV Appendix C8.d Alnml,4- 0.001ATIME + 0.398A1ny, -o.70559, -0.227AY,«+0.390Aln(M1,4/P,) (0.20) (11.43) (-2.29) (-10.93) (4.47) R?-0.732 SEEx10-0.393 0-w—2.192 186 Appendix D1 Recursive Regression and Simulation Results Constant 1ny, 1973/1 ~1982/IV -1983/IV -1994/1v -1985/IV -1986/IV -1997/1v -1988/IV -1999/11 -l.477 0. .34)(8. (4 -1. (4. -1. (4. -l. (4. -l. (4 -l. (5. -l. (5. -1. (5. 36 0. 33)(8. 330 0. 62)(8 292 0. 66)(9. 228 0. .77)(9. 352 0. 26)(9. 357 0. 53)(9. 342 0. 60)(9. 352 55) 361 30) 368 .97) 363 14) 355 28) 365 78) 379 74) 378 79) R, -0.930 (3 .18) -0.761 (2. -0. (2. -0. (2 -0. (2 -0. (2. -0. (2. -0. (2 66) 719 66) 644 .46) 602 .41) 642 638 55) .59) (Ml Level) . 1n Yt (M1,-1/P,) 92 SEExlO -0.226 0.584 0.957 0.338 (9.79)(6.25) -0.222 0.514 (9.14)(5.89) -0.224 0.478 (9.77)(6.10) -0.221 0.482 (9.91)(6.26) -0.214 0.477 (9.99)(6.59) -0.218 0.492 50)(10.28)(7.33) -0.221 0.458 (9.95)(6.95) 619 -0.218 0.454 (9.91)(7.25) 0.965 0.358 0.975 0.345 0.978 0.345 0.981 0.338 0.984 0.343 0.986 0.359 4 Quarters D-W rho 1.883 0.718 (5.44) 1.815 0.671 (4.83) 1.779 0.676 (5.09) 1.777 0.670 (5.12) 1.831 0.649 (5.08) 1.826 0.675 (5.54) 1.911 0.618 (5.11) 0.987 0.357 1.893 0.601 (5.02) RMSExlO 0.547 0.214 0.360 0.254 0.414 0.560 0.294 Appendix Alny, 1973/I 0. -1982/IV(9. -1993/1v 0. (10. -1994/1v 0. (10. -1985/IV 0. (11. -1986/IV 0. (11. -1987/IV 0. (11. -1999/1v 0. (ll -1989/II 0. (11. 367 98) 383 04) 399 94) 389 34) 383 55) 385 72) 398 .43) 398 55) D2 Recursive Regression and Simulation Results AR, .0. (3. -0, (2. -0. (2. -0. (2. -o, (2. -0. (2. -0. (2. -0 (2. 971 -0. 30)(10. 918 -0. 93)(10. 897 -0. 99)(ll. 890 —0. 99)(11. 863 -0. 95)(1l. 874 -0. 96)(11. 785 -0. 51)(ll .710 -0. 32)(11. AY, 187 (M1 First-Difference) Aln(Ml,-1/P,) 233 88) 235 50) 237 19) 238 61) 232 71) 232 78) 231 .06) 228 08) .549 .88) .521 .35) .505 .54) .531 .98) .529 .13) .542 .30) .422 .98) .395 .78) 187 0. R2 811 .789 .798 .798 .791 .781 .740 .737 SEExlO 0.355 0.379 0.366 0.364 0.359 0.362 0.390 0.390 D-W 4 Quarters 2.044 2.131 2.112 2.162 2.255 2.212 2.250 2.189 RMSExlO 0.570 0.181 0.349 0.301 0.399 0.724 0.416 188 Appendix D3 Recursive Regression and Simulation Results Constant 1ny, 1973/1 -0. -l982/IV(5. -1993/1v-0. (4 -1994/1v-0 (4 -1995/1v-0 (4. -1986/IV—0 -1987/IV-0 (4. -1988/IV-0 (4. -1999/11-0 527 0.128 04)(6. 489 0. .54)(7. .476 0. .45)(7. .466 0. 64)(8. .439 0. (4. 40)(7 .465 0. 78)(7. .478 0. 84)(7. .473 0 (4. 85)(7 88) 135 08) 142 83) 140 16) 132 .46) 132 75) 131 32) .129 .44) (M2 Level) . ln R, Y, (M2,-1/P,) R2 $991410 D-W -0 253 -0.093 0.999 0 997 0.126 2.060 (2.99) (9.10)(37.19) -0.181 -0.093 0.864 0.998 0.135 1.965 (2.06) (9.83)(36.61) -0.173 -0.095 0.845 0.998 0.135 1.972 (1.82) (9.51)(34.97) -0.166 -0.093 0.847 0.999 0.132 1.961 (1.86) (9.79)(39.40) -0.141 -0.088 0.956 0 999 0.137 1.939 (1.64) (8.87)(42.07) -0.152 -0.088 0.861 0.999 0.136 1.942 (1.74) (9.20)(43.87) -0.120 -0.086 0.866 0 999 0.143 1.930 (1.42) (8.53)(4S.88) -0.109 -0.085 0.968 0 999 0.142 1.927 (1.35) (8.61)(49.66) 4 Quarters rho RMSExIO 0.496 0.248 (3.53) 0.537 0. (3.95) 168 0.616 0. (5.00) 093 0.601 0. (5.16) 194 0.541 0. (4.64) 133 0.568 0. (5.11) 224 0.512 0. (4.65) 099 0.497 (4.59) 189 Appendix D4 Recursive Regression and Simulation (M2 First-Difference) Alny, AR, AY, A1n(M2,-,/P,) R2 SEExlO D-W 4 Quarters RMSExlO 1973/I 0.151 -0.329 -0.101 0.800 0.941 0.140 2.305 0.214 -l982/IV (8.49)(2.86) (10.98)(14.59) —1983/IV 0.151 -0.315 -0.099 0.813 0.937 0.148 2.327 0.111 (8.12)(2.59) (10.39)(l4.59) -1984/IV 0.155 -0.306 -0.100 0.806 0.939 0.145 2.291 0.116 (8.68)(2.57) (10.98)(15.10) ~1985/IV 0.152 -0.301 -0.098 0.806 0.940 0.143 2.278 0.239 (8.9l)(2.59) (11.22)(15.70) -1986/IV 0.145 -0.281 -0.093 0.810 0.928 0.152 2.289 0.118 (8.18)(2.27) (10.23)(15.13) -1987/IV 0.143 -0.277 -0.092 0.819 0.926 0.150 2.296 0.269 (8.37)(2.28) (10.47)(15.89) -1988/IV 0.147 -0.237 -0.092 0.792 0.910 0.160 2.274 0.141 (8.27)(l.85) (9.94)(14.98) -1989/II 0.146 -0.207 -0.090 0.787 0.909 0.159 2.277 (8.52)(1.66) (10.05)(15.47) 190 Appendix D5 Dufour Test Results (Ml) Level First-Difference Constant -1.475 (4.33) 1ny, 0.352 (8.55) Alny, 0.367 (9.98) R, -0.932 (3.18) AR, -0.971 (3.30) Y, -0.226 (9.79) AY, -0.233(10.88) ln(M,-1/P,) 0.584 (6.25) A1n(M,-1/P,) 0.549 (5.88) 1983/1 -0.108 (2.88)* -0.109 (2.98)* 2 -0.075 (1.56) 0.030 (0.85) 3 -0.072 (1.33) -0.000 (0.01) 4 -0.056 (0.93) 0.014 (0.38) 1984/1 -0.086 (1.36) -0.029 (0.80) 2 -0.103 (1.57) -0.017 (0.46) 3 -0.075 (1.19) 0.024 (0.67) 4 -0.078 (1.16) -0.003 (0.08) 1985/1 -0.089 (1.31) -0.009 (0.26) 2 -0.100 (1.45) -0.009 (0.26) 3 -0.149 (2.25)* -0.054 (1.47) 4 -0.105 (1.54) 0.041 (1.14) 1986/1 -0.137 (1.93) -0.029 (0.81) 2 -0.089 (1.26) 0.049 (1.37) 3 -0.110 (1.53) -0.022 (0.61) 4 -0.117 (1.56) -0.008 (0.22) 1987/1 -0.124 (1.62) -0.005 (0.15) 2 -0.087 (1.12) 0.039 (1.09) 3 -0.094 (1.21) -0.009 (0.24) 4 -0.020 (0.26) 0.073 (2.03)* 1988/1 -0.136 (1.56) -0.110 (2.86)* 2 -0.l21 (1.47) 0.015 (0.42) 3 -0.021 (0.26) 0.097 (2.68)* 4 -0.051 (0.60) -0.030 (0.83) 1989/1 -0.130 (1.41) -0.075 (2.00)* 2 -0.093 (1.07) 0.038 (1.02) R? 0.988 0.782 SEExlO 0.338 0.355 D-W 1.883 2.046 rho 0.718 (5.44) Mean Error x10 -0.933 -0.037 Mean Absolute Error x10 0.933 0.360 RMSExlO 0.985 0.480 Mean Square Error x 10 0.097 0.023 UM x 10 0.087 0.000 US x 10 0.000 0.001 UC x 10 0.010 0.023 191 Appendix D6 Dufour Test Results (M2) Level First-Difference Constant -0.527 (5.04) 1ny, 0.128 (6.88) Alny, 0.151 (8.49) R, -0.253 (2.98) AR, -0.329 (2.86) Y, -0.093 (9.10) AY, -0.101(10.98) 1n(M,q /P,) 0.889(37.l3) Aln(M,d/P,) 0.800 (14.59) 1983/l -0.041 (2.85)* -0.037 (2.59)* 2 -0.028 (1.64) 0.019 (1.28) 3 -0.024 (1.29) 0.006 (0.44) 4 -0.037 (1.94) -0.009 (0.63) 1984/1 -0.055 (2.81)* -0.017 (1.22) 2 -0.060 (2.98)* -0.003 (0.18) 3 -0.043 (2.21)* 0.015 (1.03) 4 -0.046 (2.34)* -0.000 (0.03) 1985/1 -0.053 (2.66)* -0.005 (0.38) 2 -0.036 (1.76) 0.020 (1.43) 3 -0.026 (1.32) 0.006 (0.42) 4 -0.038 (1.83) -0.009 (0.61) 1986/1 -0.056 (2.60)* -0.015 (1.04) 2 -0.018 (0.82) 0.041 (2.88)* 3 -0.021 (0.95) -0.002 (0.16) 4 -0.045 (2.02)* -0.022 (1.56) 1987/l -0.047 (2.06)* 0.002 (0.11) 2 -0.033 (1.40) 0.019 (1.31) 3 —0.016 (0.68) 0.016 (1.14) 4 -0.018 (0.79) -0.003 (0.18) 1988/1 -0.056 (2.31)* -0.031 (2.09)* 2 -0.038 (1.58) 0.022 (1.53) 3 -0.002 (0.06) 0.036 (2.55)* 4 -0.022 (0.91) -0.020 (1.39) 1989/1 -0.046 (1.77) -0.017 (1.16) 2 -0.029 (1.15) 0.021 (1.41) R2 0.999 0.930 SEExlO 0.126 0.140 D-W 2.099 2.324 rho 0.496 (3.53) Mean Error x10 -0.358 0.012 Mean Absolute Error x10 0.358 0.158 RMSExIO 0.386 0.194 Mean Square Error x 10 0.015 0.004 UM x 10 0.013 0.000 US x 10 0.000 0.000 UC x 10 0.002 0.004 Constant 1ny, 85 Y t ln(M,-1 /Pt) D2 D2lnY, 02R, D2Y, D21n(M,q_/P,) R2 SEEXIO D-W rho 192 Appendix D7 Interactions Test Results (M1) -1. 0 -0. -o. 0. -0 0. 1 Level 456 (4.87) .342 (8.28) 932 (3.61) 222 (9.68) 604(7.17) .496 (0.77) 104 (1.36) .309 (1.29) 0.013 (0.27) -0. CHOO 197 (1.44) .989 .327 .968 .638 (5.42) F(5.56) - 3.36** x25 -17 . 32** First-Difference Alny, 0.367 (10.04) AR, -0.971 (3.32) AY, -0.233(10.94) Aln(M,1/P,) 0.549 (5.91) D2AlnY, 0.079 (1.12) D2AR, 1.418 (1.26) D2AY, 0.030 (0.67) 02A1n(M,1/P,) -0.278 (1.66) 0.784 0.353 2.181 F(4.58) - 4.44** x3,-17.62** Appendix D8 Level Constant ~0.509 (4.89) 1ny, 0.124 (6.52) R, -0.242 (3.06) Y, —0.091 (8.62) 1n(M,q_/P,) 0.892(39.75) D2 -0.054 (0.23) D21nY, -0.014 (0.36) D2R, 0.350 (0.93) D2Y, 0.038 (1.68) D21n(M,q_/P,) 0.020 (0.48) R? 0 999 SEExlO 0.128 D-W 2.022 rho 0.414 (3.18) F(5.56) - 3.90** x25 -19.74** 193 Interactions Test Alny, 435 AY, A1n(M,_1/P,) D2AlnY, D2AR, D2AY, D2Aln(M,1/P,) Results (M2) First-Difference .151 (8.39) .329 (2.82) .101(10.85) .800 (14.42) .022 (0.65) .825 (1.89) .046 (2.44)* .051(0.52) .928 .014 .240 F(4.58) - 5.11** x24 -19 . 94H BIBLIOGRAPHY BIBLIOGRAPHY Akerlof, G. A. and Milbourne, R. D. (1978). ”New Calculations of Income and Interest Elasticitiies in Tobin's Model of Transactions Demand for Money". Review of Eggngmics and Statisticg 60 (November): 541-46. Artis, M. J. and Lewis, M. K. (1976). "The Demand for Money in the United Kingdom: 1963-73". fianchester School 44 (June): 147-81. Bank of Korea. Annual Repogg. Various Years. . Eggnohic Stat1§§ig§ Iggzhooh. Various Years. . Mghthly Bullehih. Various Issues. . Week Domest c and nternat onal c o . Mghey ghd Bagging Stagistics. 1984. . Suggey of Savings haghgt. 1984, 1989. No. 1400 (December) 1988. Barnett, W. S. (1980). ”Economic Monetary Aggregates: An Application of Index Number and Aggregation Theory". Journal of Econometricg 14 (September): 11-48. . (1982). "The Optimal Level of Monetary Aggregation". u o e ghedit, and Banking 12 (November) : 687-710. , Offenbacher, E. K. and Spindt, P. A. (1981). "New Concepts of Aggregated Money". Jghghal Qf Finghce 36 (May): 497-505. Barro, R. J. (1976a). ”Rational Expectations and the Role of Monetary Policy". Jougnal of Monetahy Ecohomics 2 (January): 1-32. (1976b). "Integral Constraints and Aggregation in Inventory Models of Money Demand". Jghghgl 9f Einance 31 (March): 77-88. and Gordon, D. B. (1983). "A Positive Theory of Monetary Policy in a Natural Rate Model". Jgurnal of Poligiggl Egohghy 91 (August): 589-610. Batten, D. S. and Thornton, D. L. (1983). ”M1 or M2: Which is the Better Moneytary Target7". Federal Reserve Bank of St. Louis, Review 65 (June/ July) 36-42. Baumol, W. J. (1952). "The Transactions Demand for Cash: An Inventory Theoretic Approach". thhterly Journal 9f Egohgmics 66 (November): 545- 56. . (1961). "Pitfalls in Contracyclical Policy: Some Tools and Results". Rgvigw hf Ecohgmigs ghd Shagistigg 43: 21-26. 194 195 Ben-Horim, M. and Silber, W. L. (1977). ”Financial Innovation: A Linear Programming Approach". Jgurnai 9f Banking and Einance 1 (November): 277-96. Blinder, A. S. (1986). ”More on the Speed of Adjustment in Inventory Models". MW 18 (August): 355-65. Boughton, J. M. (1979). "The Demand for Money in Major OECD Countries". OECD Economic Qntlook, Occasional Studies (January): 35-57. (1981). ”Recent Instability of the Demand for Money: An International Persvective". We]. 47 (January): 579-97. Brainard, W. (1967). ”Uncertainty and the Effectiveness of Policy“. e can Ecgngnig Revigw 57 (May): 411-25. Brown, R. L., Durbin, J. and Evans, J. M. (1975). "Techniques for Testing the Constancy of Regression Relationships Over Time". ggurnai 9f the Royal Stngisgicai Sogiety 37: 149-92. Brunner, K. and Meltzer, A. H. (1963). ”Predicting Velocity: Implications for Theory and Policy". Janina; oi Einangg 18 (May): 319-54. Cagan, P. (1956). "The Monetary Dynamics of Hyperinflation". In Sgnnies in WW1. ed. M. Friedman. pp. 25417. Chicago: University of Chicago Press. and Schwartz, A. J. (1975). "Has the Growth of Money Substitutes Hindered Monetary Policy?"- WM 7 (May): 137-59. Carlson, J. A. and Frew, J. R. (1980). ”Money Demand Regressions With the Rate of Return on Money: A Methodological Critique". Jgugnal oi Pglitigai Ennnnny 88 (June) 598-607. Carr, J. and Darby, M. R. (1981). "The Role of Money Supply Shocks in the Short-Run Demand for MOMY". WM 7 (September): 183-99. Chow, G. E. (1960). ”Tests of Equality between Sets of Coefficients in Two Linear Regressions". EEQQQEQEIIEQ 28 (July): 591-605. Cole, D. C. and Park, Y. C. (1983). Financial Develonment in Kogea, 1945-28. Cambridge: Harvard University Press. Cooley, T. F. and LeRoy, S. F. (1981). "Identification and Estimation of Money Demand". hnerican Economic Review 71 (December): 825-44. Davis, R. G. (1982). "Monetary Targeting in a 'Zero Balance' World". In Asilomar Conference on W. pp. 20-60. Federal Reserve Bank of San Francisco. 196 Dufour, J. M. (1980). ”Dummy Variables and Predictive Tests for Structural Change”. Economis_Le££er§ 6: 241-47. (1982). ”Predictive Tests for Structural Change and the St. Louis Equation". Proceedings of the Business and Economic Statistics Section, American Statistical Association. Fair, R. C. (1970). ”The Estimation of Simultaneous Equation Models with Lagged Endogenous Variables and First Order Serially Correlated Errors". Econometrica 38 (May): 507-16. Frenkel, J. A. (1977). ”The Forward Exchange Rate, Expectations, and the Demand for Money: The German Hyperinflation". c v w 67 (September): 653-70. Friedman, B. (1981). "The Relative Stability of Money and Credit 'Velocities' in the United States: Evidence and Some Speculations". NBER Working Paper No. 645. (1982). "Using a Credit Aggregate Target to Implement Monetary Policy in the Financial Environment of the Future”. Paper presented at the Federal Reserve Bank of Kansas City Conference on Monetary Policy Issues in the 19808. .(1983). ”The Role of Money and Credit in Macroeconomic Analysis”. In na2r2ec2n2mi2sl_£rise§l_ans_flsenti£ies. ed. J. Tobin. pp- 161-99- New York: Basil Blackwell. Friedman, Milton. (1956). ”The Quantity Theory of Money-A Restatement". In Srudiea_inatbe_Quan£i£x_1hecrx_2f_ngnex. ed. M. Friedman. pp- 3-21- (1957). A_Ihe2r1_2f_the_gonsuantign_£unction. Princeton: Princeton University Press for NBER. . (1959). ”The Demand for Money: Some Theoretical and Empirical Results". I2srnal_2£_figli£ical_ficgnomx 67 (August): 327-51. (1968). "The Role of Monetary Policy”. Angrigan Ecgnomig Rgvigw 58 (March): 1-17. (1969). Ihe_Q2simsa_Qsan2i51_2£_ugnex_ans_flrher_§§§axs. Chicago: Aldine. (1982). "The Yo-Yo Economy". Newgwgeh (Feburary 15), p. 72. . (1984). "Monetary Policy for the 19805". In Io Eggngtg Eognezigy; Q, S. pgngggig 2911c! in nhe hid-i9803, ed. John H. Moore, pp. 26-63. Stanford, California: Hoover Institution Press. and Schwartz, A. J. (1963). A_M2nsLaII_fl1at2Ix;2£_£he_uni£ed_§£e£esl 1862-1960. Princeton: Princeton University Press for NBER. (1982). Monggngy Iggnns in ghg Q. S. nng UI K, Chicago: University of 197 Chicago Press for NBER. Garbade, K. (1977). "Two Methods for Examining the Stability of Regression Coefficient8"- l2nrna1_2fltbe_Ameri2an_fitatistical_A§sccia£ign 72 (March): 54-64. Garcia, G. and Pak, S. (1979). "Some Clues in the Case of the Missing Money". American.£22n2mic_nexiex 69 (May): 330-34. Goldfeld, S. M. (1973). "The Demand for Money Revisited". n a e n Econ2m12_Ac£ixi£x 3: 577-538- (1976). ”The Case of the Missing Money”. ok n ers cono c w 3: 693-730. Goodfriend, M. (1985). ”Reinterpreting Money Demand Regressions". In gnggzstanding hgnetngy Regines, pp. 207-42. Carnegie-Rochester Conference Series on Public Policy 22. Amsterdam: North-Holland. Goodhart, C. A. E. (1981). ”Comment". In hnnghn;y_1n;ggn§, ed. B. Griffiths and G. E. Wood, pp. 129-31. New York: St Martin's Press. (1984). n2ne£a£1_Iheerx_and_£ractice. London: Macmillan- Gordon, R. J. (1984). "The Short-Run Demand for Money: A Reconsideration." e d a 16 (November): 403-34. (1985). ”The Conduct of Monetary Policy". In Mongtnry Pgiiny in Onr Iingg, ed. A. Ando et al, pp. 45-81. Cambridge, Mass.: MIT Press. (1986). Mnnxngnnnnning, 4th edition. Boston: Little, Brown. Gould, 8., Mills, J. and Stewart, S. (1981). unngnngi§n_g;_£;n§ngli;x. London: Macmillan. Gould, J. P. and Nelson, C. R. (1974). "The Stochastic Structure of the Velocity of Money". Angricnn Econgnic Review 64 (June): 405-18. Granger, C. W. J. and Newbold, P. (1974). ”Spurious Regressions in Econometrics”. Jgugngi 9f Egonomegzics 2 (July): 111-20. Griliches, Zvi. (1967). "Distributed Lags: A Survey". Ecgnometricn 35 (January): 16-49. Gurley, J. G. (1960). ”Liquidity and Financial Institutions in the Postwar Economy". Study Paper 14a, Joint Economic Committee, 86th Congress, 2nd Session, Washington. . and Shaw, E. S. (1960). hgney in g Ihgggy 9f Finangg. Washington D. C. :The Brookings Institution. 198 Hafer, R. W. (1981). ”Much Ado About M27". Federal Reserve Bank of St Louis, Rgxign 63 (October): 13-18. (1985a). ”Choosing Between M1 and Debt as an Intermediate Target for Monetary Polio?"- In W. PP- 89-132- (1985b). "The Stability of the Short-Run Money Demand Function, 1920- 1939". W 22 (July): 271-95. Hafer, R. W. and Hein, S.E. (1979). "Evidence on the Temporal Stability of the Hall, Demand for Money Relationship in the United States”. Federal Reserve Bank of St. Louis, Rngigg 61 (December): 3-14. (1980). "The Dynamics and Estimation of Short-Run Money Demand". Federal Reserve Bank of St. Louis, Revigw 62 (March): 26-35. (1981). "Investigating the Shift in Money Demand: An Econometric Anelyeio". In MW. 917- 1-28- Proceedings of a Conference held at the Center for the Study of American Business, Washington University. (1984). "Financial Innovations and the Interest Elasticity of Money Demand: Some Historical Evidence”. Journal of Money, J9n1nnl_nf_hnnny, W 16 (May): 2747-52. R. E. (1983). "Macroeconomic Policy under Structural Change". In 1 , pp. 85-111. Proceedings of a Conference held at the Federal Reserve Bank of Kansas City. Hamburger, M. J. (1977). "Behavior of the Money Stock: Is There a Puzzle2". Hart, WW 3 (July) : 265-88. (1982). "Financial Innovation and Monetary Targeting". In Einnnninl ..1.1°H:,-1 ' It, I”: 04110 et ,'0 24' _4= : 9: ‘ , pp. 105-10. Proceedings of the Seventh Annual Economic Conference. held at the Federal Reserve Bank of St. Louis. "Discussion" on Monetary Crises. A. G. (1982) "Comment". In Chisgs in ghe Egonnnig nnd Finangiai SEIEEEBIQ. ed. P. Wachtel, pp. 87-95. New York: Lexington Books./ D. C. Heath and Company. Hatanaka, Michio. (1974). "An Efficient Two-Step Estimator for the Dynamic Hein, Adjustment Model with Autoregressive Errors”. lnn;nnl_n£_§nnnnng§;in§ 2 (September): 199-220. S. E. (1980). "Dynamic Forecasting and the Demand for Money". Federal Reserve Bank of St. Louis, Rgvigw 62 (June/July); 13-23. and Veuglers, P. T. (1983). "Predicting Velocity Growth: A Time Series Perspectives". Federal Reserve Bank of St. Louis, Revien 65 (October): 34-43. 199 Heller, H. R. and Kahn, M. S. (1979). ”The Demand for Money and the Term Structure of Interest Rates”. Jgnnnni hf Eoiinigai Economy 87 (Feburary): 109-29. Hester, D. D. (1981). "Innovations and Monetary Control". figgokings Panerg on W 1: 141-89- Hetzel, R. L. (1984). "Estimating Money Demand Functions". Jou nal o e W 16 (May); 185-93. Hicks. J. R. (1967). W- Oxford: Clarendon Press. Hwang, H. S. (1985). ”Test of the Adjustment Process and Linear Homgeneity in a Stock Adjustment Model of Money Demand". Rgvigw of Econoning nng Shnhinhinn 67 (November): 689-92. Jaffee. 0- H. (1971). W. New York: Johnwiley. Johnston, J. (1984). Eggnongtrig Methods, 3rd edition. New York: McGraw-Hill. Judd, J. P. (1983). ”Deregulated Deposit Rates and Monetary Policy". Federal Reserve Bank of San Francisco, Eggnonic Rgview, (Fall): 27-44. Judd, J. P. and Scadding, J. L. (1981). "Liability Management, Bank Loans, and Deposit 'Market' Disequilibrium". Federal Reserve Bank of San Francisco, WM (Summer): 21-43. (1982a). "The Search for a Stable Money Demand Function: A Survey of the Post-1973 Literature". Journgi oi Egononic Literature 20 (Septem- ber): 993-1023. (1982b). "Financial Change and Monetary Targeting in the United States". In Asilomar Conference on I te e te Dere tio Enligy, pp. 78-106. Federal Reserve Bank of San Francisco. (1982c) "Dynamic Adjustment in the Demand for Money: Test of Alternative Hypotheses”. Federal Reserve Bank of San Francisco, Eggnomic Rnyigg (Fall): 19-30. Judd, J. P. and Motley, B. (1984). "The 'Great Velocity Decline' of 1982-83: A Comparative Analysis of M1 and M2". Federal Reserve Bank of San Francisco, Ennnnnin_figyign (Summer): 56-74. Kahn, M. S. (1973). "A Note on the Secular Behaviour of Velocity within the Context of the Inventory-Theroetic Model of Demand for Money". W22]. 41 (June): 207-13. (1977). "Variable Expectations and the Demand for Money in High- Inflation Countries". Manghgsger Schggi 45 (September): 277-93. 200 Kane, E. J. (1981). "Accelerating Inflation, Technological Innovations, the Decreasing Effectiveness of Banking Regulation". J2n;nnl_nfi_£ingnn§ 36 (May): 335-67. Kareken, J. H., Muench, T. and Wallace, N. (1973). ”Optimal Open Market Strategy: The Use of Information Variables". Angrican Economic Review 63 (March): 156-72. Klein, B. (1974). "Competitive Interest Payments on Bank Deposits and the Long-Run Demand for Money”. Anglin§n_figgnnni§_ggyigg 64 (December): 931-49. Kopcke, R. W. and Dockser, M. (1982). "Appendix” to "Do the Monetary Aggregates have a Future as Targets of Federal Reserve Policy?". Federal Reserve Bank of Boston, Ngn_Enginng_finnnnnin_figyign(March/ April): 12- 14. Kydland, F. E. and Prescott, E. C. (1977). "Rules rather than Discretion: The Inconsistency of Optimal Plans". l2nInnl_gfi_£nli§innl_finnnnmy 85 (June): 473-91. Laidler, D. E. W. (1980) "The Demand for Money in the United States- Yet Again". In Qn_the_State_2f_uacr2;E22n2mics. pp. 219-71. Carnegie- Rochester Conference Series on Public Policy 12. (1982a). Mnnghnzi§§_£§1§ngggiyg§. Cambridge, Mass.: Harvard University Press. (1982b). "Comments" on "Financial Changes and Monetary Targetting in the United States" by J. P. Judd and J. L. Scadding. Asilomer Conference on Intere§t_Rate_neregslatign_ans_ugnetarx_figlicx. pp- 127-33- . (1985). The_Demand_f2r_u2nsx1_Ihe2r1l_il_e11_Exigense_an§_zrgblema. 3rd edition. New York: Hafer & Row Publishers. Lane, T. D. (1984). "Instrument Instability and Short-Term Monetary Control". I2srnal_of_ngneterx_ficongmics 14 (September): 209-24. . (1985) "The Rationale for Money-Supply Targets: A Survey". c ter finhnni 53 (June): 179-207. Lieberman, C. (1977). "The Transactions Demand for Money and Technological Change"- Esxies.2f_Economic§_an§_§£etistics 59 (August): 307-17. (1979). "Structural and Technological Changes in Money Demand". American.£22n2mic_fiexies 69 (May): 324-29. . (1980). "The Long-Run and Short-Run Demand for Money Revisited". l C B ki 12 (Feburary): 43-57. Lindsey, D. (1977) "The Implications of Removing the Demand Deposit Rate 201 Prohibition for Monetary Control and the Conduct of Monetary Policy". Special Studies Paper No. 104. Board of Governors of the Federal Reserve System. Lucas, R. E., Jr. (1976). "Econometric Policy Evaluations: A Critique”. In Ihg Phi11ina_Qurxe_and_Lahgr_Markets. pp- 19-46. Carnegie-Rochester Conference Series on Public Policy l. (1978). "Asset Prices in an Exchange Economy". EQQDQEQEILQQ 46 (November): 1429-45. (1986) "Adaptive Behavior and Economic Theory". Jgurnai 9i aninggg 59 (October): 401-26. Paper presented at a Conference held at the University of Chicago. Marty, A. L. (1961). ”Gurley and Shaw on Money in a Theory of Finance”. I2srnal_gf_zelitical_§cen2mx 69 (May): 56-62- Mayor, T. H. and Pearl, L. R. (1984). "Life-Cycle Effects, Structural Change and Long-Run Movements in the Velocity of Money". gnuznai 9f unney, gggdit, nnd BanRing 16 (May): 175-84. McKinnon, R. I. (1973). d Ca ta E m c eve o . Washington D. C. : Brookings Institution. Mehra, Y. P. (1978). “Is Money Exogenous in Money Demand Equations7". innznni 2f.£211£12§l.§22n2m1 35 (April): 211-28- Meltzer, A. H. (1963) ”The Demand for Money: The Evidence from the Time Series". I2ernal_gi_zgliticel_ficsnomx 71 (June): 219-46. Milbourne, R. D. (1983). "Price Expectations and the Demand for Money: Resolution of a Paradox". Review of Ecgnonicg nnd Sgntiggigg 65 (November): 633-38. (1986). "Financial Innovations and the Demand for Liquid Assets". Jnnznni 9f hgngy, CIQQIE. and Ranking 18 (November): 506-11 Miller, M. H. and Orr, D. (1966). ”A Model of the Demand for Money by Firms". Qnarrerlx_iesrnel_2f_ficgngmics 80 (August): 413-35. Minsky, H. P. (1957). ”Central Banking and Money Market Changes”. ter Journal_2f_ficengmics 71 (May): 171-87. (1969). "Private Sector Asset Management and the Effectiveness of Monetary Policy”. Jouhnal 9i Einance 24 (May): 223-38. (1982). "Financial Innovations and Financial Instability: Observations and TheorY“- In Einan2ia1_Isnoretinn§1_Iheir_ImRact_2n_nenetarx_figlisx nnd Einanciai Marketg, pp. 21-41. Modigliani, F., Rasche, R. and Cooper, J. (1970). "Central Bank Policy, the 202 Money Supply, and the Short-Term Rate of Interest". Jpnnnni_nf_hnngyy W 2 (May): 166-218. Morris, F. E. (1982). "Do the Monetary Aggregates Have a Future as Targets of Federal Reserve Policy7”. Federal Reserve Bank of Boston, ng Rngiand Eggnonig Rgview (March/ April): 5-11. Nelson, C. R. and Plosser, C. I. (1982). ”Trends and Random Walks in Macroeconomic Time Series". lnnLnni_n£hflnng§n;y_§gnnnnin§ 8 (September): 139-62. Niehans, J. (1982). ”Innovation in Monetary Policy". lnnrn§1_nfi_§§nking_§ng Einnnng 6 (March): 9-28. (1983). ”Financial Regulation, Multinational Banking and Monetary Policy”- I2srna1_2f_hanking_and_fiinance 7 (December): 537-51. Offenbacher, E. K. (1981). "Two Papers on Money Demand”. Special Studies Paper No. 157. Federal Reserve Board of Governors. Phelps, E. S. (1968). "Money Wage Dynamics and Labor Market Equilibrium". l2urnal_2f_£211tical_flconomx 76: 678-711- Pierce, J. L. (1982). ”Panel Discussion" on "Alternative Intermediate Instruments of Monetary Policy". In Asilomar Conference on Ingehggg Rate Dereg2latign_ang_ngnererx_zglicx. pp. 215-220. Pindyck. R. and Rubinfeld. D. (1981). E22n2metric_nggela_ené_ficcn2mic Egrggnggn. New York: MacGraw-Hill. Plosser, C. I. and Schwert, G. W. (1978). "Money, Income, and Sunspots: Measuring Economic Relationships and the Effects of Differencing.” I2arnal.2flugnetarx_522ngmics 4 (November): 637-60. Podolski, J. M. (1986). Einnngini innovntign nng the honey Supply. New York: Basil Blackwell Inc. Poole, W. (1982). "Discussion" on ”Monetary Targeting in a 'Zero Balance' World" by Davis, R. G. In Asilomar Conference on Ingepgst Rnte Dereg21a£129.9ne_ngnetarx_£2112x. pp- 61-69- Porter, R. D. and Offenbacher, E. K. (1982). "Financial Innovations and Measurement of Monetary Aggregates". In Einnnnini_innnyn§ipn§;_1hgi; inpngt 9n honegary Rgiicy and Financiai Marketg, pp. 49-98. (1983). "Empirical Comparisons of Credit and Monetary Aggregates Using Vector Autoregression Methods". Federal Reserve Bank of Richmond, Egonomig Review 69: 16-29. Porter, R. D., Simpson, T. and Manskopf, E. (1979). "Financial Innovation and Monetary Aggregates". fizpphings Rapei§ in Epongnip Aphivity 1: 213-29. 203 Quandt, R. E. (1960). "Tests of the Hypothesis that a Linear Regression Obeys Two Separate Regimes". ou a t er c at st Ass at 55 (June): 324-30. Rasche, R. H. (1986). "Ml-Velocity and Money Demand Functions: Do Stable Relationships Exist7'. Econometrics and Economic Theory Paper No. 8705, Michigan State University. Paper persented at the Carnegie-Rochester Conference Series on Public Policy 27, 1987. . (1988). "Monetary Policy and Financial Deregulation in the United StateS". Ere—diLRLLKapitel 21: 451-68 and Johannes, J. M. (1987) ”Controlling the Growth of Monetary Aggregates”. Boston, Mass.: Kluwer Academic Publishers. Roley, V. V. (1982). "Comments" on "Financial Changes and Monetary Targeting in the United States” by J. P. Judd and J. L. Scadding. In Asilomar Conference on WM. PP- 134- 38. (1985). ”Money Demand Predictability" WW9 finnhing 17 (November): 611-41. Paper presented at the Enterprise Institute Conference on Monetary Policy in a Changing Financial Environment. Samuelson, P. (1970). "Reflections in Recent Federal Reserve Policy". inngni of Money, Shedig, nnn Snnhing 2 (Feburary): 33-44. Sargent, T. J. (1976). ”A Classical Macroeconomic Model for the United States." WW9: 8‘4 (April): 207-38. and Wallace, N. (1975). ”Rational Expectations, the Optimal Monetary Instrument and the Optimal Money Supply Rule”. l9n;nni_nf_£nli§innl Ennnnny 83 (April): 241-54. Schmidt, P. and Waud, R. N. (1973). "The Almon Lag Technique and the Monetary Versus Fiscal Policy Debate”. t e e c s A§§2§l§£12£ 68 (March): 11-19. Schumpeter, J. A. (1964). Enginesg Sycieg. New York: McGraw-Hill. ShaW. E. (1973). MW. Oxford: Oxford University Press. Shin, H. C. (1986). "The Money-GNP Relationship in Korea". Paper presented at the Seventh Pacific Basin Conference on Economice Modeling, Sydney, Australia. Silber, W. L. (1975). "Towards a Theory of Financial Innovations“ In Einnncial Innpyngipng, ed. W. L. Silber, pp. 53-85. Farnborough, Lexington Books. (1983). ”The Process of Financial Innovations". e c o c 204 Rgvigw 73 (May): 89-95. Simmons, H. C. (1936). "Rules versus Authorities in Monetary Policy". innznni Wm 4‘4 (Feburary): 1-30. Simpson, T. D. (1979). ”A Proposal for Redefining the Monetary Aggregates". WW1: 65: 13-42. . (1980). "The Redefined Monetary Aggregates". ede a es rve 1e n 66: 97-114. and Porter, R. D. (1980) "Some Issues Involving the Definition and Interpretation of the Monetary Aggregates". In Qpnnhniiing_§hg_hpng§§ry Aggzggn§§§_111, pp. 161-234. Federal Reserve Bank of Boston Conference Series 23. Spencer, D. E. (1985). “Money Demand and the Price Level". v w 0 cs Matias 67 (August): 490-96. Taylor, J. B. (1985). ”What Would Nominal GNP Targeting Do to the Business Cycle?” In W. PP- 61-84- Teigen, R. L. (1964). “Demand and Supply Functions for Money in the United States: Some Structural Estimates". Ennnnnggpipn 32 (October): 476-509. Theil, H. (1966). Appiigfi_£nnngnin_£g;gnn§§ing. Armsterdam: North-Holland. (1971). W. New York: John Wile?- Tinsley, P., Spindt, P. and Friar, M. (1980). "Indicator and Filter Attributes of Monetary Aggregates: A Nit-Picking Case for Disaggregation". Special Studies (mimeo). Board of Governors of the Federal Reserve Systems. Tobin, James. (1956). "The Interest Elasticity of Tansactions Demand for Cash". Rgvigw of Ecgnomicg ang Statiggicg 38 (August): 241-47. . (1980). "Stabilization Polity After Ten Years”. fignnhing§_£npg;§_2n W 1: 19-72- (1983). "Monetary Policy: Rules, Targets, and Shocks". inn;nn1_nf Money, Qredit, nnd Ranking 15 (November): 506-18. (Feburary): 57-83. Viner, Jacob. (1962). "The Necessary and Desirable Range of Discretion to be Allowed to a Monetary Policy". In Sgnzph pi n Mpnggagy anstitugion, ed. Yeager. L. B., pp. 244-74. Cambridge: Harvard University Press. Wai, U. T. (1957). "Interest Rates Outside the Organized Money Markets of Underdeveloped Countries". IMF Staii Rnpegg 6 (November): 80-142. Wenninger, J. (1984). "Financial Innovation - A Complex Problem Even In a 205 Simple Framework". Federal Reserve Bank of New York, anrteziy Revigw 9 (Summer): 1-8. and Sivesind, C. M. (1979). "Defining Money for a Changing Financial System”. Federal Reverve Bank of New York, Qnarpeziy Rgview 4 (Spring): 1-8. White, W. H. (1978). "Improving the Demand-for-Money Function in Moderate Inflation”. 1h£_S;§fi£_£np§;§ 25 (September): 534-72. (1981) "The Case for and against 'Disequilibrium' Money". F ta Rnpgpg 28 (September): 564-607. Williams, D. (1978). I'Estimating Levels or First Differences: A Defense of the Method Used for Certain Demand for Money Equations". e c o qunnai (September): 564-68. Wilson, A. L. (1978). "When is the Chow Test UMP?" Ihg_Ang;nnn_S§n§i§§iginn 32 (Feburary): 66-68. Wojniloner, A. M. (1980). ”The Central Role of Credit Crunches in Recent Financial Hietory"- Wu 21: 277-326- "7'11 11911171111999"