_

I:;...- I
: .2 v.

j. . ‘ . am

.5.

 

”nun":

Rh . . .

again ...
«3.51.564.

3
.
.

5.. . 3‘
35.: 5.72:... v

9 a
. . I i...“-
Juana...

z. .I
.313. a}!

1371!...7’2‘0...

:

5. he“. a... 551.12....
: a. 1.. .
:mxt ‘vh. finialuflok s
.. .. . . a
it: fligwi :3. lat...
, r..:x...l§§t?. (0655
31‘:

"haiku” .u z A...“ 3..
2» 3.x.3alfi 15-...

11.4 ...I.l.2|§..l

I); fall.
ti)! . {833‘
{3.9.5.922 ‘0. ‘2!

)N x...”
2).! 2.1 :3.
.tisastztxi... fi.?..i::).~i= ...
:u7x_u.flxsf.§x.‘..! .vdrfikht- :2
x . . 3.1.2231. it 2-!
3.55.1531, $4. . . 1....qu

 

35.)}: ve$|’§111.( .
3.3.3333...

a I I .
1...; (59...... ’11

 

 

 

 

 

 

 

 

3.3119
)ér.?t.i§..rz‘
. .1} v T

 

V
. .32?“ a
cum-mm. 54..."? in

 

.- ,
.15.?

.3 ix

fig ma. fisfi

 

 

1HEs£

Q

\
5: Us

This is to certify that the

dissertation entitled

Three Essays on the Microstructure of
the Turkish Stock Market

presented by

Sadettin Aydin Yuksel

has been accepted towards fulfillment
of the requirements for

Ph . D . degree in Einance

Maj professor

Date 7/26/00

 

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

 

 

 

LIBRARY
University

 

 

 

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

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 c-JCIRC/DaieDmpGS—pjs

 

 

 

THREE ESSAYS ON THE MICROSTRUCTURE
OF THE TURKISH STOCK MARKET

By

Sadettin Aydin Yuksel

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Department of Finance

2000

ABSTRACT

THREE ESSAYS ON THE MICROSTRUCTURE
OF THE TURKISH STOCK MARKET

By

Sadettin Aydin Yuksel

This is the first study to use intraday data from the Istanbul Stock
Exchange (ISE). The purpose of the study is to investigate short-term price
dynamics in this emerging market. The data set covers the 30 most active stocks
on the ISE over 14 months and contains about nine million transactions. The
ISE is an order-driven market and its most distinguishing feature is the tick rule it
employs. It appears that the ISE aims to keep a large relative tick constant for all
price levels.

One goal of the study is to determine whether lSE’s large tick restricts
trader behavior more than do tick rules in other markets. The results suggest
that traders on the ISE use predominantly one-tick and occasionally two-tick
rounding. Perhaps due to the narrow tick regime width, no variation in clustering
is found within a tick regime. In addition, the examination of spread and price
change frequencies reveals they hardly ever exceed the tick size, which
indicates that the tick size is binding.

The literature states that mean reversion in short-term price dynamics is
necessary to make limit order trading an attractive strategy. A trading rule used

in studies of the New York Stock Exchange (NYSE) and Paris Bourse is

employed here to compare the expected profitability of limit and market orders.
Unlike the NYSE and the Paris Bourse, the ISE has excessive marketwide price
movements and these hide the short-term mean reversion in price. The market-
adjusted returns show that, on average, executed limit orders perform better than
market orders, but the opposite holds for unexecuted limit orders. A comparison
of the fraction of executed limit orders in these three markets reveals that prices
are more volatile on the ISE than on the other two exchanges. This can be
attributed to two factors. First, the absence of an opening call auction may
negatively affect price discovery on the ISE. Second, there may be insufficient
depth in the limit order book, which is unexpected, given the use of a large
relative tick by the ISE. Therefore, one can hypothesize that the weak balance
between limit and market order submission rates may be one reason for the
ISE’s choice of an unorthodox tick rule.

Finally, the short-term relationship between price and volume is
examined. Both display strong intraday and weak interday variability on the ISE.
Similar to other equity markets, the ISE displays an asymmetric
contemporaneous relationship between volume and volatility. Moreover, there is
a feedback relationship in the Granger sense between volume and volatility.

This differs from Jain and Job (1988), who found a unidirectional relationship

(VOIatility causes volume) on the NYSE.

 

 

ACKNOWLEDGMENTS

I would like to thank my dissertation chair, Dr. G. Geoffrey Booth, for his
guidance from start to finish. I have benefited immensely from our discussions. I
am also grateful to him for his patience, for his strong moral support, and for
exerting the right amount of pressure. Needless to say, this dissertation could
not have been done without him. I also thank the other members of my
committee, Dr. Kursat Aydogan, Dr. Kirt Butler, and Dr. Ted Fee, for their helpful
comments throughout the process.

I am very grateful to Dr. S. Tamer Cavusgil for his enormous support in
very difficult times. I thank my friend Umit Mersinli, of the Istanbul Stock
Exchange, for providing data. I am grateful to my friend Yigit Kavurmacioglu for
his help in improving my programming skills and to my friend Dr. Salim Hiziroglu
for being there whenever I needed moral support. Finally, I am grateful to my

parents, to whom this dissertation is dedicated.

TABLE OF CONTENTS

LIST OF TABLES ....................................................................................... viii
LIST OF FIGURES ..................................................................................... xi
LIST OF DETAIL TABLES .......................................................................... xii
CHAPTER 1 Introduction ....................................................................... 1
1.1 Purpose and Relevance of the Dissertation ....................... 7

1.2 Contributions of This Study ................................................ 8

CHAPTER 2 The Istanbul Stock Exchange ............................................ 12
2.1 History of the Securities Market in Turkey ......................... 12

2.2 Organization of the ISE ...................................................... 13

2.3 Trading Mechanism for Stocks ........................................... 14

2.4 ISE Member Firms ............................................................. 19

2.5 Foreign Investment on the lSE ........................................... 19

2.6 Data .................................................................................... 20
CHAPTER 3 Price Clustering on the Istanbul Stock Exchange .............. 25
3.1 Introduction ........................................................................ 25

3.2 Literature Review ............................................................... 28

3.2.1 Early Evidence of Clustering ................................... 28

3.2.2 An Overview of Clustering Studies .......................... 29

3.2.3 The Extent of Clustering in Different Markets .......... 30

3.2.4 Determinants of Clustering ...................................... 31

3.2.5 The Effect of a Change in Tick Size ........................ 35

 

3.3
3.4
CHAPTER 4
4.1

4.2

4.3

4.4

4.5

CHAPTER 5

5.1

5.2

3.2.6 Relation to Other Microstructure Aspects ................
Empirical Analysis ..............................................................
Summary ............................................................................
Limit Order Profitability on the Istanbul Stock Exchange
Introduction ........................................................................

A Comparison of the ISE with the Paris Bourse and the
NYSE .................................................................................

The Viability of a Pure Limit Order Market .........................

4.3.1 The Tradeoff between Market and Limit Order
Strategies ................................................................

4.3.2 The Method .............................................................
4.3.3 The Experimental Design ........................................
4.3.4 Limitations of the Method ........................................
4.3.5 Empirical Evidence in Previous Studies ..................
Empirical Analysis ..............................................................
4.4.1 Data .........................................................................
4.4.2 Results ....................................................................
Summary ............................................................................

Relationship between Trading Volume and Price Change
on the ISE ..........................................................................

Introduction ........................................................................
Literature Review ...............................................................
5.2.1 Early Research on Price-Volume Relationship .......

5.2.2 More Recent Work ..................................................

vi

36
40
49
51

51

54

60

60
61
66
68
70
73
73
75

87

89
89
91
93
98

5.2.3 Intra and lnterday Patterns in Volume and Return

Volatility ................................................................... 102

5.3 Empirical Analysis .............................................................. 104

5.3.1 Introduction .............................................................. 104

5.3.2 Data ......................................................................... 105

5.3.3 Time of Day and Day of Week Effects .................... 108

5.3.4 Contemporaneous Price-Volume Relationship ........ 110

5.3.5 Causality .................................................................. 113

5.4 Summary ............................................................................ 116
CHAPTER 6 Summary ............................................................................ 117
APPENDIX A Tables ................................................................................ 123
APPENDIX B Figures ............................................................................... 180
APPENDIX C Detail Tables ...................................................................... . 188
LIST OF REFERENCES ................................................. ‘ .............................. 199

vii

Table
Table
Table
Table
Table
Table

Table

Table

Table

Table

Table

Table
. Table

Table

Table

Table

Table

10

11

12
13

14

15

16

17

LIST OF TABLES

Annual ISE Trading Activity, 1986-1998 .................................... 124
Monthly ISE Trading Activity during the Sample Period ............... 124
Comparison of Emerging Markets by Size, in Millions of Dollars 125
Number of Stocks Listed on the ISE ............................................ 126
Types of Normal Order ................................................................. 126
Minimum Trade Size for Special Orders ...................................... 127
Number of ISE Members Authorized to Trade on the Stock

Market, 1986-July 1999 ................................................................ 127
Top 20 ISE Brokerage Houses Sorted by Total Transaction Value
Generated, January-December 1998 ........................................... 128
Top 20 ISE Brokerage Houses Sorted by Wholesale and Special
Transactions, January-December 1998 ....................................... 129
Equity Portfolio Holdings by Foreign Investors, 1995-1999, in

Millions of Dollars ......................................................................... 130
Total Market Value of Companies Traded on the ISE, 1986-1999,

in Millions of Dollars ..................................................................... 130
Some Characteristics of ISE Stocks in the Sample ........................ 131
Weekdays without Trading during the Sample Period ................. 132
Number of Sessions Stocks were Traded during the Sample

Period ........................................................................................... 133
Monthly Volume Generated by Foreign Investors during 1998

among Stocks in the Sample ........................................................ 134
Stock Split Adjustments for Sample Firms during the Sample

Period ........................................................................................... 135

Dividend Adjustments for Sample Firms during the Sample Period 136

viii

Table

Table

Table
Table

Table

Table
Table
Table
Table
Table

Table

Table
Table

Table

Table

Table

Table

Table

Table

18
19

20
21
22

23
24
25
26
27

28

29
30

31

32

33

34

35

36

Number of Transactions for Sample Stocks ................................. 137
Maximum Number of Shares Limit per Month for Sample Stocks

during the Sample Period, in Lots .................................................. 139
Maximum Trade Size in Lots for Normal Orders (Type 1) ............ 140
Identification of Special Orders .................................................... 140
The Tick Rule on the ISE and Six Other Exchanges that Use Step
Functions ............................................................................. 141
The Extent of Clustering in Different Markets ............................... 143
Determinants of Clustering .................................... -. ...................... 144
Distribution of Tick Size per Sample Stock .................................. 145
Actual Frequencies ....................................................................... 147
Frequencies Adjusted for Serial Dependence in Price ................. 150
Distribution of Bid-Ask Spread as a Multiple of Tick Size for

Sample Stocks .............................................................................. 153
Price Change in Consecutive Transactions ................................ 154
Variation in Clustering within Tick Regimes ............................ 155

Parameter Values and the Number Of Stock-Windows Used in the

Experiment ................................................................................... 1 59
The Number of Stock-Windows Conditional on Execution and
Nonexecution of Limit Orders ....................................................... 159
Average Standardized Execution Price of Limit Orders ............... 160
The Difference between Execution and Triggering Prices for

Executed Limit Orders .................................................................. 161
Average Time to Execution in Hours for Naturally Executed Limit
Orders .......................................................................................... 161
Unconditional and Conditional Returns for Market and Limit

Orders .......................................................................................... 162

Table
Table
Table
Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

Table

37
38
39
40

41

42

43

45

46

47

48

49

50

51

Differential Returns ....................................................................... 163
Standardized Price Following the Order Submission Day ............ 165
Sensitivity of Costs to the Choice of Investment Vifindow Length 166
Market-Adjusted Differential Limit Buy Order Returns .................. 167

Sensitivity of Market-Adjusted Differential Limit Buy Order Returns
to the Choice of Investment Window Length .................. 168

The Experiment of Simultaneously Submitting Limit Buy and Sell
Orders at :t k Ticks from the Equilibrium Price ............................. 169

Frequency of Trade Sequences in Which Price Changed
Consecutively ......................................................................... 1 70

Distribution of Intervals Based on Price Change and No Price
Change Classification ................................................................... 172

Stationarity of Price and Volume Series ....................................... 173

Average Turnover during 15 Minute Intervals by Weekday, in

Percentage ................................................................................... 174
Average Return Volatility during 15 Minute Intervals by Weekday
(x104) ............................................................................................ 175
Average Return during 15 Minute Intervals by Weekday ............. 176
Regression Results for the Test of a Contemporaneous

Relationship between Volume and Return ................................... 177
Granger Causality: Relationship between Return and Volume 179

Granger Causality: Relationship between Return Volatility and
Volume ......................................................................................... 179

Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Figure 7
Figure 8
Figure 9
Figure 10
Figure 11
Figure 12

LIST OF FIGURES

Standardized Price Level of ISE100 Index and Equal Weighted

Portfolio of Sample Firms ........................................................... 181
Tick Rule on the ISE .................................................................. 182
Distribution of Nontrivial Spread over Time ................................ 182
Timing of Events ........................................................................ 183
Standardized Price Following the Order Submission Day ......... 183
Index Price Level over Time ....................................................... 184
Percentage Turnover over Time ................................................ 184
Index Return over Time .............................................................. 185
Index Volatility over Time ........................................................... 185
Average lntraday Turnover ........................................................ 186
Average lntraday Volatility ......................................................... 186
Average lntraday Return ............................................................ 187

xi

Table

Table

Table

Table

Table

Table

52

53

54

55

56

57

LIST OF DETAIL TABLES

Unconditional and Conditional Returns for Market and Limit
Orders, Investment Window of 5 Days ...................................

Differential Returns, Investment Window of 5 Days ...............

Unconditional and Conditional Returns for Market and Limit
Orders, Investment Window of 7 Days ..............................

Differential Returns, Investment Window of 7 Days .............

Market-Adjusted Differential Limit Buy Order Returns,
Investment Window of 5 and 7 Days ................................

Details for the Experiment of Simultaneously Submitting Limit
Buy and Sell Orders at :l: k Ticks from the Equilibrium Price

xii

189

190

192

193

195

196

 

Chapter 1

Introduction

During the last two decades many factors have contributed to increased
competition in cross-border financial exchanges. The most important of these are
improvement in information technology; elimination of foreign investment
barriers; an increased supply of equities, partly fueled by the large number of
privatizations; and investor demand for international diversification. To reduce
the transaction costs of their stocks, world-class companies seek listings in all
major markets. In addition to the international competition, exchanges in
developed countries face domestic competition. For example the New York
Stock Exchange (NYSE) competes with Nasdaq, regional exchanges, third-
market firms that make over-the-counter markets in NYSE stocks, and crossing
networks (such as POSIT, the Crossing Network, and Instinet) that allow
investors to cut out the middlemen.

The increased competition forces organizers of these markets to
reconsider the optimality of their design. Although there are similarities across
exchanges, many significant differences can be observed. The trading
mechanism may be either a call or continuous auction, or a market may open
with one mechanism and switch to the other later in the trading day.1 A market

may be order or quote driven. It may use floor trading, electronic trading, or

 

‘ For example, the New York, Milan, and Helsinki stock exchanges.

 

both. Players in the market may differ. A continuous market may have a
specialist (called a registered trader on the Toronto Stock Exchange) and floor
traders, as does the NYSE, or it may simply match the order of public traders, as

does the Stock Exchange of Singapore. Rules on transparency, short selling,
maximum price change limit, and settlement date are all factors about which an
exchange organizer must decide.

One feature of an exchange is the tick rule it employs. The minimum
price increment (tick) determines what prices traders use. In most markets this
increment is a decimal fraction, but for the NYSE and the American Stock
Exchange (AMEX) it is based on negative powers of two (currently 1/16)? Some
markets, such as the London Stock Exchange, have only informal customs that
dictate the minimum price increment. Even among exchanges with a formal tick
rule, there are differences. Some use single absolute tick size for all the listed
stocks, whereas others employ a tick rule that is a step function of stock price.
Among others, the London gold market and the major North American
exchanges (NYSE, AMEX, and Toronto Stock Exchange, except for stocks with
extremely low prices) have single absolute tick size; Helsinki, Hong Kong,
Singapore, Sydney, and Tokyo stock exchanges are examples of markets that

use step functions.

 

2 It seems likely that the US stock markets will soon switch to decimal pricing.
Some newspapers have already started to report prices in decimals.

 

The effect of tick size on trader behavior as well as the quality and
competitiveness of markets has been the subject of academic and industry
debate. Tick rules changed very infrequently in equity markets until the 19905.3
At least partially due to concerns about competitiveness and the increase in
academic research in this area, owing to the availability of detailed trade data,
stock exchanges have begun to experiment with tick rules. During the last
decade, exchanges in Australia, Singapore, Hong Kong, Canada, and the United
States have decreased tick size.

A stream of research analyzes how traders use the sets of discrete prices
available to them. The existence of minimum price variation provides a natural
benchmark in assessing trader behavior directly from prices. This literature
dates back to Osborne (1962), and shows that asset prices deviate from random
walk, and some prices are observed more frequently than others when
underlying asset value is uniformly distributed over the range of feasible prices.
The so—called clustering literature presents evidence of and explanations for this
anomaly. Price clustering is shown to exist in all markets analyzed thus far,
although its extent varies across markets.

An economic explanation of this anomaly comes from Ball at al. (1985)
and Harris (1991). Ball at al. developed the price resolution hypothesis, that is,

clustering is the manifestation of the haziness about asset values. In other

 

3 In some other markets, such as commodity futures, rule changes have been
quite common.

words, investors will use coarser discrete price sets than the set based on the
regulated minimum price variation if they cannot determine the true price of an
asset with enough precision. Harris took the argument a step farther and
pointed out that even if investors can determine asset values with high precision,
they may avoid using a fine price set if it is costly to do so. His negotiation
hypothesis states that the use of a coarser grid arises from the incentive to lower
negotiation costs, which include the time it takes to strike a bargain and the extra
amount of information that traders need to track when they use a fine grid (such
as close attention in recording). The extent to which traders use a coarse grid
depends on the price resolution in the market. A high resolution will result in a
small dispersion among traders’ reservation prices, and if a coarse grid does not
include a price acceptable to both parties in a trade, gains from trade may be
lost. The negotiation hypothesis focuses on the tradeoff between lower
negotiation costs and lost gains from trade.

Research on tick size and trader behavior emphasizes a market’s liquidity
and competitiveness as well as the reaction of issuing firms. Depending on
market structure - whether quote driven, order driven or both - liquidity may be
provided by exchange members as well as by public traders who submit limit
orders. A market is quote driven if dealers announce the prices at which other
market participants can trade; it is order driven if some investors, by placing limit
orders, establish the prices at which other participants can buy or sell shares. In
a quote-driven market, member firms, which provide market-making services,

should earn a minimum amount of profit to cover their fixed costs. Anshuman

4

and Kalay (1998) estimate annual discreteness-related profits and show that
observed minimum tick size in US. exchanges is consistent with maximization of
member profits.

In an order-driven market, public traders can choose to trade via limit
order and supply liquidity to the market, or they can choose to trade via market
order and demand liquidity from the market. In order for a public trader to forgo
the immediate execution of his order, there should be some compensation for
trading via limit order.

The problem with liquidity provision is that a limit order/dealer quote
reveals part of the information owned by that trader and creates a free-rider
problem. Without a minimum tick, given a strict price-time priority rule,
competition among liquidity providers will result in front-running on existing limit
orders/dealer quotes by other investors and/or dealers. The tick size poses a
barrier to competing forces, thereby creating positive expected profits for market
makers, as pointed out by Grossman and Miller (1988). In exchanges that use a
sharing rule rather than enforce strict time priority, the effect of front-running may
be attenuated, as suggested by Porter and Weaver (1997). To minimize this
problem, some exchanges, such as the Paris Bourse, allow traders to display
part of their limit orders, which decreases the transparency of the market.

Another issue is the effect of tick size on interrnarket competition. A
mandatory tick size may cause an exchange to lose part of the order flow to
other trading mechanisms. The emergence of such nonprice competition

practices as preferencing (also known as payment-for-order flow) and its

 

extreme case, called internalization, are at least partially attributed to mandatory
tick size. Brokers direct orders to other market makers off the floor and receive
payments in return. Chordia and Subrahmanyam (1995) show that these
practices may lead to inferior order execution. They argue that a decrease in tick
size will make competition for order flow more transparent, and orders will flow to
the least-cost liquidity provider.

All these issues affect transaction costs, and this will be of concern to the
issuing firms. Amihud and Mendelson (1986) found that higher transaction costs
as measured by bid-ask spreads are associated with higher rates of return. In
general, however, firms have some flexibility in adjusting relative tick size of their
stocks by using stock splits. Angel (1997) provides us. and international
evidence in support of the view that one motive for splits is to move to the
optimal relative tick size.

Various empirical studies examine the effect of tick size on trader
behavior indirectly by measuring the change in a number of market quality
variables caused by an event that alters relative tick size. These studies
examine three events: the modification by an exchange of its tick rule; stock
splits, and, in markets that use a step function, the transition of stocks into a
different tick regime as their prices change. The market quality variables are the
size of bid-ask spread, market depth at the best quotes or beyond those, and
trade volume. The empirical results are mixed. All studies find a positive
relationship between relative tick size and bid-ask spread, but there is limited

evidence that the total depth in the limit order book decreases with an increase

6

 

in relative tick size. However, no significant change in either the trading volume

or dealer profits has been reported.

1.1 Purpose and Relevance of the Dissertation
Because of data limitations, most of the evidence in this area is from

North American markets. In the United States, for instance, intradaily transaction
data first became widely available in the mid-19803, and access to quotation
data was made possible at the beginning of the 1990s. In recent years,
automation has expanded data accessibility, and evidence is starting to
accumulate for other markets. Because some of these exhibit features that are
not shared by the major markets, an examination of them advances the market
microstructure literature by providing a different environment for analysis. This
dissertation examines an emerging market that represents a polar case because
of its apparently large tick size relative to stock price. On the Istanbul Stock
Exchange (ISE), relative tick size is at least 120% and as much as 2,200%
greater than that used by other exchanges. No previous analysis has examined
the issues related to the lSE's tick size choice, and empirical evidence from this
market may be of interest to policy makers there as well as to policy makers in
other markets who question the optimality of their tick size. The purpose of this
study is to investigate short-term price dynamics on the ISE.

Three issues that are related to the choice of tick size are addressed in this
research. First, mandatory minimum price change rules should not restrict trader

behavior. Clustering studies measure the effect of tick size on trader behavior

directly from prices. This issue is of interest to regulators in order to assess the
optimality of their tick rule.

Second, it is argued that mean reversion in short-term price dynamics is
a necessary condition for the limit order trading strategy to be attractive (Handa
and Schwartz 1996). Tick size is related to the magnitude of mean reversion
due to its effect on the front-running practices of limit order submitters.
Therefore, a tick rule can serve as a policy variable to market regulators for the
purpose of affecting the extent of liquidity in a market.

Third, a large tick size causes observed prices to differ from underlying
stock values. Consequently, the speed and precision of the price adjustment
process is likely to be hampered, which may affect the short-term relationship

between price and trading volume. This is important to investors.

1.2 Contributions of This Study

This dissertation advances the clustering literature by identifying and
examining a market that uses a large relative tick for all price levels. The goal of
Chapter 3 is to determine whether the tick size is as large as it appears to be.
Depending on market conditions, a seemingly large tick may turn out to be an
optimal choice.

One explanation for the lSE’s large tick is that the price resolution in this
market is low, and it is consistent with the typical precision required by investors.
In this case, one would expect to see a level of price clustering comparable to

those observed on other markets analyzed thus far. Alternatively, the minimum

price variation may be binding. In this case, the choice of tick size may reflect
other concerns of regulators.

Empirical examination supports the latter scenario. It shows very weak
clustering on the ISE, which distinguishes it from all other markets examined
thus far. Moreover, analysis of spread size and consecutive price changes
indicates that these hardly ever exceed the tick size. Taken together, the
evidence suggests that the extent of uncertainty in the market is not likely to
dictate the choice of a large relative tick. Therefore, one wonders whether there
is a rationale for the choice of a binding tick size.

This study also contributes to the order choice decision literature by
applying a trading rule introduced by Handa and Schwartz (1996) in an emerging
market, where liquidity provision is likely to be of greater importance than in the
markets in which the rule has been applied previously. Tick size is likely to affect
the balance between limit and market order submission rates and short-term
price dynamics. A large relative tick attenuates the front-running problem, which
means that it is likely to shift the balance in submission rates in favor of limit
orders. Moreover, a large relative tick also makes price discreteness more
restrictive, which is likely to affect short-term price dynamics.

The viability of an order-driven market depends on sufficient limit orders
to provide liquidity at all times. For an order-driven emerging market, such as
the ISE, temporary lack of liquidity is not uncommon. Therefore, the ISE tick rule
may reflect an extra incentive to the providers of liquidity in this market. The

performance and thus attractiveness of a limit order trading strategy require

 

frequent deviation of the market price from its equilibrium value and its correction
within a reasonably short period so that there is mean reversion in short-term
price dynamics.

The last contribution this study makes to the finance literature is that it
examines the price-volume relationship in a market where the speed and
precision of the price adjustment process are hampered by extreme price
discreteness as well as the absence of an opening call auction. The price-
volume relationship is important because it reveals equilibrium dynamics in a
market. The microstructure literature is concerned with the question of how
equilibrium is reestablished following the arrival of new information, in particular
the random rate of information arrival, the dissemination of private information to
market participants, and the learning process of uninformed investors from
informed trades. These are the respective focus of the mixture of distributions,
the sequential information, and the asymmetric information model.

After the arrival of new information, an equilibrium will be attained at a
new price, and no matter whether the information is private or public there will be
increased trading activity during the adjustment period. Therefore, a positive
relationship between price volatility and trading volume is predicted. Empirical
observation of such financial market conditions as serial dependence in trades,
lagged price adjustment, and the persistence in volatility suggest that this
relationship is dynamic. Contemporaneous relationship and Granger causality

between volume and volatility are examined in this analysis.

10

 

This dissertation is organized as follows. Chapter 2 describes the
institutional rules of the ISE and the data used in the study. Chapter 3 examines
the extent of price clustering in this market. Chapter 4 compares the profitability
of limit and market order trading strategies. Chapter 5 presents univariate and
bivariate analyses of price change volatility and volume on the ISE. Chapter 6

offers conclusions.

11

 

Chapter 2

The Istanbul Stock Exchange

2.1 History of the Securities Market in Turkey

An organized securities market in Turkey has roots in the second half of
the nineteenth century. Following the Crimean War, the first such market in the
Ottoman Empire was established in 1866, the Dersaadet Securities Exchange. It
created a medium for European investors who were seeking higher returns in the
vast Ottoman holdings. Following the proclamation of the Turkish Republic, a
law was enacted in 1929 to reorganize the fledgling capital markets under the
name Istanbul Securities and Foreign Exchange Bourse.

The bourse contributed substantially to the funding requirements of new
enterprises across the country, but the 1929 Depression and World War II had
serious consequences for the embryonic business world in Turkey. During the
industrial drive of the postwar decades, there was a continuous increase in the
number and size of joint stock companies, which began to open up their equity to
the public. Those mature shares faced a strong and growing demand from
mostly individual and some institutional investors.

By the early 19803 there was a marked improvement in Turkish capital
markets, both in regard to the legislative framework and the institutions required
to set the stage for sound capital movements. In 1981 the Capital Market Law

was enacted. One year later, the main regulatory body responsible for

12

supervision and regulation of the Turkish securities market was established, the
Capital Markets Board, based in Ankara.

A new decree in October 1983 laid the groundwork for security exchanges
in Turkey, and in October 1984, "Regulations for the Establishment and
Functions of Securities Exchanges" was published in the Official Gazette.
Operational procedures were approved in subsequent extraordinary sessions of
the General Assembly, and the Istanbul Stock Exchange was formally
inaugurated in late 1985. Turkey has one of the most liberal foreign exchange
regimes in the world, with a fully convertible currency as well as a policy that
allows foreign institutional and individual investment in securities listed on the
ISE since 1989. There are no restrictions on foreign portfolio investors trading in
Turkish securities markets. Decree No. 32, passed in August 1989, removes all
restrictions on overseas institutional and individual investment in securities listed
on the ISE. Hence, Turkish stock and bond markets are open to foreign
investors, without any constraints on the repatriation of capital and profits.

Decree No. 32 also allows Turkish citizens to buy foreign securities.

2.2 Organization of the ISE

The ISE is the only exchange in Turkey to provide trading in equities,
bonds and bills, revenue-sharing certificates, private sector bonds, foreign
securities, real estate certificates, and international securities. It is governed by
the Executive Council, composed of five members elected by the General

Assembly. The person who holds the posts of both chairman and chief

13

executive officer is appointed by the government. The four other members of the
council represent the three categories of ISE members: development banks,
commercial banks, and brokerage houses. As an autonomous, professional,
semipublic organization, the ISE is allowed a high degree of self-regulation. Its
revenues are generated from fees charged on transactions, listing procedures,
and miscellaneous services. The profits of the ISE are retained to meet
expenses or undertake investments and are not distributed to any third parties.
The ISE has its own budget. It is supervised by the Capital Markets Board (the
regulatory and supervisory authority for Turkish capital markets), which not only
ensures the proper operation of both the exchange and its members but also

protects the interests of both the public and the investing community.

2.3 Trading Mechanism for Stocks

The ISE is a fully automated continuous auction market that matches buy
and sell orders on a price and time priority basis. It was founded on December
26, 1985, and the first transaction was executed on January 3, 1986. Full
automation occurred on October 21, 1994. This is a rapidly growing market, as
revealed by various measures of total trading activity shown in Table 1 and Table
2. During the last five years, annual dollar volume tripled, share volume
increased more than twentyfold, and the number of contracts quadrupled.

Table 3 compares the ISE to other emerging markets in terms of annual
trading volume and market capitalization. One notable characteristic of the

Turkish market is its high trading activity. The ISE generates more trading

14

 

 

volume than most of the markets that have a larger market capitalization.

There are two trading sessions on the ISE: from 10:00 am. to noon and
from 2:00 pm. to 4:00 pm. Unlike some other exchanges, the trading
mechanism does not change during a trading day.

The national market has 262 firms listed (as of the end of 1998), and
there are several much smaller markets. The regional market contains smaller
stocks that cannot be listed in the national market. Young firms are listed in the
new companies market . The so called wholesale market involves trades
exceeding 10% of paid-in capital of a firm, and it is used for block trades of either
existing shares to predetermined/unidentified buyers or public offerings. Trading
in the wholesale market is conducted from 9:15 am. to 9:45 am, just before
the other markets open. Unlike some other exchanges, the ISE does not have a
parallel upstairs market that operates during the normal trading hours. Finally,
the watch list companies market is reserved for firms under special surveillance
and investigation due to extraordinary circumstances such as incomplete,
inconsistent and/or untimely disclosure of information to the public; failure to
comply with rules and regulations; or other situations leading to delisting and/or
dismissal from the related market temporarily or permanently in order to protect
investors' rights and the public interest. The number of stocks listed in those
markets is shown in Table 4. During the last five years, the quantity listed in the
national market has increased by 50%.

In the ISE all orders are submitted in the form of a limit order. The

standard trade size, one lot, contains 1,000 shares. Investors place orders with

15

brokers, who in turn enter these into the electronic limit order book. Brokers can
see the aggregate order size at each price level in real time. In addition to that
information, the member codes are also displayed for executed orders.
Settlement occurs on the second day following a transaction.

At present, there are no fixed commissions for the trading of stocks. The
fee an ISE member may charge clients ranges between 0.2% and 1% of the
transaction value. Depending on the amount and frequency of trading, the fee is
negotiable between the member and the client. For each transaction, members
have to pay an exchange fee whether the order is for a customer or a trade on
their own account. This fee amounts to 1.4 10“ % of trade value, and it is paid
separately by both members on each side of a transaction.

Short selling is allowed on the ISE. A customer must deposit cash or
security with value equal to at least 50% of the short sale. If the most recent
price is an uptick, the short sale can be made at the same price. Otherwise, it
should be at a price higher than the most recent transaction price.

Both the tick size and the maximum price change limit to be used in a
session depend on the stock price. For each stock, the weighted average price
(WAP) as a result of filling normal orders during a trading session is used to
calculate the base price and tick size for the next session. Base price is
obtained by rounding the WAP to the nearest tick. Transaction prices during a
session must be within a 20% band around the base price, again rounded to the

nearest tick. Upper limits are rounded upward and lower limits downward, so the

16

 

i10% limit may be exceeded slightly.

Two examples show the calculation of base price and tick size in a
session. First, a WAP of TL2,528 falls in the third interval (2,500 , 5,000]. Tick
size in this interval is TL50. The base price for the next session is found by
rounding the WAP up or down to the nearest permissible price, in this case
TL2,550. Like the WAP, the base price also falls in the third interval (2,500 ,
5,000]. Therefore, the tick size in the next session becomes TL50. Second, a
WAP of TL10,083 falls in the fifth interval (10,000 , 25,000]. Tick size in this
interval is TL250, so the base price for the next session is found by rounding the
WAP to the nearest permissible price, which is TL10,000. The base price falls in
the fourth interval (5,000 , 10,000]. Therefore, the tick size in the next session
becomes TL100.

There are three different order types in the ISE: normal, special , and odd-
lot. Each has its own limit order book. Normal orders have four different
subtypes, for all of which a limit price is specified. The difference concerns the
maximum trade size allowed and the status of the unfilled portion of the order.
All normal orders are subject to a “maximum trade value" upper limit of TL500
billion. One category - ordinary limit orders — is subject to a lower “maximum
number of shares” upper limit. This second limit is expressed in lots, and since it
depends on prior trading activity, it differs among stocks.

The subtypes of a normal order are as follows. (1) Ordinary limit orders:

Both the limit price and trade quantity are specified. If the order cannot be filled

17

partially or completely, it waits in the limit order book. “Maximum number of
shares” is the relevant upper limit. (2) Fill or kill: The unfilled portion is canceled
immediately, and “maximum trade value” is the relevant upper limit. (3) Limit
orders that do not restrict the transaction value: Order size is not fixed. lt
transacts with all the counter orders up to the specified price. “Maximum trade
value” is the relevant upper limit. (4) Limit orders that restrict the transaction
value: Order size is not fixed. It transacts with all the counter orders up to the
specified total trade value. “Maximum trade value” is the relevant upper limit.
Table 5 summarizes the differences among these four types of normal order.

Special orders are transactions that exceed the minimum trade size lower
limit, expressed in lots (for normal orders, this is one lot), and fall below the block
sale (10% of paid-in capital) upper limit. These orders need the approval of an
exchange official and cannot be filled partially. For an earlier order to have time
priority over a later order, both price and quantity of the two orders should be
identical, so the time priority rule is of minor importance for these orders. The
minimum trade size limit depends on the base price and is given as a multiple of
“maximum number of shares”, which is the upper limit for ordinary limit orders.
Table 6 shows the rule used In determining the minimum trade size limit.

Odd-lot orders are for quantities less than a single lot. These are

executed at the same price as the most recently traded round-lot order.

18

2.4 ISE Member Firms

All ISE members are incorporated banks and brokerage houses.
According to an arrangement by the Capital Markets Board on August 15, 1996,
banks that intend to operate in the stock market must transfer their relevant
operations to the brokerage firms they control. Table 7 shows the number of ISE
members over time. There were 140 brokerage houses executing customer
orders at the end of 1998. Tables 8 and 9 list the decomposition of total
transactions by brokerage house. In those tables, transactions are classified into
three categories: primary market, executed wholesale and special orders, and
other transactions. The last category contains executed normal orders as well
as transactions in the rights coupon and the official auction markets.

Overall, the respective shares of primary market transactions, executed
wholesale and special orders, and other transactions during 1998 were 0.07%,
2.42%, 97.51%.

Table 8 lists the top 20 brokerage houses in terms of total transaction
volume. These accounted for 53.83% of volume in 1998. Table 9 lists these in
terms of wholesale and special transaction volume. The top 5 brokerage houses

captured 73.63% of volume in this category during 1998.

2.5 Foreign Investment on the ISE
As shown in Table 10, since December 1995, foreign investment in the
ISE has more than tripled. As can be seen from Table 11, comparison of foreign

investment to total market value of the companies traded on the ISE, assuming a

19

 

float rate of 20%, suggests that about half the floating equity in this market is

owned by foreigners.

2.6 Data

The sample used in this study consists of 30 stocks that made up the
lSE30 index as of February 26, 1999. These are the most actively traded stocks
on the ISE. The sample period covers 14 months, from January 1998 through
February 1999. The data were provided by the ISE and included transaction
number, time, session, day, price, size, and an indicator variable showing the
type if two different types of the same stock were traded simultaneously.

Table 12 shows some characteristics of the sample. The median firm had
been listed for about seven and a half years as of January 1998. The median
stock price is TL 14,750 (US. dollar value of about four cents on March 12,
1999). The median firm has a market value of $467 million. The last column
shows the fraction of shares kept in the ISE Settlement and Custody Bank,
which is a proxy of the fraction of shares held by the public. The median float
rate is 20%, a low figure. There are two reasons for that. First, most of the firms
are controlled by families, as in Italy and some other countries. For example,
nine of the 30 firms (Arcelik, Koc Holding, Migros, Otosan, Turk Otomobil Fab.,
Akbank, Akcimento, Aksigorta, and Sabanci Holding) are controlled by the Koc
and Sabanci families. Their unwillingness to share control of these companies is
likely a reason for the relatively low float rates. Second, some firms (Petkim,

Petrol Ofisi, Tupras, and Turk Hava Yollari) were completely state-owned

20

enterprises. In the first step of a privatization plan, the state reduced its holdings
in these firms. However, it still has majority ownership.

During the sample period there were 302 weekdays. As Table 13 shows,
19 of these were weekday holidays. In addition, there was trade only during the
first session on January 28, 1998, and October 28, 1998, which were the
beginning of Ramadan and a national holiday, respectively. This leaves 564
trading sessions during the period. The total number of sessions available per
firm and an explanation for differences are given in Table 14. VVlth the exception
of two stocks (Dogan Yayin Holding and Efes Yatirim) that were listed during the
sample period, the minimum number of sessions per firm is 556.

The sample is representative of the entire market. Based on information
in a local newspaper on August, 28 1998, the 30 firms generated 67.9% of total
trading volume (TL27,890 billion) in the first session and 72.3% (TL28,471
billion) in the second. In relation to general market movement, Figure 1
compares the standardized price level of the sample and the ISE100 index over
the 14 months. In calculating the index, the market value of shares held by the
ISE Settlement and Custody Bank, rather than total market capitalization, is
used, and only capital gains are considered. Over the sample period the
correlation between the index and the equal weighted sample average is 0.976.
These 30 most actively traded stocks reflect most of the trading and price
change activity in the market.

The extent of foreign involvement in the 30 stocks during 1998 is revealed

in Table 15. Using monthly data, the table allocates the volume of purchases

21

 

and sales by foreign investors among the sample. On average, trading in
sample stocks constituted 81.65% of total monthly foreign volume in 1998.
Based on a comparison of first trading prices with market prices at the
close of market on March 12, 1999 (refer to Table 13), and recalling that the
consumer price index in Turkey rose from 88.5 in January 1987 to 72,4069 in
March 1999 ( average annual inflation of about 75%), it is fair to say that real
stock prices have fallen considerably. An explanation is given in Table 16, which
shows that there were stock splits in two-thirds of the sample firms, some as
large as 12 to 1. Almost all involved a bonus issue, and some included 3 rights
issue. The rights issue price was usually small in comparison to the presplit
stock price. Table 16 also shows the price adjustment on split dates, which is
important, since the adjusted presplit WAP determines the tick size used during

the first session following the split. The adjusted WAP is calculated as follows:

 

- .. WAP+ P* Sm - D .

Adjusted WAP - 1 + SBI + Sm + D , (1)

where:

WAP = weighted average price during the previous session;

P = price used in rights issue;

SW = shares issued in rights issue as a percentage of number of
existing shares;

83. - shares issued in bonus issue as a percentage of number of
existing shares; and

D = dividend to be paid to existing shares before the split.

Table 17 shows dividend payments of stocks in the sample and the
adjustment on ex dividend dates. The formula for the adjusted WAP is as

follows:

22

Adjusted WAP = WAP - D. (2)

The number of transactions per firm is given in the first part of Table 18.
There were 8,751,940 transactions in the sample, and the number per firm
ranges between 87,988 and 767,309. The second part of the table shows the
periods over which the so-called new shares were traded and the total number of
transactions involving these shares. When there is an increase in capital before
dividend is paid in a given year, newly issued shares are not entitled to receive
the next dividend payment. Until the payment of the next dividend, there is a
separate market for “old” and “new” shares. In the sample, five stocks have the
“new” shares traded.

As was shown in Table 8, combined wholesale and special order
transactions accounted for less than 2.5% of aggregate transaction volume
during 1998. The ISE data do not identify order type, so it is not possible to sort
round-lot transactions in the sample into normal and special orders.

Distinguishing a normal order (types 2, 3, and 4) from a special order is
problematic. An order value above TL500 billion clearly indicates a special
order, but a transaction below that amount yet above the minimum trade size
limit for special orders can be a normal order or special order. Two groups of
transactions can be identified: those that exceed TL500 billion, and those
transactions that exceed the minimum trade size for special orders (which
contains the first group). All transactions in the first group are special orders, but

cases in the second group may be either normal or special.

23

To find the minimum trade size for special orders, the maximum number
of shares limit must be calculated. By the fifth day of each month, the ISE
announces that information. This figure depends on the average trade size per
transaction during the previous month. In calculating this quantity, the ISE uses
all transactions except odd-lot orders. Once this quantity is found, the rule
shown in Table 6 can be used to determine the minimum trade size for special
orders.

The above procedure is followed to determine the number of transactions
in the sample that are likely to be special orders. For all months, the average
trade size is calculated for each stock after eliminating the odd-lot orders. This
way, the relevant maximum number of shares limit is obtained for all months
except the very first one. From Table 19 it can be seen that maximum number of
shares limit is quite stable over time unless there is a rights or bonus issue.
Therefore, it is assumed that the figure found for the second month is also
correct for the first month in the sample period.

The number of identified special orders and gray cases per stock is shown
in Table 21. Only 23 special orders can be identified, and there were 6,357 gray
transactions. Even if all the latter are treated as special orders, the fraction of

executed special orders in the sample is less than 0.1%.

24

 

Chapter 3

Price Clustering on the Istanbul Stock Exchange

3.1 Introduction

In this chapter, the ISE tick is evaluated by empirically examining the
usage of different discrete price sets by traders.

According to Grossman et al. (1997 p. 26),

The optimal minimum tick will be smaller than the unit of trade typically
used during periods of normal trading activity. If market participants
typically used the minimum tick as the unit of trade, then the market would
lack the flexibility to reduce the minimum tick to allow for smaller units
when appropriate. The minimum allowable increment is not chosen to be
the typical degree of precision required but, rather, to reflect the most
precision required, that is, the relatively rare event.

The minimum tick on the ISE is a step function of stock price. Table 22
compares it with step function rules used in six other equity markets: Helsinki,
Hong Kong, Paris, Singapore, Tokyo, and Toronto stock exchanges. The
relative tick on the ISE is at least 120% and as much as 2,200% larger than
those on other exchanges.“

One reason for the ISE’s apparently large relative tick size may be that
price resolution in this market is low. The generation of information in the

Turkish market is not the same as in developed markets. If uncertainty about the

true value of stocks is higher due to inadequate information, then investors may

 

‘Average relative tick is 1.36% in Istanbul, 0.61% in Hong Kong, and 0.06% in
Fans.

25

avoid using a smaller tick size even if it is allowed. In this case, trader behavior
would depend on the hypothesized tradeoff between price resolution and
negotiation costs. Therefore, one would expect to observe a level of clustering
comparable to that in other markets.

An extremely low level of clustering would be consistent with the view that
the choice of a large tick reflects other concerns of the exchange. In this case,
trader behavior would not depend on the above-mentioned tradeoff, simply
because the enforcement of a large tick does not give traders the flexibility of
choosing the desired tick. A consequence of this may be the slow evolution of
prices in the market. If traders can resolve share values with a higher precision
than is allowed by the tick size, then prices will be sticky. Until the true price
moves significantly close to the next available discrete price, the transaction
price will not change.

In markets where the minimum price variation does not limit the use of the
desired tick, tick size is not a determinant of spread. Other than the concern that
clustering may indicate anticompetitive behavior in dealer markets, the positive
relationship between clustering and spread size found in Gwilym et al. (1998b)
can be attributed to the overlap in their determinants. Yet, regarding the assets
for which minimum price variation is binding, there is evidence about a positive
relationship between tick size and spread. Low-priced stocks on the NYSE by
definition have a large relative tick, and the spread equals tick size for a high
fraction of these stocks. This accords with the Grossman and Miller (1988)

argument that tick size puts a floor onspreads. If the ISE tick size is binding,

26

then spreads may decrease after a reduction in tick size. Moreover, it is a
common belief that lower spread is associated with higher volume, but empirical
research provides limited support for this belief. Hameed and Terry (1998)
report that a decrease in absolute tick size increases volume only for actively
traded stocks. Therefore, it is arguable that a decrease in spread caused by a
reduction in tick size necessarily increases trading volume in a relatively thin
market such as the ISE. If the concern of the ISE is the liquidity provision, then
market-making incentives may have high priority. Although there are no official
market makers, it may be critical to keep profits of voluntary liquidity providers
sufficiently high. The ISE is still in its infancy and market depth may be important
for the viability of the exchange.

The experience of the Stock Exchange of Hong Kong (SEHK) may show
the delicacy of this issue. In early 1994, the SEHK proposed to reduce the tick
size by half, which the brokerage industry immediately opposed. The SEHK
compromised and agreed to a four-month evaluation program; beginning on
June 1, 1994, tick size was reduced by 50%. Volume dropped in the evaluation
' period, and the exchange reverted to the original tick size for stocks trading
below HK$10. The decline in volume may partially have been caused by the
gaming behavior of the brokerage industry.

The purpose of this chapter is to determine whether the ISE’s relative tick
size is so large that it deprives traders of flexibility. The next section reviews the
literature on clustering. Then, additional details are presented about the lSE’s

tick rule. The following section contains an empirical analysis, and ends with a

27

brief conclusion.

3.2 Literature Review
3.2.1 Early Evidence of Clustering

Osborne (1962) and Niederhoffer(1965, 1966) present empirical evidence
of clustering on the NYSE and show that it depends on price level and variability.
These studies do not give any economic justification for the anomaly, but rely
instead on behavioral explanations.

Niederhoffer (1965) relates clustering to the behavior of limit order
submitters. According to this explanation, traders place limit orders in numbers
with which they are accustomed, that is, even rather than odd fractions. The
argument is supported by using data from a specialist book. Given the
congestion of limit orders at even fractions, it is hypothesized that one would
expect a higher degree of clustering in high-priced issues, since a specialist
would be reluctant to maintain price continuity by trading for his own account,
which would result in a large dollar change in his inventory. The data confirmed
this hypothesis and the examination of stocks with a closing price above $50
showed heavy clustering.

Another variable related to clustering was explored in Niederhoffer (1966).
The hypothesis was that occurrences of consecutive fractional prices in a given
stock on a single day are not independent. Thus, consecutive price changes of
118 would result in equal proportions of even and odd eighths. Niederhoffer used

intraday transaction data for randomly selected stocks on seven days, to classify

28

 

 

transaction prices into three groups according to the amount of consecutive price
change: no change, a change of 1/8, and a change larger than 1/8. Given the
way the groups were constructed, the first two were more likely than the third to
contain stocks with lower price variability. The results showed moderate and
high levels of clustering in the first and third groups, respectively, but no
significant clustering in the second. The findings suggest that as price variability
increases, so does clustering. The lack of clustering in the second group is
consistent with the hypothesized negative effect of consecutive price changes on

clustering.

3.2.2 An Overview of Clustering Studies

All of the previous studies display the extent of clustering and most of
them examine its determinants. Two studies investigate the effect of a change in
tick size on clustering More recent papers analyze whether clustering is related
to other microstructure issues. Those studies give evidence from open outcry,
continuous/call auction, and dealer markets. Depending on the market, analysts
focus on single or multiple assets.

Single-asset studies look at variation in price over time. Researchers
have examined gold, foreign exchange, long-term government bonds, and
financial derivatives. Multiple-asset studies look at cross-sectional variation in
equity markets. Table 23 shows the market(s) analyzed in each study.

All this research provides evidence about the nature of clustering. Harris

(1991) shows its persistence through time. Transaction price distributions for the

29

four most actively traded securities on the New York Stock & Exchange Board
between March 22 and April 15, 1854, are qualitatively identical to the more
recent Center for Research in Security Prices (CRSP) sample of daily closing
stock prices for the period January 1963 to December 1987. Both Harris (1991)
and Gwilym et al. (1998a) show clustering in quotes and transaction prices as
well as in intradaily and closing prices for the same assets. Neither study reports
any significant or consistent difference in clustering patterns. Harris (1991) and
Booth at al. (2000) compare the clustering of same or similar assets across
different market structures. Using NYSE/AMEX and Nasdaq transaction and
quote data, Harris shows that Nasdaq has a higher degree of clustering in both
transaction and quoted prices for similar stocks. Booth et al. examined the same
stocks in terms of trading during continuous session and after hours on the
Helsinki Stock Exchange (HSE). Transactions can be executed in either the
downstairs or upstairs market during the continuous trading session and in the
upstairs market after hours. It was found that clustering during these two periods

is very similar.

3.2.3 The Extent of Clustering in Different Markets

The extent of clustering varies across markets. Extreme cases are
financial derivatives on the London lntemational Financial Futures and Options
Exchange (LIFF E), silver futures on the Commodity Exchange (COMEX ), and
common shares on the Helsinki Stock Exchange. In the London case, about

98% of all trades occur on even ticks for the FTSE100 index futures, FTSE100

30

 

index options, and FTSE250 index futures. For silver futures, 92.2% of trade
prices fall on zero and five out of ten available final digits. On the HSE, the
share of transaction prices that fall on the same two digits varies between 55%
and 78% in the three tick regimes analyzed. As suggested by Grossman et al.
(1997), one possible measure of the extent of clustering is standardized range,
that is, the range between the highest and lowest frequencies divided by the
expected frequency per unit. The standardized range is an ordinal measure that
can be used to rank the degree of clustering on different markets, but it cannot
be used to assess the relative amount of clustering on those markets. The last
column of Table 23 shows the standardized ranges reported in various studies or

calculated for this study by using reported frequency distributions.

3.2.4 Determinants of Clustering

The attraction of round numbers explanation suggested by Goodhart and
Curcio (1991) argues that discrete trading prices are obtained from continuously
distributed true values by rounding to the nearest available final unit, but the
attraction of each integer varies. Most studies do not provide evidence in favor
of that hypothesis, which is supported only in Aitken et al. (1996); evidence in
Harris (1991), Brown et al. (1991), and Goodhart and Curcio is inconsistent with
the hypothesis.

Ballet al. (1985) put fonrvard the price resolution explanation, which states
that clustering manifests haziness about asset values. Building on that

hypothesis by adding a counteracting force, Harris (1991) argues clustering is

31

due to the incentive to lower negotiation costs, and its level is limited by the
price resolution in the market. High price resolution will result in a small
dispersion among traders’ reservation prices, and if a large tick does not include
a price acceptable to both parties in a trade, then gains from trade may be lost.
Most empirical research adopts the price resolution hypothesis, perhaps
because it is difficult to find measures of the level of negotiation cost in a market.
In fact, even though Harris (1991) proposed the negotiation hypothesis, he does
not rely upon it in developing explanatory variables.

The price resolution hypothesis suggests a number of variables that proxy
the stock of information in the market. Ball et al. (1985) argue that high volatility
is an indication that existing infomration is obsolete. Ball et al., Harris (1991),
Gwilym et al. (1998a), Brown et al. (1991), and Booth et al. (2000) all report a
positive relationship between volatility and clustering. The only exception is
Hameed and Terry (1998) who find an insignificant relationship for the Singapore
Stock Exchange. Aitken et al. (1996) differentiate between market and individual
stock volatility, but both kinds display a significant positive association with
clustering.

Harris (1991) justifies firm size as another proxy for the stock of
information in the market on the premise that larger firms release more
information and are also followed by a larger number of analysts. He finds
evidence of a negative relationship between firm size and clustering, but Aitken
et al.(1996), and Christie and Schulz (1994) could not detect any significant

relationship.

32

Proxies to incorporate the ease of valuation of an asset are used by Harris
(1991), Aitken et al. (1996), and Christie and Schulz (1994). Harris uses a
dummy variable for close-end funds, and Aitken et al. use dummy variables for
optioned stocks, and for those that can be sold short, since these two factors are
believed to increase the efficiency of stock prices. Both studies find a negative
relationship between these variables and clustering. In an analysis of Nasdaq,
Christie and Schulz use dummies for dual listed stocks and stocks with listed
options, but both of these prove to be insignificant.

Since frequent trading will lead to the incorporation of the most recent
information into prices, variables that proxy trading activity, namely, volume and
number of trades, are expected to increase price resolution. Aitken et al. (1996),
Booth et al. (2000), and Harris (1991) all find a significantly negative relationship
between transaction frequency and clustering, but Gwilym et al. (1998a) report a
significant relationship in the opposite direction. This may be due to the open
outcry trading mechanism used by LlFFE, which requires traders to stay in the
pit for an order to execute. As suggested by Brown et al. (1991), pit traders have
incentives to speed up trading by using a coarser price set to participate in a
larger number of trades. Hameed and Terry (1998) find a negative relation
between trading volume and clustering.

The only study that tests the negotiation hypothesis extensively is by
Brown et al. (1991), who consider the implications of negotiation costs from the
aspect of pit traders in COMEX silver futures. They test trade size as a variable,

because a large trade increases both total surplus and benefits from negotiation.

33

This variable was suggested by Harris (1991) but not tested. Gwilym et al.
(1998a), Booth et al. (2000), and Aitken et al. (1996) also include it in their
analyses. The first two find, along with Brown et al., a negative relationship
between trade size and clustering, but Aitken et al. report a relationship in the
opposite direction for the Australian Stock Exchange. Based on the price
resolution hypothesis, Aitken et al. argue that trade size may proxy the existence
of new information, since larger orders are sometimes associated with informed
traders. Two factors may weaken this effect, however. First, informed traders
usually split their orders to hide themselves (stealth trading), so the association
may be questionable. Second, even if a positive connection between large
trades and informed trading is assumed, this may give rise to a negative
relationship between clustering and trade size, since informed trading is
expected to improve price discovery.

Brown et al. (1991) point out additional reasons for volume and volatility
as determinants of clustering: lost trade opportunity and higher inventory
variance from the aspect of pit traders. Because silver futures trading uses an
open outcry mechanism, which requires face-to-face negotiation in the pit, it is a
suitable setting for testing various implications of the negotiation hypothesis.
Furthermore, detailed data about parties in trade and timing information are
available. Brown et al. report that, in addition to such variables as volatility and
trade size, whether a trader executes an order for a customer or on his own
account makes a difference. Moreover, they show that odd tick orders take

longer to consummate and create delay in clearing a trade.

34

The discussion in Ball et al. (1985) and the application in Harris (1991)
both assume that, for a given price resolution, traders use discrete price sets
based on minimum price variation that is a constant fraction of price. The
relation between desired and mandated tick, both of which are expressed
relative to price level, captures the effect of mandated tick size on clustering by
using price level as a proxy. This implies more clustering for high-priced stocks
in a market. The presumed relationship could be discontinuous in markets that
use a step function tick rule. Aitken et al. (1996), Ball et al. (1985), Booth et al.
(2000), Christie and Schulz (1994), Hameed and Terry (1998), and Harris (1991)

all confirm the hypothesized positive relationship of clustering to price level.

3.2.5 The Effect of a Change in Tick Size

As discussed above, almost all studies use relative tick size to control for
the effect of tick size on clustering. For example, Aitken et al. (1996) and Booth
et al. (2000) examine stock exchanges that use step function tick rules, and they
either consider a single regime or analyze different regimes separately; neither
compares different tick regimes. Two studies estimate the use of discrete price
sets if a lower tick is used. Harris (1991) bases his prediction on usage
frequencies at the prevalent tick size. Hameed and Terry (1998) provide some
evidence that absolute tick size may be a factor and that the method of
prediction used by Harris (1991) overstates the use a of smaller tick. They
examine the effect of a change in tick size by comparing the level of clustering in

two adjacent regimes after controlling for cross-sectional determinants of

35

clustering: price level, trading volume, and volatility. The results show that price

clustering increases when the absolute tick size is reduced.

3.2.6 Relation to Other Microstructure Aspects

More recent studies examine the relationship of clustering to other
microstructure aspects such as operating efficiency, transaction costs, collusion
in the market, and the practice of preferencing.

Brown et al. (1991) show that low degree of rounding harms the
operating efficiency of the market. Odd tick orders confuse pit traders, make
them leave the pit temporarily, and take longer to execute. Emphasizing the
tradeoff between anticompetitiveness and market viability, Brown et al. (1991 p.
68) argue,

Intuition suggests that the optimal minimum tick size is a very small
number, because a small tick maximizes the flexibility of traders in
establishing transaction prices. However, there is no guarantee with a
small tick size that the competitive solution is viable given the cost
structure of those providing services to the market.

Gwilym et al. (1998b) analyze the relationship between clustering and
transaction costs. They examined the intradaily pattern of the bid/ask spread
and clustering in quotes. Spread is significantly higher when the market opens
and significantly lower when it closes. Even after controlling for the effect of this
intraday behavior, they found a significant positive relation between quoted

spread and the level of clustering in quotes. Since volume is high both around

opening and closing, they conclude that the use of odd ticks is driven more by

36

the desired quoted spread than by volume of trading.

Hameed and Terry (1998) report limited evidence from Singapore about
the effect of tick size on trading volume. They used data on four stocks for which
tick size decreased from S$0.5 to S$0.1 on July 18, 1994. For actively traded
stocks, they found that trading volume rose and volatility declined after a
decrease in tick size, but no significant relationship was detected for thinly traded
stocks.

Three studies explore the issue of whether collusion is a factor in
clustering. Christie and Schulz (1994) examined the distribution of pooled inside
bid and ask quotes and found that the lack of one-eighth spreads can be traced
to an absence of either inside bid or ask quotes ending in odd eighths. Those
stocks whose market makers rarely use odd eighth quotes have a mean duration
of odd eighth quotes lower than that for even eighth quotes; for the remaining
stocks, the mean duration of both is quite similar. Cross-sectional logistic
regressions show that the factors which affect the width of spread have little
power in identifying firms whose market makers avoid or use odd eighths. The
authors argue that one explanation of this evidence is implicit agreement among
market makers to keep spreads of at least at $0.25 by not posting quotes on odd
eighths.

Bessembinder (1997) tests the implicit collusion hypothesis by examining
the relationship between price/quote rounding frequencies and measures of both
investors’ trade execution costs and market maker profits. Cross-sectionally,

after controlling for determinants of spread size (price level, firm size, and

37

volatility), Bessembinder found that both quoted and effective spreads on the
NYSE and Nasdaq are significantly and positively related to price and quote
rounding frequencies. The decomposition of effective spread into price impact
and realized spread shows, for NYSE stocks, the observed relation between
price rounding and trade execution costs can be explained by similar variation in
information costs. For Nasdaq, the significant and positive relationship between
realized spread and rounding frequencies cannot be due to the information
content of trades. This implies that larger trade execution costs associated with
rounded prices can be justified by variation in observable market making costs
for NYSE but not for Nasdaq stocks, which supports the collusion hypothesis.
According to Bessembinder, a partial explanation for the positive relation
between quote rounding and trading costs comes from the examination of price
improvement probabilities. A comparison of these probabilities for spreads
posted using odd and even eighths showed that trades are less likely to receive
price improvement when even eighths are used on both NYSE and Nasdaq, but
the differential for Nasdaq was more dramatic.

Grossman et al. (1997) indirectly examined the level of transaction cost
for NYSE/AMEX and Nasdaq by using two different methods. The direct way to
compare transaction cost would require the information on market depth, rather
than spread size, but the former is difficult to measure. Since higher transaction
costs will be passed to firms when they issue equity, comparable firms would find
it advantageous to switch to the NYSE if it has lower transaction costs. A

comparison revealed, however, that there are more secondary issues of Nasdaq

38

stocks than NYSE/AMEX stocks. Also, transaction cost has implications for
trading volume. One would expect the trading volume of comparable stocks to
be higher in markets with lower transaction cost. Examination of comparable
stocks showed that those on Nasdaq have higher trading volume than those on
the NYSE, which is inconsistent with Nasdaq’s higher transaction cost.

Godek (1996) hypothesizes an indirect link between preferencing and
clustering through spread size. He argues that the focus on implicit collusion in
Nasdaq understates the importance of preference trading as an institutional
determinant of quoted spreads. A competing market maker who offers an
unusually good price can be expected to receive less order flow from brokers
and other market makers. Narrowing the spread below a certain level not only
may fail to attract trades, but also could actually tend to repel them. Godek
shows that even eighth spreads are quoted by using even eighth prices on both
NYSE and Nasdaq. It is not the different quoting behavior but the higher
spreads driven by the practice of preferencing that causes the low frequency of
odd tick quotes for Nasdaq.

Godek uses the same exogenous economic factors as Christie and
Schultz (1994) to identify firms whose market makers avoid or use odd eighths,
but he reaches the opposite conclusion. Cross-sectional variation in clustering
can be explained by factors that determine the spread size. Godek also gives
evidence that in this large sample the bimodal distribution of clustering remained

stable after the Christie and Schultz study data period.

39

Booth et al. (2000) provide evidence on the extent of internalization and
its relation to price clustering by using trades of 73 issues listed on the Helsinki
exchange during 1993-1995. lnternalization, the extreme case of preferencing,
is the practice of brokers executing orders in-house. Preferencing is believed to
be an important institutional factor affecting price formation. The study
compared the degree of clustering between internalized and nonintemalized
trades. After controlling for other factors, Booth at al. found that internalization is
related positively to the level of clustering in continuous trading session, but its
effect is small in comparison to that of control variables. In after-hours trading,

internalization was found to be insignificant in explaining the level of clustering.

3.3 Empirical Analysis

The ISE tick rule was compared in Table 22 to the step function rules
used on the Helsinki, Hong Kong, Paris, Singapore, Tokyo, and Toronto stock
exchanges. One feature of the ISE rule is that the average relative tick is much
larger than for the other six exchanges (1.36% for the ISE, versus 0.06% Paris,
0.18% Helsinki, 0.19% Tokyo, 0.40% Toronto, 0.55% Hong Kong, and 0.61%
Singapore).5 Moreover, the step function used by the ISE differs from the others.
It has a larger number of tick regimes, and the width of each is narrow. On

average, an increase by 114% or less would move a stock price into the next

 

5 The first and last regimes are excluded in the calculation of average relative tick
sizes.

40

regime.“3 Regime widths in Helsinki, Paris, Tokyo, and Toronto are several times
as large as those in the lSE and this is true for at least some regimes in
Singapore and Hong Kong.

The ISE tick rule limits both the time a stock may spend in a regime and
the percentage decrease in relative tick size as stock price rises. As shown in
Figure 2, relative tick size starts at 2% or 2.5% for each regime (left end) and
falls to 1% (right end). This type of tick rule is exactly the kind that Angel (1997,
p. 678) recommends not be employed by exchanges,

If the exchange adopts a step function, however, it may find that it has set
relative tick sizes either too high or too low, without giving firms a means
to adjust them. For this reason, it seems prudent for an exchange to set a
small number of absolute tick sizes and give firms the flexibility to modify
their own relative tick sizes through stock splits.

The stability of nominal stock prices over time (refer to Table 12) and the
frequency of stock splits and dividends during the sample period (refer to tables
16 and 17) raise an interesting question. Persistently high inflation has
decreased real stock prices on the ISE over time. Moreover, the nature of the
tick rule makes changing stock price through splits a less effective mechanism
' for firms to determine relative tick size. That being the case, it is not clear why
splits have been a common practice of firms listed on the ISE at the expense of

lowering real stock prices further. This is inconsistent with Angel’s (1997) view

that the motivation for stock splits is an adjustment of relative tick size through a

 

° Ignoring the first and last tick regimes, the maximum percentage change
required is 96% (144%) in five (three) of the remaining eight regimes.

41

change in share price.

For each stock the distribution of tick size over the sample period is
presented in Table 25.7 Apparently due to the narrow tick regimes, persistently
high inflation rate, and frequent use of stock splits, ISE stocks used multiple tick
sizes during the sample period. The cross-sectional mean, median, and mode of
the number of tick regimes used are 3.0, 3.0, and 2.0, respectively. The
corresponding numbers for 18 firms that had stock splits are 3.6, 4.0, and 4.0.8

As was shown in Table 22, the ISE has ten tick regimes. For convenience
in presentation, these are collapsed into three groups: (1) regimes 100, 1,000,
and 10,000, (2) regimes 25, 250, and 2,500, (3) regimes 50, 500, and 5,000.9
Based on the final digits of transaction prices, there are ten, four, and twenty
price categories in these respective groups.

Table 26 reports actual frequencies of the final digits of intraday
transaction prices for all stocks in the sample. Part A shows the frequencies for
the first group. The null hypothesis that prices are uniformly distributed across
the ten price categories in this group is rejected, but neither the individual tick

regime nor combined distributions are consistent with the pattern of clustering

 

7 As explained in Chapter 2, the tick size used during a trading session depends
on the WAP in the previous session. Since special orders are not included in the
calculation of WAP and it is not possible to identify these orders in the sample,
WAP data were hand collected from daily Turkish newspapers.

° One of the firms, Migros, had a minor split during the sample period and is not
included in this calculation.

9 The first tick regime contains only 1,075 observations. Therefore it is excluded
from the analysis.

42

that has been found elsewhere. The 00 and 50 price categories are not
necessarily more common than the other categories. Moreover, the frequencies
of even and odd final digits are almost identical.

Part B of Table 26 displays the frequencies for the second group. Both
individual tick regime and combined results show weak clustering consistent with
the pattern found elsewhere. First, the 00 price category is more common than
the 50 category, which is more common than the other two price categories in
this group. Second, even numbers are more frequent than odd numbers, which
agrees with the observations in other studies.10 Nonetheless, the standardized
range measure calculated by using the combined frequencies is 0.18, implying a
lower level of clustering than other markets as shown in Table 24.11

Part C of Table 26 displays the frequencies for the third group. Prices are
not uniformly distributed, and the frequency of even final digits is larger than that
of odd final digits. Moreover, the examination of combined frequencies shows
that, for all the adjacent price categories, even categories are more frequent than
odd categories. This last observation implies the existence of a weak form of
clustering. To interpret the pattern, one can refer to the reasoning used in Ballet

al. (1985) or in Harris (1991), both of which involve multiple classes of traders

 

‘° Since the tick regimes in the second group contain four price categories, this
result follows directly from the first result.

" The standardized range is an ordinal measure that can be used to rank the
degree of clustering in different markets, but cannot be used to assess the
relative amount of clustering within those markets.

43

who use different types of rounding. The extent of clustering reflects the relative
proportions of these trader classes in the market. Consider the TL50 regime.
Traders can round the price using any one of the six available pricing grids: 50,
100, 200, 250, 500, and 1,000. For any pricing grid, it is assumed that the
possible final three digits of the rounded price are equally likely to occur. For
example, for the first grid in which prices are rounded to the closest TL 50, the
occurrences of all the 20 possible final three digits (000, 050, 100, 150,...,800,
850, 900, 950) are equally likely. In this setting, the prediction that the final three
digits are more likely to be 000 than 500 relies critically on the existence of
traders who use TL1,000 rounding. By following this line of reasoning, the
observed pattern of clustering can be explained as follows. If, due to a large
relative tick, there are only two classes of investors on the ISE using either one-
tick or two-tick rounding, then for all the adjacent price categories, even
categories will be more frequent than odd categories. Moreover, if the proportion
of traders who use two-tick rounding is much less than 50%, then the extent of
clustering will be low. The pattern of clustering observed in the second and third
groups (parts B and C of Table 26) are consistent with this scenario.

The observed prices depend on the initial price level, and for a stock with
infrequent price change it may be the case that only part of the entire tick regime
will be swept over time. To take this serial dependence into consideration, the
adjustment discussed in Harris (1991) is used. In this method each price change
creates domain events. A domain event occurs when prices change and price

path passes over or arrives in a different price category. If prices do not often

44

visit the region near a given category, then the frequency for that category is
adjusted upward. If prices dwell in that region, then the frequency is adjusted
downward. One property of this estimator is that the adjusted frequencies add
up to one. Another property that makes this estimator desirable in the analysis
of the ISE is that prices can cluster even if all price changes are equal to tick
size.” This is because zero price changes are not counted as domain events.

For each price category the adjusted frequency is given by the following

expression:

fadjusted = factual T funiform "' fdomain ; (3)
where:
funm the expected frequency given uniform distribution; and

fdmm the frequency of domain events.

Three cases are discussed to understand the effect of the Harris (1991)
adjustment: (1) If price does not pass over or arrive a price category, then for
this category: factual = foam," :> fadjusued = funnorm. (2) If some price categories are
skipped when price changes, then for those skipped categories: fauna, < fdmm :>
fadjum < fumfom. (3) The effect of zero price change observations on
corresponding price categories: fem. > fdomain :> fadjusmd > fumm.

The adjusted frequencies are shown in Table 27. Qualitatively the
conclusions from this table are identical to those from Table 26. The only

notable difference is, with the exception of the TL 5,000 tick regime, the pattern

 

‘2 See Harris (1991, p. 401) for further properties of this estimator.

45

of clustering in the third group becomes more consistent with the above
explanation. The standardized range measure, calculated by using the
combined frequencies, is 0.26 for the third group.

To interpret the evidence provided by adjusted frequencies, univariate
statistics are presented about bid-ask spread size and transaction price
dynamics in the sample.

Bid and ask quotes at the close of the second trading session were
compiled from Dunya, a daily Turkish financial newspaper. The data were
screened for instances of no reported spread, missing bid or ask quotes, and
days in which systematic errors were found in the newspaper. The resulting
distribution of spread size in ticks is presented in Table 28 and Figure 3. Overall,
in 7,486 of the 7,643 observations (98%) the size of bid-ask spread is at the
minimum value of one tick. Hence, unlike other exchanges, the ISE can be
considered a single-tick-spread market, which is consistent with the explanation
that the tick size constraint on bid—ask spread is binding. Similar findings has
been reported for low-priced stocks on the NYSE and AMEX. Harris (1994), for
example, found a bid-ask spread equal to the tick size in 66.8% of the
observations for stocks traded below $10, compared to 34% for stocks priced
above $20.

The change in price from one transaction to the next, expressed as
multiples of tick size, is presented in Table 29. In about 92% of the

observations, there is no change. When price change occurred, it was equal to

46

the tick size in more than 99% of the observations.13 The evidence is consistent
with the explanation that, due to large relative tick sizes imposed by the ISE,
small changes in equilibrium price cannot be reflected in transaction price.
Furthermore, since the price change in consecutive transactions rarely exceeds
the minimum possible amount, the argument that the typical price change in the
market justifies the enforced tick size is not supported. Taken together, the
evidence about spread size and price change probabilities suggests that the tick
rule may be constraining trader behavior to a greater extent on the ISE than in
other markets.

As Table 29 shows, price changes predominantly by one tick on the ISE.
Therefore, skipping a final digit category is not an issue in this market. This
means factual = fdmin and therefore f,,,,,,,,,,,, = funiform ignoring the zero price change
cases. Hence, if there is any clustering, it will be driven by zero price change
cases. Based on this interpretation, adjusted frequencies show that the time

spent on each price category does not display significant variation on the ISE.

As discussed in the literature survey, clustering has been shown to
increase with a decrease in relative tick size. Figure 2 reveals that the relative
tick jumps at least 100% when a small increase in stock price moves the price
into the next available tick regime. The weak clustering demonstrated in

previous tables was attributed to a large relative tick size. On the lSE, relative

 

‘3 Table 29 also shows that unconditional up and down movement probabilities
are nearly identical during the sample period.

47

tick size never falls to a level comparable to those in other markets. In spite of
that, one would expect to observe relatively more clustering at the right end of a
tick regime, where relative tick size reaches its minimum of 1%. This can be
explored by comparing the extent of clustering just before and after a stock
moves into another tick regime. The advantage of this method is that
observations will be taken during adjacent periods. Therefore, it is unlikely that a
change in determinants of clustering will occur during the sampling period.

Alternatively, the possibility of a variation in clustering can be examined
within a tick regime. If there is some clustering on the ISE, its extent should be
highest (lowest) at high (low) price levels. This method enhances the
comparison, since all observations are taken within the same tick regime. Table
30 reports the results by using this method.“ Three price ranges within each
regime are defined: low, medium, and high. The definition of each range is
shown in the first part of the table.15

Although the results from these three groups do not display a consistent
pattern, none of them supports the hypothesis that within regime variation in
relative tick leads to a positive relationship between relative tick and the extent of
clustering. For example, the extent of clustering is a U-shaped function of

relative tick in the second and third groups as shown in Part C and Part D of

 

“ Only combined frequencies are reported to increase the number of
observations in each price range.

‘5 The second range in some cases (tick=25, 250, and 2,500) does not contain
the entire set because of narrow regime width.

48

Table 30. Therefore, it is fair to conclude that even a relative tick size of 1% may
be too large to result in usage frequencies consistent with the pattern of

clustering that has been found elsewhere.

3.4 Summary

It was shown that ISE’s relative tick is up to 900% greater than that of
other exchanges. The major purpose of this study was to examine whether large
tick size restricts trader behavior in this market. It was found that prices show
very weak clustering. It appears that traders use predominantly one— or two-tick
rounding, and the proportion of the latter is small.

The examination of spread and consequent price change frequencies
revealed that these hardly ever exceed tick size, which indicates that the tick size
is binding. This suggests that the large relative tick in this market cannot be
attributed to low price resolution. Given that price changes predominantly by
one-tick, the only way prices will cluster in this market is if price is more likely to
change when it falls on certain categories than others. The results show that this
is not the case.

Based on the evidence from other markets, it was hypothesized that price
clustering within a tick regime increases with an increase in stock price. Possibly
due to the narrow regime width, it was found that within-regime variation in

clustering is not consistent with this hypothesis.

49

 

Taken together, empirical findings in this analysis suggest that the large
relative tick size on the ISE restricts trader behavior. Since a large relative tick
applies to all price levels with little variation, it appears that tick size is used as a

policy variable by the exchange.

50

Chapter 4

Limit Order Profitability on the Istanbul Stock Exchange

4.1 Introduction

The ISE is an order-driven market, and liquidity provision differs in this
market structure from the other two types of continuous trading systems. In
quote-driven and specialist systems, it is the responsibility of dealers or
specialists to provide liquidity at all times, but in an order-driven market no party
has such an obligation. Public traders provide liquidity by their use of limit
orders.16 From the perspective of individual traders, a limit order can be
preferred when there is a need to rebalance their portfolios. Alternatively, they
may find this a profitable strategy by itself and act voluntarily as market makers.

The extent of liquidity provision within continuous markets depends on a
number of factors: the anonymity of trading, the informational advantage of
liquidity providers, their ability to avoid trading with possibly informed traders,
and the number of liquidity providers.

In a quote-driven market, trading is not necessarily anonymous.
Preferencing and quote-matching practices create a certain kind of competition.
Brokers direct uninformed customer orders to dealers with whom they have a

close relationship. In effect, dealers do not routinely accept all incoming orders

 

‘6 Brokerage houses also can submit limit orders when they trade on their own
account.

51

but choose on a case-by-case basis. This is done by posting unattractive quotes
initially and matching the best bid or offer when dealer wants to take the other
side of a transaction. In an order-driven market, trading is anonymous, and
priority rules make the completion of a trade automatic once a counterparty
arrives. Therefore, liquidity providers cannot make case-by-case decisions.

Specialist market is a hybrid mechanism. The specialist as broker
executes limit orders left with him by other brokers; as dealer, buys and sells for
his own account. Because he observes the total order flow, he can use this
advantage to decide when to step in ahead of the limit orders left with him for
execution. Nevertheless, the specialist must optimize average profits, just as do
liquidity providers in an order-driven market, due to the anonymity of arriving
market orders.

The number of liquidity providers changes relatively infrequently in a
quote-driven market, but may change substantially over time in an order driven
system because there are no direct costs of entry. In the former there is a
commitment to provide liquidity, but it may not necessarily be obtained at the
minimum possible cost due to the nature of competition in the system. In the
latter, although there is no such commitment, when the number of potential
liquidity providers is large enough, the cost of liquidity may be driven down to
the perfect competition level.

It is clear that market makers in specialist and quote-driven systems are
given certain benefits that protect them from the problem of adverse selection in

return for their commitment to provide liquidity. Although the viability of an order-

52

driven market requires the use of limit orders, the lack of explicit protection from
adverse selection is noteworthy.

A line of research that examines the order choice decision has emerged
in recent years. Harris and Hasbrouck (1996) studied usage frequencies,
execution rates, and profitability of limit and market orders on the NYSE. They
found that trader behavior depends on the order size, the prevailing spread, and
the side of the market, and they examined the ex post optimality of trader
behavior. Handa and Schwartz (1996), who analyzed the rationale of limit order
trading, argue that the viability of an order—driven market requires short-term
price dynamics that follow a mean reverting process. They predict and test
whether limit order trading is profitable only for traders with well-balanced
portfolios due to their low opportunity cost of nonexecution. Hamon et al. (1995),
drawing on findings in Handa and Schwartz (1996) and Hamon et al. (1993)
compared the profitability of limit order trading on the NYSE and the Paris
Bourse (the CAC system).

The balance between limit and market order submission rates is important
for the ISE, because in an emerging market there is concern about the extent of
liquidity provision. Therefore, the method used by Handa and Schwartz (1996)
will be employed in this chapter to examine the profitability of limit and market
order trading strategies.

The chapter is organized as follows. The next section compares the ISE
to the Paris Bourse and the NYSE regarding certain aspects of their

microstructure. The following section discusses the work of Handa and

53

Schwartz (1996), in particular the implementation and limitations of their one
period model. It also describes the evidence given in Handa and Schwartz
(1996) and Harris and Hasbrouck (1996) and compares the methods used in
these two studies. Then limit order profitability on the ISE is analyzed

empirically. The chapter concludes with a summary.

4.2 A Comparison of the ISE with the Paris Bourse and the NYSE

The ISE will be compared to the two exchanges in which the order choice
decision has been studied. The aspects examined are market transparency,
price change limits, the use of alternative trading mechanisms, and the
settlement of transactions.

Transparency. The extent to which trading information is made available
after each discrete market event can be classified as ex ante and ex post. The
former is information available beforehand that enables the trade to take place,
that is the information required to resolve transaction price uncertainty. Ex post
transparency is the immediate publication of transaction prices and sizes to
resolve uncertainty about future prices. Handa, Schwartz, and Tiwari (1998)
argue that an increase in transparency helps traders better assess the execution
probability of limit orders and thus improves market efficiency.

In Paris, the CAC system provides three levels of trading information. The
first level that is available to everyone shows the last trade, current prices and
quantities at the best ask and bid and other daily summary statistics. The

second and third are available to brokers only. The second level consists of the

54

last five transactions and the current state of the book; including broker
identification numbers. The third level is similar to the second, but contains the
entire transaction record for the day.

Under the rules of the NYSE, the content of the book of limit orders is
known only to the specialist and may not be disclosed publicly. The specialist
quotes best bid and ask prices along with maximum order sizes for which the
execution is guaranteed at these quoted prices.

In Turkey, total quantities at each price level are disseminated to
members in real time. Moreover, for executed orders the identity of both sides of
the transaction are displayed.

Traders who supply liquidity but are concerned about the free option value
of their orders are encouraged to supply such liquidity in some markets by
allowing them to hide the full size of their order. Systems that incorporate this
option almost invariably impose some cost regarding the execution priority of the
hidden part.

In the CAC system, traders can hide part of their limit orders. This allows
' limit order submitters not to reveal their information or strategy to the public. The
hidden part of the order preserves price priority but not time priority. On the ISE,
hiding a limit order is not allowed.

On the NYSE, limit orders better than the current quote are not
automatically posted as new quotes. The rationale is conjectured to be the
desire of the NYSE to obscure true market prices. This action raises the

marginal costs of non-dominant competitors by necessitating a search to

55

discover true prices. Unlike the CAC procedure, the NYSE gives the specialist
the discretion to hide a limit order unless he is instructed by the submitter to
display it. Thus, it is easier for submitters to hide limit orders on the Paris Bourse
than on the NYSE.

Price Change Limits. In Paris, stocks are traded in either the so-called
forward or cash markets. In the forward market, for example, a trading halt of
five minutes occurs if price exceeds the closing price of the previous session by
more than 10%. If the price pressure continues, then a second halt occurs, and
the maximum possible price change during a day is 20%.

The maximum price change regulation on the ISE is similar to that on the
Bourse. Price cannot exceed the preannounced base price by more than 10%
during a trading session.

On the NYSE, the whole market closes if the index trips a circuit breaker,
but suspension of individual stocks is discretion based rather than rule based.
Order imbalance halts are initiated by the specialist, after consulting with floor
officials. A news halt may be initiated either by the exchange or by specialists.17
To achieve a fair and orderly market, the NYSE provides price continuity-depth
guidelines for specialists. These vary across stocks, since they depend on the

price range and normal trading volume of each stock. Adherence to the

 

‘7 Bhattacharya and Spiegel (1990) found that 49.1% of suspensions in their
sample were brought by the announcement of news, and another 48.5% were
caused when specialists observed a severe order imbalance. The typical price
change during a suspension was 6.7% during 1974-1988.

56

guidelines is one of the several criteria by which NYSE specialists are evaluated,
and those who perform poorly risk not being assigned more profitable stocks in
the future or may even have their stocks reassigned to others. Trading
suspensions can take the form of delayed openings or intraday suspensions."3
NYSE policy concerning opening prices places particular emphasis upon
minimizing subsequent price volatility and price reversals.

The treatment of large market orders in some pure limit order markets
can be interpreted as an indirect way of imposing price continuity rules. In the
CAC system, due to the concern for transaction price, large market orders may
not get immediate and full execution. After consuming the entire depth at the
best quote on the opposite side of the book, the unexecuted part of a large
market order is converted into a limit order at the price of partial execution. This
limit order waits the arrival of an opposing order.

On the ISE there are no market orders.19 Helsinki Stock Exchange’s HETI
system has the same feature. In addition, HETI does not allow a limit price to
exceed the best price level on the opposite side of the book. Whereas

aggressive traders on the Paris Bourse can submit a suitably priced limit order

 

‘3 Regardless of the cause, all trading delays that exceed 29 minutes for delayed
openings and 14 minutes for intraday suspensions must be officially approved by
a floor governor or a duly appointed floor official. Once trading has been halted,
the suspension must continue for at least 15 minutes to permit the
announcement of suspension on the exchange ticker and to provide time for
investors to respond by changing outstanding orders and submitting new orders.
‘9 This is not of any practical consequence. A trader has to use the best limit
price on the other side of the market to get immediate execution.

57

rather than a market order to get immediate execution for large order sizes, this
opportunity does not exist in the HETI system.

In the early 19905, the competition from London’s quote-driven dealer
system, Stock Exchange Automatic Quotation (SEAQ), forced Paris to
implement an innovation. Jacquillat, Schwartz, and Hamon (1995) give a
description of and rationale for PIBAL (Programme d’lntervention en Bourse pour
I'Amelioration de la Liquidité), which is used in the CAC system to add liquidity to
the market and increase its efficiency. Companies listed on the Bourse use their
own shares and cash to set up a fund to buy shares in a falling market and sell
shares in a rising market. The fund trades in the opening call according to a
prescribed formula. The idea is as follows. The higher the elasticity of the
supply curve, the lower will be the price impact of a shift in the market demand
curve. Thus, the fund dampens price volatility by adding liquidity to the market.
The fund also has an indirect effect on liquidity. An improvement in liquidity is
generally believed to attract additional order flow, which leads to further
improvement. Since the use of the fund delays the adjustment of price to its new
equilibrium value, the addition of liquidity cannot be done effectively in this way in
a continuous market. In a call market, the fund will trade after observing the total
imbalance in the market, but this is not possible in a continuous market, and the
operation of such a fund would give some traders the opportunity to make
unjustified profits.

Alternative Trading Mechanisms. Both the Paris Bourse and the NYSE

have call auction at the opening of each trading day, a feature not shared by the

58

ISE. Whereas entire order flow information is disseminated in Paris before the
call auction takes place, only some imperfect information about the extent of the
order imbalance is given by the specialist on the NYSE.

A parallel upstairs market exists on both the NYSE and the Paris Bourse,
but block trading on the ISE is conducted from 9:15 am. to 9:45 am. just before
the other markets open. In the block trading of existing shares, if the trade leads
to a change in the control of the corporation, there is no restriction on the
transaction price. Othenrvise, the transaction price should be within 1 20% of the
average WAP during the past 15 days. In Paris, since 1989, blocks can be
negotiated off the exchange. The procedure is similar to that on the NYSE.
Blocks are negotiated by upstairs traders and then brought to the exchange.
Unlike the procedure for small trades, block trades are allowed to occur outside
the bid-ask spread on the Paris Bourse, but the limit orders triggered by the
block price must be cleared at their limit prices. On the NYSE, limit orders are
traded at the block price.

In Paris, small orders can trade off the exchange by following the rules on
the transaction price. ISE members can execute transactions off the exchange
for odd lots only.

Settlement of Transactions. French companies are listed in either the
official list or the second market, the difference being that the official list contains

stocks with high trading activity. The most active stocks on the official list are

59

traded on a forward basis, and the settlement occurs at the end of each month?0
Because the extended time to settlement introduces additional payment risk,
there is a performance margin requirement on these fonNard contracts. Unlike

the Bourse, all NYSE and ISE stocks are traded on a cash basis.

4.3 The Viability of a Pure Limit Order Market
4.3.1 The Tradeoff between Market and Limit Order Strategies
The following tradeoffs affect traders’ order choice decision. Market

orders receive certain and immediate execution at a cost equal to half the
prevailing spread, assuming the quote midpoint represents the equilibrium value.
Limit orders benefit when liquidity trading moves price temporarily so that they
can be executed at a favorable price. There are two costs associated with
trading via limit order: the risk of nonexecution and the risk of being picked up by
informed traders due to the free option nature of limit orders. The first cost
measures the risk of having to transact at a disadvantageous price if a limit
order does not get executed. The second risk is costly, although the order is still
executed nominally at a better price than would be the case for a market order
submitted at the same time and in the same direction as the limit order.

A The equilibrium of the limit order book has been examined in the literature

(e.g. Glosten 1994 and Sandas 1999), but these studies ignore investor

 

2° The remaining stocks in the official list and all stocks in the second market are
traded on a cash basis.

60

eagerness to trade as well as other aspects of the order choice decision.
Abstracting from these issues and assuming the presence of a large number of
traders willing to place limit orders, the equilibrium of the limit order book at any
given time can be modeled as follows. The book contains orders at different
price levels. Execution probabilities and deviation of limit prices from the
equilibrium value of the stock at the time of execution vary among the existing
limit orders in the book. Since submitters cannot condition on the size Of the
order against which their limit order will transact, and there is competition among
a large number of potential submitters, the depth at each price level is
determined in a way to make the expected profitability of the marginal limit order

equal to zero.

4.3.2 The Method

The Hypotheses. Handa and Schwartz (1996) inject hypothetical orders
into the actual transaction data to measure performance of limit and market
orders. That procedure will be used in this research to test of the following
hypotheses.
Hypothesis 1: The expected total gain from limit order trading is negative.
Hypothesis 2: Traders who submit limit orders must be those for whom the

opportunity cost of forgoing a trade is low.

Handa and Schwartz (1996) claim that there are two conditions necessary

to maintain a balance between the limit and market order submission rates in an

order—driven market. First, there should be enough mean reversion in short-

61

term price dynamics. Second, there should be traders in the market for whom
the opportunity cost of forgoing a trade upon the nonexecution of their limit
orders is low. The second requirement describes a trader characteristic, and the
first refers to the mechanism that reestablishes equilibrium after it is disturbed.
The total cost of limit order trading can be decomposed into two parts: the
execution cost and the nonexecution cost. Without mean reversion in short-term
price dynamics, the first cost is positive, and the second is zero. Therefore, no
one will want to submit a limit order. Wrth mean reversion, however, the sign of
the first cost is ambiguous, and the second cost is positive. Since the total of
these two costs is positive, the only traders who may find limit order trading
desirable are those who are “patient”, that is, for whom the cost of forgoing a
trade is low. These traders will submit limit orders if the cost of execution is
nonpositive. This means that the deviations from the equilibrium price caused by
liquidity shocks are systematically corrected in the short run.

When the equilibrium between order submission rates is disturbed, for
example, when the share of limit orders in the market is disproportionally low,
short-term price volatility will increase. That will increase limit order profitability
and induce more limit orders to be submitted, which will reestablish the
equilibrium in order submission rates.

To see whether an investor who wants to trade but needs to decide on the
order type has an incentive to prefer a limit order strategy over a market order

strategy, the relative performance of these two trading strategies needs to be

62

examined. The expected value from the market order trading strategy is zero.21
The expected value of the limit order trading strategy is derived by using a
simple one-period model. The model makes predictions about the signs of the
total cost of a limit order trading strategy and its components, with and without a
mean reversion in stock price dynamics.

The Model. Consider the limit order book for a stock in a one-period
model. Assume that stock price at the beginning of the period, Po reflects all the
available information. A particular limit buy order, placed at the price Pun, < Po ,
will be examined. This limit order is assumed to have the highest time priority at
this price level. During the period, exactly one trader arrives at the market and
submits a market buy or sell order. VVlth the probability p he is informed (l), and
with probability q=1-p he is a liquidity trader (L). The probability distribution of
the end-of-period price, P,, , depending on the type of the arriving trader, is given
by two marginal densities f(P,, H) or f(Pn |L) , regardless of whether the limit order
under consideration is among those orders executed against the market order. If
the market order submitted by the arriving trader does not execute the limit buy
order under consideration, then it is converted into a market buy order and

executed immediately at the price prevailing at the end of the period.22

 

2‘ The bid-ask spread is ignored.
22 This price depends on the size and direction of the market order submitted by
the arriving trader.

63

Equation (5) in Handa and Schwartz (1996) gives the unconditional
expected cost of this limit order trading strategy that will guarantee execution by

the end of the period:
Pin Pi-m
EC: p. I(P....-Pn)f(Pn|l)dPn + q. I(P....-Po)f(Pn|L)dPn +
q. [(Pn-Po)f(Pn|L)dPn+ p- J(Pn-Pn)f(Pn|I)dPn;
Piim Plan

with E(P,,|L)=Po. (4)

The first term represents the expected cost when the limit order is
triggered by an informed trader. The second term is the corresponding
expectation when a liquidity trader takes the other side of the transaction. The
third term is the expected cost when the limit order does not get natural
execution after the arrival of a liquidity trader. The last term represents the
expected cost when there is no natural execution after the arrival of an informed
trader. In this case, the forced execution occurs at the price that equals the
expected value of the stock, given all the available information in the market and
thus the last term is equal to zero.“ 2‘ Hereafter, the first two terms combined
will be referred to as the bagging cost, and the third term will be called the

nonexecution cost.

 

’3 It is assumed that market price adjusts immediately and completely to the new
information.

2" One problem with this model is that it incorporates a limit order book, but it
uses a single market price by avoiding the spread.

A liquidity motivated trade leads to a deviation of market price from its
equilibrium value. It is assumed that this deviation is corrected in the short run.
The market price following an informed trade reflects the new equilibrium price.
Consider two scenarios. In the first, all traders are informed. In other words, q=0
in the expected cost equation. In this case, there is no mean reversion in the
price dynamic, and the bagging cost is positive, since trading against an
informed trader results in a certain loss. The nonexecution cost is zero, since
the market price equals the equilibrium value in this case. Therefore, the total
cost of limit order trading is positive.

In the second scenario, there are both informed and liquidity traders in the
market. In other words, q is positive in the expected cost expression. Since the
price impact of liquidity trading fades quickly, there is mean reversion in the
short-term price dynamics. The sign of the bagging cost is ambiguous, since the
gain from trading with a liquidity trader may compensate the loss from trading
with an informed trader. The sign depends on three factors: the likelihood that
the market order submitter is informed, by how much his information will change
the equilibrium price, and the likelihood that the transaction will change the price
so that the limit order will be executed, given that the market order submitter is
uninformed. The expected value of the nonexecution cost is positive. More
specifically, if the arriving trader is informed, then the expected nonexecution
cost is zero, since the market price at the end of the period equals the
equilibrium value. Otherwise, P0 is still the expected value of the true price,

given all the available information. Moreover, the market price at the end of the

65

 

 

period is expected to be higher than Po, since the fact that there is no natural
execution eliminates the possibility of this price being in the interval [0,P.,,,,].
Therefore, if there is sufficient liquidity trading in the market and prices are
corrected in the short run , the bagging cost will be negative. Nonetheless, it can
be shown that the expected total cost is positive even in this case. Therefore,
only traders for whom the cost of nonexecution is lower than the one assigned

by this analysis can have negative total cost from limit order trading.

4.3.3 The Experimental Design

In the experiment, hypothetical one-share limit and market buy orders are
used to assess both the profitability of these two trading strategies and whether
there is a mean reversion in price dynamics. The use of one-share orders has
two implications. First, if these orders had actually been submitted, they would
not have changed the transaction record. Second, since the hypothetical limit
orders are assumed to be executed when price falls to or below the limit price,
the results are only valid for small order sizes.

Timing of Events. Figure 4 describes the timing of events in the
experiment. Following the order submission at time zero, if the price falls to or
below the limit price by the end of the trading window, then the limit order is
assumed to be executed naturally, as shown in part A of the figure. In the figure,
the trading window is two days, and the execution occurs during the first day.
After execution, the stock is held during a period called the investment window.

In the figure, this period is three days, and the stock is sold at the end of the third

66

day following the execution. The investment window represents the time allowed
for the hypothesized mean reversion in price to take place.

If there is no natural execution during the trading window, as depicted in
part B of Figure 4, then the limit order is converted into a market order and gets
immediate execution. To be consistent with the natural execution case, the
stock is held during an investment window of the same length before its sale.

Alternatively, as shown in parts A and B of Figure 4, a market order gets
executed as soon as it is submitted, and the position is closed at the end of an
identical investment window. Irrespective of any news arrival during the trading
and investment windows, these naive strategies are strictly followed.

As Figure 4 shows, for each limit order there exists a corresponding
market order. Both the start and the length of the investment window for market
and limit orders do not match perfectly. Given the parameter values in the figure,
there is a minimum lag of one day and a maximum lag of two days between the
starting points. The length of investment window is exactly three days for market
orders, and it may vary between three and four days for limit orders, depending
on the time of execution. Any setup that arranges immediate and sure execution
for market orders but postponed and risky execution for limit orders cannot use
identical periods to compare the profitability of these two strategies. The
nonidentical periods will introduce some noise but will not bias the reported
values.

The Definition of Prices and Returns. Since the data set contains bid-

ask prices at market close, the end of the trading day is chosen as the time of

67

order submission. A trader can submit a market order and get immediate
execution at the prevailing ask price, or he can use a limit order, for which the
earliest time of execution is the next morning when the market reopens.

Returns are calculated as the difference in the logarithm of purchase and
selling price. For market orders, the ask price at the time of order submission is
the purchase price, and the selling price is the bid price at the end of the market
order investment window. For limit orders, the selling price is the bid price at the
end of the limit order investment window. For executed limit orders, the
purchase price is the limit price, and for unexecuted limit orders it is the ask price
at the time of forced execution.25 Limit prices are defined with respect to the
midpoint of bid and ask prices at the time of order submission. The limit price is
chosen k ticks below the quote midpoint. In the four limit order test categories, k

takes on the values 0.5, 1.5, 2.5, and 3.5.

4.3.4 Limitations of the Method

The trading rule is fairly successful in classifying hypothetical limit orders
correctly as executed or unexecuted, but in a few situations it does not work.
When quotes move so that the limit price equals the best ask price and no

transaction occurs until the end of the trading window, the hypothetical limit buy

 

25 Since the aim is to assess the profitability of hypothetical limit orders, when
there is natural execution, the limit price is taken as the execution price, even
though it may be greater than the price that triggers the trade.

68

order will incorrectly be classified as unexecuted. If the hypothetical limit buy
order were actually displayed in the limit order book, then the trader who
submitted the limit sell order at the best ask price would use a market sell order
and trade against the limit buy order. The trading rule is also less reliable when
used in a market with spreads typically larger than the tick size. In that case, the
price of the hypothetical limit order might fall between the best bid and ask
prices. If the hypothetical limit order were actually displayed in the limit order
book, then the price of immediacy would decrease for sellers in the market.
Therefore, if such a hypothetical limit order is labeled as unexecuted by the
trading rule, then it is likely to be misclassified. Since spreads hardly ever
exceed the tick size on the ISE, as was shown in chapter 3, only the first case
can lead to misclassification.

Another problem with the trading rule is that it follows a naive strategy. An
information event may occur before the natural/forced execution of a limit order,
but the naive strategy does not allow the original limit order to be replaced with
one that incorporates the change in equilibrium value brought about by that
information event. Yet, since positive and negative information events are
equally likely, their effects will cancel each other in the calculation of average
returns. Therefore, the use of a naive limit order submission strategy should not

bias the results.

69

4.3.5 Empirical Evidence in Previous Studies

The Handa and Schwartz (1996) method was used by Hamon et al.
(1993) to examine the profitability of limit order trading in fonivard and cash
markets on the Paris Bourse?6 In a later work, Hamon et al. (1995) compare
and contrast the evidence to test the validity of their conjecture that any viable
order-driven market requires that prices follow a mean reverting process. They
found that in Paris as well as on the NYSE bagging cost is negative and
nonexecution cost is positive. Overall, the limit order trading strategy
outperforms the market order trading strategy. Moreover, a limit order trader,
who is not facing competition from a specialist performs better on the Paris
Bourse than on the NYSE. In all the three markets (cash and fonivard markets in
Paris and the NYSE), the higher the difference between limit price and
equilibrium value of the stock at the time of order submission, the higher is the
nonexecution cost. lnterrnarket comparison shows that a high nonexecution cost
is associated with a high execution rate, which compensates the cost. The
authors concluded that market structure can only lead to second-order effects on
the basic compensation that traders require to supply liquidity to a securities

market.

 

’6 The stocks in the fonlvard market have higher transaction frequency, smaller
relative spread, and higher maximum price change limit in comparison to stocks
in the cash market.

70

Harris and Hasbrouck (1996) examined the profitability of limit and market
order trading strategies using both transaction and quoted price data. Ex ante
and ex post performance measures were employed to compare returns. The
first measure compared execution price and the price at the best quote on the
opposite side of the market at the time of order submission. The second
measure looked at these prices on the same side of the market five minutes after
execution. The former was computed for all limit orders, but the latter was
calculated only for limit orders that were naturally executed. The sum of the ex
ante and ex post measures gives the round-trip return for executed limit orders.
Had the ex post measure been computed for all limit orders, both natural and
forced execution, the sum would be analogous to the total return measure in
Handa and Schwartz (1996). Harris and Hasbrouck (1996) report the results
using prevailing spread, size and side of the order, and limit price position as
control variables.

The comparison of limit order strategies that differ in terms of limit price
position shows that those performing best are the most commonly used.
Compared to market orders, at-the-quote limit orders achieve better average
performance. Moreover, bettering the quote in markets where the prevailing
spread is larger than one tick is a better strategy than placing at-the-quote limit
orders. Harris and Hasbrouck (1996) also tried to infer the profit of a
hypothetical public trader who acts as a dealer. In contrast to the finding of
negative bagging cost in Handa and Schwartz (1996), they determined that the

round-trip return for executed limit orders is negative, even without including

71

commissions

To interpret these inconsistent findings regarding the sign of the bagging
cost, the empirical methods and samples in the two studies need to be
compared. First, the performance measures used are not the same. The ex
post measure in Harris and Hasbrouck resembles the so-called bagging cost in
Handa‘ and Schwartz, but there are two differences. (1) Rather than use a
closing trade, Harris and Hasbrouck compare the execution price to the same-
side quote. This is like using a market order to close the position, but the price
improvement factor is ignored. (2) The length of the investment window differs.
In the Harris and Hasbrouck study, the closing trade occurs five minutes after
natural execution, but this period varies between one and three days in the
Handa and Schwartz study.

The ex ante performance measure in Harris and Hasbrouck is somewhat
similar to the total expected gain from limit order trading in Handa and Schwartz.
Yet, the former uses the opposite-side quote at the time of order submission
rather than the price of the closing trade as the benchmark to compare the
execution price. The computation period also differ. In Harris and Hasbrouck it
is less than a trading day whereas in Handa and Schwartz it varies between four
and six days.

Second, there are differences regarding the segment of the total NYSE
order flow used in the two studies. Handa and Schwartz consider indirectly the
total order flow by using the entire transaction tape, but Harris and Hasbrouck

limit their sample to Superdot orders. Using their trading rule, Handa and

72

Schwartz examine only limit orders with the highest time priority at a given price
level, whereas Harris and Hasbrouck analyze all the limit orders in their sample,
which should result in a less attractive picture from the perspective of a limit
order trader. Clearly, natural execution is more likely in Handa and Schwartz
than in Harris and Hasbrouck.

In addition, recall that the derivation and empirical results in Handa and
Schwartz show that bagging cost is less than the total cost of limit order trading.
In other words, the nonexecution cost is positive. This implies that ex post
performance should be better than ex ante performance in Harris and
Hasbrouck, and this proved to be the case ignoring the problems caused by the
choice of time interval in their analysis. Therefore, the two studies agree on the

sign of the nonexecution cost.

4.4 Empirical Analysis
4.4.1 Data

The parameter values and number of stock-windows for each limit order
test are shown in Table 31. 283 trading days in the sample period are divided
into 10 subperiods of 28 days each, and every subperiod is further divided into
windows. Depending on the limit order test, the window length varies from four

days for the 0.5 and 1.5 tick categories to five and six days, respectively for the

73

2.5 and 3,5 ticks?7 For these four limit order tests, the sample period contains
2,100, 2,100, 1,500, and 1,200 potential windows, respectively.28

There is one limit order and one market order observation per window.
About 21% of potential observations were lost due to missing values for the bid-
ask spread and another 2% were lost due to the omission of an observation if
either a stock split or a dividend payment occurred during a window. When a
stock-window was removed from the limit order sample, it also was removed
from the market order sample, and vice versa. The final sample contains 1,546,
1,543, 1,181, and 953 windows in the four respective limit order test categories
after eliminating the problem cases.

The subperiod average per stock is the basic unit of observation. Results
are reported both as the cross-sectional average per subperiod and the overall
average for all subperiods. Table 32 reports for each limit order test and
subperiod the number of stock-windows and the number of stocks used to
calculate the subperiod average in the subsequent tables. The figures are
arranged into two categories based on whether execution of the limit buy order is

natural or forced in windows within each subperiod.

 

2’ The last three and four days of every subperiod were not used for the 2.5 tick

and 3.5 tick limit order tests, respectively.
28 For example, for k=0.5 tick there are 30 stocks x 7 windows per stock and

subperiod x 10 subperiods =2,100 windows.

74

4.4.2 Results

Descriptive Statistics. Table 33 shows the average execution prices
standardized by the market price at the time of order submission. Part A of the
table reports the results for all limit orders. Overall, the unconditional limit order
execution price is not different from the market order execution price. Part B of
the table displays similar figures for forced execution and reveals there is a
penalty: Had a market buy order been used, the stock would have cost from
4.21% to 6.89% less, depending on the limit order test category.

Table 34 reports the difference between standardized execution price for
limit orders and the triggering price. Overall, the latter is significantly less than
the submitted buy price by about 0.55% for all limit order tests.

Table 35 reports the time to execution in hours for naturally executed limit
orders. On average, about 22 minutes are required for the 0.5 tick, 58 minutes
for the 1.5 tick, 172 minutes for the 2.5 tick, and more than one trading day
(about five trading hours) for the 3.5 tick.

Overall Average Returns. Table 36 reports average returns for limit and
market orders, along with two related statistics. On average, for the four test
categories the limit price was set 0.779%, 2.310%, 3.861%, and 5.423%,
respectively below the market price at the time of order submission. Out of all
the limit orders submitted, the proportion naturally executed is 87.65% for the
0.5 tick, and the corresponding values are 66.36%, 58.00%, and 52.47% for the
other test categories, respectively. Therefore, as the price concession required

by a limit order increases, the probability of its execution declines.

75

The data on overall returns show the following picture. Unconditional
returns are significantly negative for both market and limit orders, although limit
orders show better average performance than market orders. Interestingly, for
the two lowest categories the limit order returns are significantly lower in the
case of natural rather than forced execution. As expected, limit order return
conditional on execution is as good as or better than the unconditional market
order return. Surprisingly, except for the highest test category, limit order return
conditional on nonexecution is significantly better than unconditional market
order return.

A regularity in Table 36 is the decrease in limit order return conditional
on nonexecution as the limit order test value increases. Two factors depend on
the value of limit order test. First, the upward tendency of stock price should rise
as the limit order test value declines. The information that price never falls by
0.5 tick during the trading window implies a larger upward trend in price than
suggested by the information that price has never fallen by 3.5 ticks during the
trading interval. Second, delay in purchasing a stock is costly in a rising market.
The time of forced execution is by design a nondecreasing function of the limit
order test value. Therefore, an increase in limit order test value will make a limit
order strategy less attractive in a period of rising stock prices. These two factors,

together or individually, are consistent with the decrease in limit order return as

the limit order test value increases.

76

Differential Returns. To measure the total, bagging, and nonexecution
costs, the difference between limit order return (both unconditional and
conditional) and unconditional market order return is used. Tables 37A, 37B,
and 370 report the unconditional differential returns, differential returns
conditional on execution, and differential returns conditional on nonexecution,
respectively. Unconditional market order returns serve as a benchmark to
assess the profitability of a limit order trading strategy.

Overall values for conditional differential returns, reported in Tables 37B
and 370, are not equal to the difference in corresponding values from Table 36.
For example, if there are no forced execution observations for a stock in a
particular subperiod, then the unconditional market order return of this stock will
not affect the differential return conditional on nonexecution for this subperiod.
The averages given in Table 36 do not reflect this constraint, which arises from
the definition of differential return. The difference between the values reported in
tables 36 and 37B-C can be quite significant. As was shown in Table 32, for the
0.5 tick category there were subperiods in which almost all limit orders were
naturally executed.

Except for the 0.5 tick limit order test, bagging costs are significantly
negative, which is consistent with the sufficient price mean reversion

requirement suggested by Handa and Schwartz (1996). For the two lowest test

77

categories, the nonexecution cost is negative; this result is inconsistent with the
Handa and Schwartz model regardless of mean reversion requirement. The
nonexecution cost increases with limit order test value but is significantly positive
only for the 3.5 category. Total costs do not support the predictions of the
model. These are negative for all the test categories and are significant only for
ticks of 1.5 and 2.5. These results are qualitatively similar to those reported in
Handa and Schwartz (1996) and Hamon et al. (1993) for the NYSE and the Paris
Bourse, respectively.

Investment Window Length. The choice of a three-day investment
window is arbitrary. It was selected so that results would be comparable to those
in previous studies of the NYSE and the Paris Bourse. In general, there is a
tradeoff in the choice of investment window length. On the one hand, it should
be long enough to allow time for the hypothesized mean reversion to take place.
On the other hand, it should be short enough to avoid picking up noise from
another event that may follow the hypothesized liquidity event.

To determine whether the three-day window length is appropriate, the
midpoint of closing bid and ask prices was tracked for 10 days following the
order submission. Table 38 reports the average prices standardized by the
spread midpoint at the time of order submission for both executed and
unexecuted orders. Figure 5 graphs the data for executed orders. It appears
that for the 2.5 and 3.5 test categories there is mean reversion in price, but a
window of at least eight days is necessary for this process to be completed. For

the 0.5 and 1.5 test categories, there is no reason for price correction following a

78

small liquidity shock to take that long. Nonetheless, as Figure 5 shows, price
rises, on average till the end of the seventh day, an observation for which no
explanation is offered in this analysis.

The experiment was repeated twice, using investment windows of five and
seven days. The definition of subperiods was not changed to maintain the
comparability of results with those previously obtained. Details about the
findings are reported in the Appendix. The number of valid observations for the
four test categories falls to 980, 969, 975, and 636 when the investment window
is five days. The corresponding figures are 621, 617, 710, and 478 when the
investment window is seven days.

Table 39 summarizes the results by showing the overall average of the
three costs for each choice of investment window. Except for the 3.5 tick test
category, the bagging cost becomes significantly positive as the investment
window lengthens from three to five days. Had the bagging cost decreased, it
would mean that an investment window of three days is insufficient for the mean
reversion to be completed. Because the bagging cost increases significantly in
three of the test categories, it appears that the mean reversion cycle is shorter
than five days. The bagging cost is significantly negative for all the categories
when the investment window is extended to seven days. These sign changes
most likely indicate the effect of new information or other liquidity events that
follow the original shock. If for each stock new events are independent and
equally likely to occur in positive and negative directions, then they should not

bias the cross-sectional averages that are used to measure the effect of the

79

original liquidity event. It is possible that marketwide events dominate firm-
specific events. If so, then subsequent events will significantly distort the
measurement of bagging cost associated with the original shock. This possibility
will be examined in the next section by using a market adjustment to concentrate
on firm-specific events.

Market Adjusted Returns. If prices in the market move together, in other
words, if the marketwide information is the major factor that moves prices, then it
is appropriate to test the predictions after making adjustment for marketwide
events.

In order to find market-adjusted differential return for stock k during
window i in subperiod s, the average market order return for stock k during
windows in subperiod s is subtracted from limit order return for stock k during
window i in subperiod s:

Rm= rlkis - rm... - (5)
In the same way, the differential return of a limit order strategy for a portfolio that
includes all stocks except k is formulated by:
R9,,= r‘pis - rm” . (6)
The differential stock return is regressed on a constant and the differential
portfolio return for all the stocks in the sample:
Rm; ork + [3,, R9,, + a”, Vk . (7)
The residuals e,,, from this regression are stacked into a vector, which

contains the differential limit order returns after removing the effect of

80

 

marketwide events. This vector of residuals is regressed on a constant and a
dummy E, which equals one if the limit order is naturally executed, zero
otherwise.

e... = n. + m E... + V... - (8)

The coefficients of the second-stage regression measure the effect of
finn-specific events only. -n, shows the nonexecution cost, and -(n, +112)
indicates the bagging cost. The assumption of a mean reversion in price
dynamics predicts negative and positive signs for 111 and 112. respectively.

The results of this second-stage regression are shown in Table 40. For all
the test categories, whenever a coefficient is significant in a subperiod, its sign is
consistent with the prediction. Therefore, this is evidence that, after removing
the effect of marketwide information events, there is mean reversion on the ISE.
It appears that marketwide information events affect short-term price dynamics to
a greater extent on the ISE than on the NYSE or the Paris Bourse.

Table 41 compares the overall regression results for the three choices of
investment window.29 Unlike the unadjusted returns discussed in the previous
I section, the signs of the adjusted regression estimates are not sensitive to
window length. Although the magnitudes of i1, and nzvary, the bagging costs are
fairly similar. These regression results also indicate that the bagging cost

becomes more negative as the test category increases.

 

’9 The regression results for investment windows of five and seven days are
given in Table 56 (Appendix C).

81

Profits from Voluntary Market Making. So far, it has been assumed
that the problem is to determine the preferred trading strategy once the decision
to trade has been made. Thus, the method of trading was part of a more
complicated investment decision. In this section, limit order trading is examined
as an independent strategy. The question is whether a trader who acts as a
voluntary market maker can earn profits. Therefore, the market order trading
strategy, no longer serves as a benchmark.

Consider another experiment in which a trader submits simultaneously
both limit buy and limit sell orders. The orders are submitted at the price of ink
ticks from the equilibrium price. Each time an execution occurs, the limit orders
are renewed, and it is assumed that the most recent price that triggers execution
is the new equilibrium price. This strategy is pursued for three subperiods, each
93 days long. At the end of each subperiod the accumulated stock inventory is
eliminated. In this experiment, total loss during 93 days is decomposed into the
bagging and the nonexecution costs. Total loss from roundtrip transactions
measures the former cost, while the loss realized at inventory closing represents
thelafler
The total gain from submitting a network of limit orders during the 93 days is:

H (P. Ns' PbNb) - Pc(Ns' Nb) (9)

(Ps' Pb) min(NsiNb) + (Pav' Pc) (Ns- Nb)

gain per roundtrip*roundtrips +

per share gain from closing inventory* inventory imbalance
- bagging cost - nonexecution cost .

82

‘ w
J
l

where:
average selling price;
average buying price;
inventory closing price;
total number of shares sold;
b total number of shares bought; and
a, P8 if N,5 > NI, or
Pb if N, > N,
The nonexecution cost is expected to be positive no matter in which direction

'0 2.2503“?

price moves during the period in which a limit order strategy is used. More
specifically: (1) If the market is bearish: Then N,>Nb.therefore P,,,,=Ps ,and Pc>
P3,. (2) If the market is bullish: Then N5<Nb, therefore P,,,,=Pb , and Pc< P8,. In
both cases (Pav-Pc)(N,-Nb) is negative showing a positive nonexecution cost.
The sign of the bagging cost is ambiguous. More specifically: (1) If trading
occurs only with informed traders, then no matter whether the market is bearish
or bullish P5<Pb. (2) If trading occurs only with liquidity traders and there is
sufficient mean reversion in short term price dynamics, then P3>Pb. The first
term in equation (9), (P,- P.) min(N,,Nb) has opposite signs in these two cases;
hence the sign of the bagging cost is ambiguous.

Therefore, if there is sufficient mean reversion in short term price
dynamics then negative bagging cost can outweigh the positive nonexecution
cost and voluntary market making may be profitable.

Adjustment for Split and Dividend. The splits and dividends of sample

firms introduce a problem in using hypothetical one-share orders.3° To give

 

3° Other than adjusting the limit prices on ex-split and ex-dividend days.

83

equal weight to all the executions during the 93-day trading period, a share index
is created for each firm. It shows for each day how many shares are equivalent
to a share of stock at the beginning of the sample period. The index takes into
consideration stock splits and dividends as well as cash dividends.31 To find the
transaction price for each execution, the actual transaction price is multiplied by
the share index. The same adjustment is made to the price used to close the
inventory. With this adjustment, the experiment examines outstanding limit
orders for one equivalent share during the whole 93 days.

Two sets of experiments were conducted using a limit price of i 2 and :3
ticks around the equilibrium price. Table 57 (Appendix C) reports the number of
shares bought and sold per firm, as well as the maximum and minimum values of
the inventory. The number of purchases and sales are fairly similar.

Table 42 reports the results where stock-subperiod is the basic
observation. Part A reports the results on TL basis. Part B expresses the result
as a percentage of the price of one equivalent share at the close of each
subperiod, which prevents higher weighting of higher priced stocks. The table
shows negative bagging and positive nonexecution costs for both test
categories. Since the absolute value of nonexecution cost is greater than that of

bagging cost, the total cost is positive for the two cases, but it is significant only

 

3‘ The money to participate in a rights issue is obtained by selling stock, and the
cash dividend is used to purchase new stock.

 

for the i3 tick category. Individual stock averages (not reported) show that the
round-trip gain is always positive. Moreover, it is found that the per-share gain
from closing inventory and share imbalance at the close always have opposite

signs. In other words, the nonexecution cost is positive for individual stocks.

In the voluntary market making experiment there is no need to choose
arbitrary parameter values for the trading and investment windows. Without
imposing these constraints, the results are qualitatively consistent with those
from the market-adjusted single limit order experiment. This strengthens the
conjecture that the inconsistent results obtained from unadjusted differential
returns arise because dominant marketwide events create a measurement
problem when the investment window is chosen different from the true cycle
length of the firm-specific mean reversion.

lnterrnarket Comparison. The fraction of executed limit orders may
provide an opportunity to compare the balance between market and limit order
submission rates on the ISE with those on the NYSE and the Paris Bourse. For
the NYSE, these proportions are 46%, 39%, and 35% for orders placed 1%, 2%,
and 3% from the equilibrium value at the time of order submission. For these
same categories, the corresponding figures are 45%, 40%, and 34% on the
Paris fonivard market and 21%, 25%, and 19% on the Paris cash market. For
the ISE, the proportions per test category are: 88%, 0.5 tick; 66%, 1.5 tick; 58%,
2.5 tick; and 52%, 3.5 tick. These test categories represent a variation from

equilibrium value between 0.8% and 5.5%.

85

These data imply that short-term volatility on the ISE is larger than on the
NYSE. One possible reason for the lSE’s volatility may be the absence of an
opening call, a useful mechanism for price discovery. In addition, the
opportunity to hide limit orders on the Paris exchange, and the existence of
specialists as well as the limited disclosure of the limit order book on the NYSE
may explain some of the difference. Because these features create transaction
price uncertainty for informed traders, they protect limit order submitters to a
certain extent, and may encourage the use of limit rather than market orders by
uninformed traders. If so, then the depth at all price levels will increase, which
reduces short-term volatility.

The choice of a large relative tick by the ISE may affect the balance of
market and limit order submission rates. That is, it may be designed to enhance
the attractiveness of a limit order trading strategy for small orders. A large
relative tick sets a floor for relative spread. An increase in the relative spread
increases both the cost of a market order trading strategy and the gains from a
limit order trading strategy for small orders. The execution rates imply that —
even with a large relative tick - the balance is weaker on the ISE than on the

NYSE and the Paris Bourse.

86

4.5 Summary

A trading rule introduced by Handa and Schwartz (1996) was used to
compare the performance of limit and market order trading strategies. This rule
injects hypothetical orders into the transaction tape to estimate the expected
round-trip returns from these two strategies.

It is conjectured that unconditional returns from limit order trading are
negative. Only traders who can forgo trading when their limit orders are not
executed can find limit order trading attractive. This requires sufficient liquidity
trading and a mean reversion in short-term price dynamics.

The results show that average short-term returns from these naive
strategies are negative. The method of using average differential limit order
returns is not reliable, because results are highly sensitive to the choice of
investment window length. The reason seems to be that marketwide rather than
firm-specific events are the major factor in stock price movement in the short run.
After adjusting the differential returns for marketwide events, the results are
stable and generally consistent with predictions of the Handa and Schwartz
(1996) model. There seems to be enough mean reversion in price so that when
firm-specific events occur, limit order return conditional on execution is greater
than unconditional market order return; in other words, the bagging cost is
negative. Moreover, for return conditional on nonexecution, the relationship is in
the other direction. Stated differently, the nonexecution cost is positive.

The results also show that a trader who acts as a market maker will make

negative average profits on the ISE. This is consistent with findings in two other

87

studies, which used the same trading rule to examine the NYSE and the Paris
Bourse.

Compared to the NYSE and the Paris exchange, the fraction of executed
limit orders is large on the ISE, which indicates higher short-term price volatility
on the Turkish exchange. Two explanations can be given. First, unlike the ISE,
trading on the NYSE and the Paris Bourse start with a call auction, which is
thought to enhance price discovery in a market. Its absence is likely to result in
high intraday volatility. Second, high volatility may reflect an imbalance in market
and limit order submission rates. This suggests that even the use of a large tick

size is not sufficient to bring the rates into balance.

88

Chapter 5

Relationship between Trading Volume and Price Change on the ISE

5.1 Introduction

In economics, changes in market equilibrium are thought to depend on
information arrival. Once information is incorporated into prices, a new
equilibrium is established. The mechanics of price formation can be modeled by
using two approaches. According to the first, known as the rational expectations
framework, investors use price information to find their equilibrium demand. The
second employs the concept of a Walrasian auctioneer who aggregates the
supply and demand schedules of individual investors and finds the price that
clears the market. Although trading occurs continuously in actual markets, both
approaches ignore the adjustment process and concentrate on the properties of
the new equilibrium.

The adjustment process is a focus of the market microstructure literature.
During the adjustment period, existing information becomes obsolete and
uncertainty about the true asset value increases. It is reasonable to predict a
positive association between asset price volatility and the strength of the new
information. Furthermore, whether the information is private or public is likely to
be important. Such public information as the announcement of earnings or a
merger is believed to be incorporated into price quickly and directly without
generating abnormal trading activity. In contrast, private information is revealed

through the timing and size of informed trades. Depending on competition

89

among informed traders and whether information is short or long lived, the price
adjustment can take some time. Microstructure theory investigates the strategic
behavior of informed traders and considers stealth trading a natural strategy.
Nonetheless, it still relies on increased trading activity to model the learning
process of uninformed traders. Although abnormal trading activity is not thought
to be part of the price adjustment process following the arrival of public
information, there is likely to be a period of high trading characterized by hedging
and speculation.

Since both price volatility and trading volume are related to market
information flow, a contemporaneous relation between them is likely.32
Uncovering the relationship between these two variables is important to an
understanding of how prices adjust to new information.

In most theoretical models in the microstructure literature e.g., Brown and
Jennings (1989) and Grundy and IllcNichols (1989), price adjustment is not
instantaneous. Empirically, transitory liquidity effects add noise to the
measurement of permanent information effects. Because the price adjustment is
not immediate, such market statistics as the price-volume sequences may be
useful. Unfortunately, theory does not provide a definite answer about what

traders can learn from market data beyond the current price. Yet, the fact that

 

32 One should be careful in this argumentation. If the new information is a public
signal and transaction level data is used, then this argument would be incorrect
since the abnormal trading activity is believed to follow the price adjustment. For
less frequent data sampling, this argument should be valid.

90

technical analysis is widespread in seemingly efficient markets indicates that
there may be a dynamic relationship between price and volume.

The interaction between trading volume and price change has been an
issue for almost 40 years (Granger and Morgenstern 1963). As pointed out in a
survey article by Karpoff (1987), one benefit of investigating this relationship is
the insight gained about the structure of financial markets. Relevant factors
noted in the literature include the flow of information, its dissemination, the extent
to which prices reflect information, and the effect of market frictions, such as the
cost of taking a short position.

This chapter examines the intraday relationship between price and
volume on the ISE. Empirical evidence is provided on both a contemporaneous
and a dynamic relationship. An examination of the univariate temporal patterns
in price and volume precedes bivariate analyses. The next section reviews the
literature on the price-volume relationship and temporal patterns. It is followed

by an empirical analysis and a brief conclusion.

5.2 Literature Review
According to Gallant et al. (1992, p. 201),

Generally speaking, the empirical work on price-volume relation tends to
be very data-based and not guided by rigorous, equilibrium models of
market behavior. The models are more statistical than economic in
character, and typically neither the optimization problem facing agents nor
the information structure is fully specified. The intrinsic difficulties of
specifying plausible, rigorous, and implementable models of volume and
prices are the reasons for the informal modeling approaches commonly
used.

91

' -. finu‘l

The evolution of theoretical research in this area can be separated into
two periods. The models in the early period saw the arrival of new
information as the factor that generates trades, but they did not distinguish
between private and public signals. The second period began with the
introduction of the private information models, which is usually credited to
Bagehot (1971), but the concept of trades themselves as signals of
information was first developed by Glosten and Milgrom (1985) and Easley
and O’Hara (1987). A new generation of work emerged after the concept of
asymmetric information was introduced into the microstructure literature.

Empirical research has identified at least two characteristics of the price-

volume relationship.33 Trading volume is positively correlated with both price
change and its absolute value. Moreover, the ratio of volume to price change for
upticks exceeds the absolute value of the same ratio for downticks. To explain
this difference, Karpoff (1987) argues that if the true relationship between the
two variables is asymmetric, then incorrect specifications that force a functional
and/or monotonic relation between them can lead to these somewhat
I inconsistent findings for upticks and downticks. Asymmetry has been confirmed
in stock and bond markets, which Karpoff believes can be a consequence of the
extra cost involved in taking a short position. His explanation is supported by
Foster (1995), who reports a symmetric relationship in crude oil futures markets,

where there is no difference in the cost of long and short positions.

 

33 See the survey article by Karpoff (1987) for a list of empirical works.

92

5.2.1 Early Research on Price-Volume Relationship

The positive correlation between volume and the absolute value of price
change is consistent with both the mixture of distributions hypothesis and the
sequential information model. An important difference between them is the
speed with which a market moves to the full information equilibrium. The former
assumes that upon the arrival of new information the new equilibrium is reached
immediately; the latter theorizes that the final equilibrium is attained after passing
through a number of incomplete equilibria.

Mixture of Distributions Hypothesis. Work by Clark (1973), Epps and
Epps (1976), and Tauchen and Pitts (1983) is consistent with the mixture of
distributions hypothesis, which states that price changes over time are sampled
from distributions with different variances.

The Clark (1973) study belongs to the body of research on the distribution
of speculative prices. In an effort to explain the leptokurtis in the distribution of
daily stock price changes, Clark points out the distinction between transaction
and calendar time. He argues that although the price change process for
individual transactions may be Gaussian with constant variance, the random rate
of daily information arrival to the market makes the central limit theorem
inapplicable. Total volume and price change over a fixed period are summations
of statistics generated from individual transactions during this interval.

Therefore, price change during a fixed period is a mixture of independent
normals, with the number of new pieces of information flowing into the market

being the mixing variable.

93

Clark (1973) conjectures that volume during a fixed period is related
positively to the number of new pieces of information. Hence, volume is a proxy
for the rate of information arrival, and this makes price change variability during a
fixed period proportional to the volume generated. Clark uses the daily change
in cotton futures prices to test this hypothesis. By grouping observations
according to the level of volume, he shows that the kurtosis of the distribution is
significantly reduced for each volume category relative to the kurtosis of the
entire sample. In other words, the distribution of price change adjusted for
operational time looks more like the normal distribution, which supports the
hypothesis.

Another version of the mixture of distributions hypothesis is presented by
Epps and Epps (1976), who examined volume-price changes from one market
clearing to the next, rather than over a fixed interval. Therefore, although their
model predicts the same relationship as Clark, their result is not driven by the
random rate of information arrival. In the Epps and Epps model, after new
information arrives, each investor updates his beliefs about the distribution of
assets” end-of-period values. This updating changes means, but not variances
or covariances of these distributions. The change in the equilibrium price is the
average of changes in traders’ reservation prices. The critical assumption in the
model is that the higher the absolute value of the average change in traders’
reservation prices, the higher is the disagreement among investors. Assuming
that high disagreement among investors is associated with high volume, a

positive correlation between the absolute value of price change and volume

94

arises.

Tauchen and Pitts (1983) formalize Clark’s idea by using an economic
model in which the change in reservation prices of investors has a common and
an investor-specific component. There is functional dependence between the
first two moments of price change and trading volume at each market clearing,
but the two variables are stochastically independent. During a fixed period, the
relationship between volume and price change variability is due to their common
positive connection to the number of new pieces of information arriving to the
market.

One goal of Tauchen and Pitts (1983) is to explain the increase in volume
and decrease in daily price change variability in the Treasury bill futures market
over time. Their model shows that both price change variance and expected
volume over a fixed interval depend on the mean arrival rate of information to the
market, the extent to which traders disagree when they respond to new
information, and the number of active traders in the market.

The mixture of distributions hypothesis implies that price change variance
over a fixed interval is heteroskedastic. Lamoureux and Lastrapes (1990) relate
the observation of persistent return volatility in financial markets to that
hypothesis. They suggest that the GARCH effects in stock returns may reflect
serial correlation in the number of information arrivals. Using daily stock return
and volume data, they found strong support for this view. When
contemporaneous volume was added to the conditional variance equation, its

coefficient was significant. Moreover, the GARCH effects disappeared after the

95

inclusion of volume.

A similar analysis by Najand and Yung (1991) used data from the
Treasury bond futures market. Unlike Lamoureux and Lastrapes (1990), they did
not find that the GARCH effects disappeared when contemporaneous volume
was included to the variance equation, volume was not significant in explaining
the variance. The suspicion of simultaneity then led them to use lagged volume
as an instrument for contemporaneous volume, and it proved significant.
Nevertheless, the GARCH effects remained significant even after solving the
simultaneity problem. A later study by Foster (1995) of the market for crude oil
futures supports the findings in Najand and Yung.

Sequential Information Model. A positive correlation between volume
and the absolute value of price change is also predicted by Copeland’s (1976)
model, in which information is received sequentially by investors. They are
classified as optimist or pessimist depending on the way they interpret the new
information. Investors shift their demand curve up or down by a fixed amount
after receiving the information. Since short selling is not allowed in the model,
the volume generated by a pessimist in response to the information is lower than
the volume generated by an optimist. Total volume and price change in this
model depend on the composition of investors and on the amount of information.
In addition, total volume depends on the order in which information is received.
Expected total volume reaches its maximum when there is complete agreement
among investors, a counterintuitive result caused by the no short selling

constraint. Expected total volume and the absolute value of price change attain

96

their minimum for the same fraction of optimists in the market, and the absolute
value of price change increases with volume. This shows the positive correlation
between the two variables.

To explain asymmetry in the volume-price change relationship, Karpoff
(1986, 1987) pointed to the cost of short selling in most markets. Another model
Epps (1975) that examines asymmetry arrives at a different explanation. Epps
creates a portfolio selection model that classifies investors as optimist or
pessimist depending on their beliefs about the end-of-period value of an asset.
They accept a new piece of information that reinforces their beliefs and
otherwise ignore it. Another critical assumption by Epps concerns the effect of
new information. Each investor changes his belief about the end of period value
of the asset in response to news shocks in such a way that the coefficient of
variation remains constant. This assumption implies that optimists have a
steeper demand function than pessimists, which leads to the result that the ratio
of volume to absolute value of price change is higher for price upticks than for
downticks.

In the models discussed thus far, the results are generally not driven by
the maximization of objective functions of different investor groups. Instead,
these models make strong assumptions that are hard to justify, such as ignoring
information that contradicts beliefs, or the artificial classification of investors as
optimists or pessimists. These features of early research affected the way

empirical studies were conducted in this area.

97

5.2.2 More Recent Work

The informational role of volume is examined in more recent works that
are not subject to the same criticisms as the early models. One approach
analyzes the volume that emerges when traders with different information signals
transact. Another approach focuses on what traders can learn from observing
volume.

Informational Role of Volume. As an example of the first approach,
Wang (1994) created a model of stock trading in which investors are
heterogeneous with regard to information and private investment opportunities.
Informed investors trade when they receive private information about the stock’s
future cash flows or to rebalance their portfolio when their private investment
opportunity changes. The uninformed are willing to trade with the informed,
since not all the trades from the informed traders are information motivated.
Investors follow dynamic trading strategies to maximize lifetime expected utilities.
Market clearing price does not fully reveal the informed investors’ private
information because they have two possible motives for trade. The effect of
information asymmetry on the behavior of volume is Wang’s major concern. As
that asymmetry increases, uninformed investors demand higher price
concessions in trading with informed investors. Therefore, trading volume is
always positively correlated with absolute price changes, and the correlation
increases with information asymmetry. Expected future returns conditional on
current volume and return depend on the extent of information asymmetry.

Thus, the dynamic return-volume relationship can reveal the nature of investor

98

heterogeneity in the market.

The second approach is exemplified by Blume, Easley, and O’Hara
(1994), who focus on the Ieaming problem that arises when traders condition on
the information conveyed by volume. If the process by which prices adjust to
information is not immediate, then market statistics may contain information that
has not been incorporated into the current market price. In this model, volume
does not merely describe but affects market behavior. Therefore, Blume et al.
give one explanation for the existence of technical analysis in seemingly efficient
markets.

In this Walrasian model, informed and uninformed traders receive signals
of different quality (precision). Both the level and the quality of the informed
traders' signal is unknown to uninformed traders. Price incorporates the
aggregated value of underlying signals, whereas volume conveys information
about the signal quality of informed traders that can be used, together with price,
to make inferences about the true value of the asset.

The relation between volume and information is not linear in the Blume et
al. framework. Both low and high volume may indicate the arrival of new
information. Volume is related to the dispersion of investor beliefs, and the link
between dispersion and information is complex. In equilibrium, volume is strictly
convex in price, and a V-shaped pattern emerges. Therefore, absolute price
change and volume are positively correlated. Both information quality and the
level of information dissemination affect the price-volume relationship. The

dispersion of the distribution of the price changes decreases with an increase in

99

information quality, while it increases with volume given a high level of
information dissemination. The latter prediction is consistent with the empirical
evidence given by Gallant et al. (1992). According to Blume, Easley, and
O’Hara (1994), past volume (and price sequences help to make inferences about
the true value of a security. Thus, a dynamic relationship between volume and
price is consistent with this model.

A number of studies ( Cheung et al. 1993, Foster 1995, Gwilym et al.
1999 , Hasbrouck 1991, Jain and Joh 1988, and Stickel and Verrecchia 1994)
give empirical evidence of the dynamic relationship between volume and price
change.

Jain and Joh (1988) and Cheung et al. (1993) use intraday data to
investigate the contemporaneous and causal links between price change and
volume in equity markets. Both studies confirm a positive asymmetric
contemporaneous relationship and find that return causes volume in the Granger
sense.

Hasbrouck (1991) follows information-based theoretical models, which
suggest that security prices respond to trading activity as a consequence of
asymmetric information. He uses vector autoregression (VAR) modeling
because he maintains that microstructure imperfections necessitate the use of
lagged price, trade size, and trade direction terms in estimation. Factors related
to public trader behavior, such as order fragmentation and price pressure, as
well as dealers’ concern for inventory control lead to serial dependence in

transactions. Moreover, certain trading rules, such as price discreteness and

100

exchange-mandated price smoothing requirements, result in lagged adjustment
of price to new information.

The dynamic structure of the VAR system permits the measurement of
price response to a single impulse in unanticipated trade size. The unanticipated
trade size gauges the effect of private information. The use of impulse response
function picks up the persistent price impact and thus eliminates the transient
inventory control and liquidity trading effects. The unanticipated price change in
the VAR model reflects the effect of public information.

Hasbrouck (1991) provides evidence that price impact is a positive,
increasing, and concave function of trade size. Another finding is that Granger
causality runs in both directions. Contemporaneous and past trades affect the
current change in price, which suggests a lagged price adjustment process. The
negative relation between current trade and past price changes is consistent with
both inventory control effects and the price experimentation hypothesis, which
maintains that market makers set quotes to extract information optimally from
traders.

Stickel and Verrecchia (1994) examine how trading volume influences
subsequent price change around earnings announcements. They hypothesize
that price changes are more likely to reverse following weak volume support
than strong volume support. It is argued that price changes reflect demand for a
stock, and higher volume increases the likelihood that the demand originates
from informed rather than uninformed trading. They give evidence that large

price changes on days with weak volume support tend to reverse the next day,

101

after controlling for the bid-ask bounce effect.

Foster (1995) reports a contemporaneous relationship between price
volatility and volume, after controlling for lagged observations of these variables.
This suggests that volume and volatility are both driven by the same factors, one
of which may be information, as is assumed in the mixture of distributions
hypothesis. The coefficient estimates of lagged volume and volatility are
significant in Foster’s study and support the prediction in the Blume, Easley, and
O’Hara (1994) model that past volume can be used to explain future price
volatility.

Gwilym et al. (1999) investigate the intraday relationship between volume
and volatility in LlFFE futures markets and show a positive contemporaneous
correlation between the two. Moreover, causality is bidirectional between
volatility and volume. The authors argue that the latter finding supports the
sequential dissemination of information in LIF FE futures markets. This is
attributed to the potential for profitable trading due to short-lived informational
advantages, which stem from the low transaction costs and margin requirements

in this market.

5.2.3 lntra and lnterday Patterns in Volume and Return Volatility

Empirical research provides evidence of temporal patterns in volume and
return volatility. Wood, Mclnish, and Ord (1985) report that intraday volatility is
U-shaped. A systematic pattern in interday returns is found in Harris (1986),

who shows that significant interday differences in returns occur during the first 45

102

minutes after the market opens. Moreover, prices drop on Monday morning, but
rise on other weekday mornings. Jain and Joh (1988) report that volume has a
U-shaped intraday and an inverse U-shaped interday pattern.

Theoretical work on this issue consists mainly of the intraday model of
Admati and Pfleiderer (1988) and the interday framework of Foster and
Visvanathan (1990). Both use the concept of discretionary liquidity traders to
derive temporal patterns. They differ primarily in their assumptions about the
nature of private information (whether it is short or long lived ) and the number of
informed traders in the market.

In the Admati and Pfleiderer model, there are multiple informed traders,
and they receive a private signal in each period. Their information is short-lived
(becomes public at the end of the period). There also are discretionary liquidity
traders, who have the flexibility of choosing the period in which they will trade.
To minimize their trading cost, they concentrate their trades in the same period.
The reaction of informed traders is to increase their trading in the same period,
because they can hide better when liquidity trading is high. When information
acquisition is endogenized, the period with the highest volume also has more
informative price and high price change volatility.

In the interday model of Foster and Visvanathan (1990), unlike the Admati
and Pfleiderer (1988) framework, private information is long-lived. A single
informed trader receives a private signal on all days, and the other market
participants get a public signal on weekdays only. The discretionary liquidity

traders can postpone their trades by no more than one day. Given the structure

103

of Foster and Visvanathan model, the largest informational asymmetry occurs on
Monday. The public signal reduces the advantage of private information. Vlfith a
highly informative public signal, there will be two days of high trading activity.
With a less informative public signal, Friday is the only day with concentrated
trading. Volume is predicted to be lowest on Monday. Moreover, when the
public signal is very precise, price change volatility will be high on Monday. In
contrast, Admati and Pfleiderer (1988) predict that volume and price change

variability move together.

5.3 Empirical Analysis
5.3.1 Introduction

Due to lack of data, the intraday price-volume relationship on the ISE has
never been examined. The analysis in this section is the first attempt to do so.
The microstructure of the ISE is likely to affect this relationship.

The ISE is not different from most other markets regarding the rules on
short selling, so the asymmetry documented in stock and bond markets is
expected to exist there as well. As discussed earlier, serial dependence in
trades and lagged price adjustment are features that Hasbrouck (1991) cites in a
dynamic volume-price relationship. One factor that leads to serial dependence in
trades, - the inventory control effect - does not exist on the ISE because there
are no designated market makers. Moreover, the ISE’s large relative tick implies
that price discreteness is more important in this market, and there are no price

smoothing rules other than the maximum price change limit. These last two

104

points are related to the issue of lagged price adjustment: The former is likely to
delay it and the latter to enhance it. Finally, the opening call auction
accommodates a significant fraction of trading volume in many markets but is
absent on the ISE. Therefore, the information collected during nontrading

periods has to be acted upon during the continuous trading period.

5.3.2 Data

The Choice of Sampling Frequency. The relationship between price
and trading activity can be analyzed in either transaction or clock time. With the
former method, trade direction can be used as a control variable. Since the ISE
data do not contain an identifier for trade direction, another method is required.
One way to classify trades is to compare the trade price to the quote prices at
the time of trade, but ISE quote data are not available. In similar situations,
researchers have used the tick method: The trade direction is inferred by
comparing the price of the current and preceding trade. Trades are classified
into four categories: an uptick, a downtick, a zero-uptick, and a zero-downtick.“

To see whether the tick method is satisfactory in classifying consecutive

transactions, Table 43 gives the frequency distribution of trades using only

 

3‘ A trade is an uptick if the price is higher than in the previous trade. A zero and
a downtick is defined analogously. A zero tick is classified as a zero-uptick if the
previous trade is an uptick. A zero-downtick is defined analogously.

105

upticks and downticks.35 As discussed in chapter 3, about 10% of transactions
result in a price change. It is clear from the table that only a small fraction of
transactions in the sample can be classified with this method. Therefore, the
distinction between seller- or buyer-initiated trades is ignored in the analysis,
which is based on clock time, by sampling price and volume information
periodically.

Although the choice of sampling frequency is arbitrary, there is a tradeoff
between the total number of sampled data points and the fraction of these for
which the price changed during the sampling period. Table 44 shows
frequencies using three interval lengths, that is, splitting a trading day into
intervals of 15, 30, and 60 minutes. The 15-minute interval is used in the
analysis.

The Price and Volume Series. To construct an equally weighted price
and volume series, the following procedure was adopted. The use of nominal
stock prices would assign larger weights to higher priced stocks. Therefore, the
nominal price series are adjusted so that the price of each stock equals 100 at
' the beginning of the sample period. The final price index relies on these

individual stock indexes.36

 

3“ Lee and Ready (1991) found that the tick method is 90% accurate, but Aitken
and Frino (1996) found only 74% accuracy for the Australian Stock Exchange
(more than 90% when zero ticks were excluded).

3“ Split and dividend adjustments were done for each stock. Five stocks had old
and new shares trading simultaneously. The price of old shares was used in
creating the price series. Since the new share volume could not be ignored, the
volume series includes the trades of both kinds of shares. Refer to Chapter 2 for

106

For most stocks the number of outstanding shares changed during the
sample period, so trading activity is measured by share turnover. Due to the
large cross-sectional variation in the float rate, share turnover is defined as the
ratio of shares traded to floating shares.

There were 283 trading days in the sample period, of which 277 were
standard trading days with sixteen 15-minute intervals. On the remaining six
days, either there was trading only during the first session or there were trading
delays/halts. As a result, there are 4,500 intervals during the sample period.

Figures 6 and 7 plot the price and percentage turnover series over time.
Figure 6 shows the effect on stock prices of the Russian crisis of August 1998,
during which month average prices fell about 50%. A similar decrease in volume
can be observed from Figure 7. The popular press emphasized the pressure on
prices exerted by the sell orders of foreign investors.

Stationarity. To decide on the specification to model the relationship
between price and volume, the stationarity of these two series needs to be
examined. If the two are integrated on the order one, then it is possible that they
are cointegrated. In this case, the short— and long-term relationship between the
two series can be distinguished, and estimation methods that preserve the
information about both forms of covariation should be employed. Table 45 shows
formal stationarity tests for price, return, and percentage turnover. Both the

augmented Dickey-Fuller and Phillips-Perron tests give consistent results. The

 

the definition of old and new shares.

107

hypothesis that the volume series contains at least one unit root is rejected at the
0.001 level of significance, so it is concluded that this series is stationary.
Because neither test indicates rejection for the price series, differencing is
appropriate in that case.

The return series is the difference of logarithmic prices at the end of
consecutive intervals. Close-to-open returns (both overnight and midday) are
excluded to eliminate any possible confounding effects from information that
arrives when the market is closed. Figures 8 and 9 graph the return and return
squared over time. As shown in Table 45, the unit root test is rejected for the
return series at the 0.001 level of significance, which suggests it is stationary.

lntraday and interday variation in stock returns and trading volume has
been shown in other markets. Since this is the first study to employ intraday
transaction data from the ISE, a univariate analysis of systematic intraday
patterns in these two variables is presented before investigating the bivariate
relationship. To eliminate measurement error, the six days with fewer than 16
intervals were excluded, which leaves 277 trading days and 4,432 intervals in the

final sample.

5.3.3 Time of Day and Day of Week Effects

lntraday Trading Volume. Table 46 and Figure 10 show average
turnover for each interval and each weekday. The overall average turnover
during a 15-minute interval is 0.467%. For each day of the week, turnover

attains maximum value during the first interval and it is about twice the amount

108

observed in the remaining intervals. Turnover is also high during the first and
last intervals of the second trading session. High trading activity during the first
interval of both sessions can be attributed to the effect of information flow during
the nontrading periods, whereas the increase during the last interval of the day
probably reflects the concern of traders to rebalance their holdings before the
market closes.

Several analysis of variance tests were performed to measure the
variability of mean turnover across intervals and days. Table 46 shows these F
tests. F,,,t tests the hypothesis of equality of mean turnover during all intervals in
a given weekday. Ffrst (anm, aneenm) tests the hypothesis that mean turnover in
interval 10:00-10:15 (14:00-14:15, 15:45-16:00) is not different from the mean
turnover in the remaining intervals (excluding those three intervals). Fday tests
the hypothesis that there is no interday difference in mean turnover during a
given interval. Overall, the results suggest weak interday but strong intraday
variation in turnover.

lntraday Volatility. Table 47 and Figure 11 show average return squared
' for each interval and each weekday. This measure is a proxy for return volatility
during an interval. The volatility of return increases during the first interval of
each session, but it is much higher during the first interval of the day. The
definition of F,,,,, Fm, F,,,,,,,,, F,,,,e,,,,,,,, and Fday in Table 47 are analogous to the F
tests in Table 46. Combined with the turnover pattern, the variation in volatility

suggests that the incorporation of new information into prices occurs during the

109

 

first interval of both trading sessions, and the high turnover at the end of the day
is due to portfolio rebalancing rather than the effect of information.

To complement the picture, Table 48 and Figure 12 show the time-of-day
and day-of-week effects for average return. Similar to the behavior of the other
two series, there seems more intraday than interday variation in average returns.
A large positive return during the last interval of the second session is the most
striking pattern. This may be caused by buyer-initiated trades to close short
positions by the end of the trading day. Positive returns on Friday afternoon, and
negative returns on Monday suggest that investors prefer to take long positions
during the weekend and liquidate their holdings on the first day of the week.

The univariate analysis so far shows systematic temporal patterns in
return, volatility, and turnover. The time-of-day rather than day-of-week effect
seems to be the dominant source. To remove seasonality, the three series were
standardized using time-of-day and day-of-week means and standard deviations.
The analysis in the remainder of this chapter uses the standardized return,
volatility, and volume series.

The link between price and volume will be examined, first, by investigating
any contemporaneous association and, second, by testing for a causal

relationship in'the Granger sense.

5.3.4 Contemporaneous Price-Volume Relationship
To examine the contemporaneous relationship between price change and

turnover, the Jain and Joh (1988) empirical model is used. Three specifications

110

of this model are estimated:

20 19
volumet = a + b|return,| + chegt |return,| + ZeiDit + ZfiDit |return,_,| +

i=1 l=1

19
Zg,(D,,|return,_,|Dneg,) + u, ; (10)

i=1

D = a dummy variable that takes on the value of one when return is
negafive;

D, = i=1-19, that is, four day-of-week and 15 time-of-day dummies. For

example, D1 equals one when the day of the observation is Tuesday,

zero othenivise; D5 equals one when the observation belongs to the

10:15-10:30 interval, zero otherwise; and

a dummy variable to incorporate the effect of Russian crises. Its value

is zero before August 1, 1998, and is one othenivise.

.9
II

The model allows the contemporaneous relationship to depend on the
sign of return. As discussed in the introduction, the theoretical models of Epps
(1975) and Karpoff (1986, 1987) predict an asymmetric relationship: Volume on
price upticks will be larger than volume on downticks.

The first specification imposes the restriction that all the coefficients
except a, b, and c are equal to zero. The second specification sets all fi and gi
equal to zero. The last specification estimates the unrestricted model.

Table 49 reports the estimated coefficients and their p values for the three
specifications. To account for heteroskedasticity and autocorrelation in

disturbance terms, the current model and the following two models used to

111

analyze causality are estimated with the Newey and West (1987) approach.37

The first specification shows a strong contemporaneous relationship
between absolute value of return and volume. Moreover, the prediction of an
asymmetric relation is also supported, although the extent of the asymmetry is
small. Positive return has a coefficient of 0.491, compared to 0.429 for negative
return.

The second specification includes time-of-day and day-of-week dummies.
None of these are significant, which suggests that the standardization removed
systematic intraday differences. The dummy denoting the down market after the
Russian crises is highly significant and shows that volume decreased following
the shock. The strong relation between volume and absolute vaer of return and
the small asymmetry displayed in the first specification are robust to the inclusion
of the above-mentioned control variables.

The third specification permits the relationship between volume and
absolute value of return, as well as asymmetry to depend on the time-of-day and
the day-of-week. In this specification, the coefficients b and c show the relation
for the first interval on Monday. As in the second specification, all day-of-week

dummies are insignificant. All time-of-day dummies that are significant have

 

3’ Newey-WeLst estimator of the covariance matrix of the least squares estimator

is. S= S 0+—z ZwIeIe eI_I[xIxI_I+xI_IxI ] where S o=—ZeIxIxI and w I=l—-L—+-jl, e,
T|= I i=l+l Tl= I

is the least squares residual, and autocorrelations greater than L are small
enough to ignore.

112

negative coefficients. This probably reflects the fact that the first interval of the
day has the highest volume, and standardization could not completely eliminate
time-of-day differences in turnover. The relation between volume and return for
the first interval on Monday is qualitatively similar to the results in the other two
specifications, except that asymmetry is much greater in this interval. The
general conclusion is that both the positive contemporaneous relation between
volume and absolute return and asymmetry regarding the sign of the return exist
at different levels in all intervals on all days. The few exceptions are that
asymmetry disappears on Tuesday and its direction changes during the interval

14:45-15:00.

5.3.5 Causality

For the two jointly covariance stationary time series x and y, x is said to
Granger cause y if knowledge of past x and y leads to better predictions of y
than would result from knowledge of past y alone. Several Granger causality
tests have been proposed in the literature.38 Bivariate vector autoregression, as
suggested by Granger (1969), is used in this study. Two specifications are

estimated.

 

3" See survey articles by Pierce and Haugh (1977) and Geweke et al. (1983).

113

Specification 1:

4 4
returnI =a+ZbIreturnI.I +ZcIvolumeI_I +yI; (11)

i=l i=1

4 4
volumeI =d-l-ZeIretumI.I +ZfIvolumeI,I +77. .

i=l i=l
This specification uses signed return and presumes that two returns of the
same magnitude but oppoSite signs have a different effect on future return and
volume. Since there is no reason to impose this restriction, a second
specification that uses absolute return is also estimated. This system can be

interpreted as testing the causality between return volatility and volume.

Specification 2:

4 4 4
IreturnI |=a+ZbI lreturnI,I l-l-ZCI volumeI_I +EddIDnegI,I +IuI ; (12)

4 4 4
volumeI = e + ZfI|returnII| + ZgIvolumeI_I + ZhIDnegI, + 77, ;
i=1

i=1 i=1
where:
Dnegk is a dummy that equals one if return is negative in interval k, zero
otherwise.

Before the results are analyzed, a few points deserve clarification. First,
the results of causality tests can be extremely sensitive to the choice of lag
length. Although the two specifications use p=4, the estimation was repeated for

p=1-5 to confirm robustness, and the results are qualitatively the same as those

reported. Second, interpretation of size, sign, and significance of individual

114

lagged terms in a vector autoregression is often meaningless.39 Inferences
using their combined effects are valid, however, and this is sufficient for the
purposes of this study, which is testing for causal relationships. Third, only the
lagged observations contained in the same trading session as observation t are
considered. The reason is to isolate the effect of information arrival during the
preceding nontrading period. Therefore, the maximum lag length that can be
used in the analysis is seven.

Table 50 reports the regression results for the first specification. The
coefficients suggest that both return and volume are positively autocorrelated.
The Wald statistics displayed in the last row of the table suggest that return
causes volume in the Granger sense. Moreover, there is no feedback, since the
coefficients of lagged volume terms as a group are not significant in the return
equaflon.

Table 51 displays the regression results for the second specification.
Both return volatility and volume are positively autocorrelated. The negative
coefficients of dummies in the volatility equation show that positive
autocorrelation in volatility is stronger when past volatility arises from an increase
in price. The significant Wald tests for both equations suggest that there is a
feedback relationship between volume and return volatility. Past return volatility
has a negative effect on future volume. If return volatility arises from a decrease

in price, then future volume falls farther. Past volume has a negative effect on

 

3“ See the discussion in Sims (1980).

115

future return volatility. In other words, an interval with high trading activity is

followed by a period in which price stays relatively stable, ceteris paribus.

5.4 Summary

This chapter examined intraday trading on the ISE. The analysis of
return, return volatility, and volume data indicates a strong time-of-day and weak
day-of-week effect for all three variables in this market.

The evidence shows a strong positive contemporaneous relationship
between the absolute value of price change and volume. As in equity markets
elsewhere, the relation is asymmetric on the Turkish exchange. Positive price
change has a larger influence on volume than negative price change. Moreover,
asymmetry shows variability within a trading day.

Two causality tests were performed. The first found a unidirectional
relationship between volume and return: Past return causes volume in the
Granger sense. The second revealed a feedback relationship between volume
and volatility: Past values of both variables help explain the current value of the
other variable.

Overall, the evidence in this chapter shows that the large relative tick size
does not result in any notable difference regarding the price-volume relationship
on the ISE. Neither the univariate temporal patterns nor the asymmetric positive
relationship between volatility and volume differ qualitatively from the evidence

reported for other markets such as the NYSE and SEHK.

116

Chapter 6

Summary

Broadly speaking, market microstructure studies investigate the effect of
trading mechanisms on the formation of prices. Most of the empirical research
examines developed markets, mainly because of data availability. During the
last decade, probably due to automation of the trading process, high frequency
data have become available for emerging markets. This is the first study to use
intraday data from one such market, the Istanbul Stock Exchange. The data set
covers the 30 most active stocks on the ISE over 14 months and contains about
nine million transactions.

The most distinguishing feature of this order-driven market is its tick rule,
a step function with 10 regimes of narrow width. North American exchanges fix
the absolute tick size, but the narrow regimes of the ISE indicate a desire to
keep the relative tick size constant. Moreover, ISE’s relative tick is up to 900%
greater than that of other exchanges.

One goal of this study was to examine whether large tick size restricts
trader behavior in this market. It was found that prices show very weak
clustering. It appears that traders use predominantly one-tick and occasionally
two-tick rounding. Possibly due to the narrow regime width, no within-regime
variation in clustering is found. In addition, the examination of spread and
consequent price change frequencies revealed that these hardly ever exceed

tick size, which indicates that the tick size is binding. This suggests that the

117

large relative tick in this market cannot be attributed to low price resolution.

This study also presents evidence about limit and market order profitability
on the ISE. In an order-driven market, liquidity is provided by public traders via
limit orders. To be an attractive strategy, limit order trading requires sufficient
liquidity trading in the market, and the deviation of price from its equilibrium value
should be corrected within a short period. A trading rule used in the literature to
analyze this aspect of the NYSE and Paris Bourse was employed here.

The analysis in this study indicates that on the ISE, average round-trip
returns are negative for both limit and market order strategies. This is not
surprising, given the short investment horizon incorporated into the trading rule.
The use of raw returns yielded inconsistent results. The choice of the period for
observing how long it takes the temporary price effect to fade seems to distort
return estimates, possibly because marketwide information events can
intervene. These are likely to affect estimated returns systematically, so
market-adjusted returns were used to measure performance of the two trading
strategies. Because the results are not sensitive to the choice of the waiting
period, they support the conjecture about marketwide information intervention.
Unlike the NYSE and the Paris Bourse, the ISE has excessive marketwide price
movements, and they hide short-term mean reversion in price.

The market-adjusted returns show that, on average, executed limit orders
perform better than market orders, but the opposite holds for unexecuted limit
orders. Therefore, patient investors who can forgo trading if their limit orders are

not executed are likely to play the role of liquidity provider on the ISE.

118

3L

Data on the fraction of executed limit orders allow comparison of the ISE
to the NYSE and Paris Bourse. On the NYSE the fraction is 46%, 39%, or 35%
depending on whether orders are 1%, 2%, or 3% from the equilibrium value at
the time of submission. The corresponding figures are 45%, 40%, and 34% for
the Paris forward market and 21%, 25%, and 19% for the Paris cash market.
For the ISE, these figures are 88%, 66%, 58%, and 52% for orders placed in the
0.5, 1.5, 2.5, and 3.5 test categories expressed in ticks (0.8%-5.5% from
equilibrium value). This comparison implies that prices are more volatile on the
ISE, which can be attributed to two factors. First, the absence of an opening
call auction may negatively affect price discovery on the ISE. Second, there may
be insufficient depth in the limit order book, which is unexpected, given the use
of a large relative tick in this market. Therefore, one can hypothesize that the
weak balance between limit and market order submission rates may be one
reason for the ISE’s choice of an unorthodox tick rule. Unfortunately, a test of
this hypothesis is not possible with the data available.

This study also presents evidence on the relationship between price
change and trading volume on the ISE. Similar to the NYSE and the Hong Kong
exchange, these two variables display strong intraday variability on the ISE. It
appears that the incorporation of new information into prices occurs during the
first fifteen minutes of both daily trading sessions, and the high turnover at the
end of the day is due to portfolio rebalancing rather than information effects.
Unlike the NYSE and Hong Kong situations, there is a weak day-of-week effect

on the ISE.

119

Consistent with previous studies of bond, equity, and futures markets,
there is a strong contemporaneous relationship between volume and price
change on the ISE. The literature documents that this relation is asymmetric in
bond and equity markets, which is attributed to the extra cost of taking a short
position. The absence of asymmetry in futures markets, where there is no cost
difference between taking a long and short position supports this hypothesis.
Asymmetry was found on the ISE. The ratio of volume to price change for
upticks is 1.52 times the absolute value of the same ratio for downticks on the
NYSE , 1.39 for the Hong Kong exchange, and 1.15 on the ISE.

The study also provides evidence of a dynamic relationship between price
change and volume. Both are positively autocorrelated. Moreover, there is a
feedback relationship between volume and volatility. This differs from Jain and
Joh (1988), who found a unidirectional relationship on the NYSE, that is, volatility
causes volume in the Granger sense.

While this dissertation has contributed to our understanding of the Turkish
stock market, with the help of increased data availability and some recent
' developments regarding Turkish stocks further work will be possible in at least
two directions. First, in order to test the conjecture that tick size is used as a
policy variable for the purpose of enhancing liquidity on the ISE, quotation and
order flow data should be used to describe the dynamics of limit order book and
the order choice decision of traders in this market.

Second, recent developments will enable to analyze the price discovery

process for the Turkish stocks. Price discovery is the process by which markets

120

incorporate information available to market participants to arrive at equilibrium
asset prices. If a single financial asset or multiple highly related financial assets
are traded on more than one market, each market may be involved in the price
discovery process. In a few months, futures and option trading will be introduced
on the ISE. Moreover, the first Turkish stock, Turkcell, was listed on the NYSE
on July 12, 2000. It is likely that the cross-listing decision of Turkcell be followed

by other large Turkish companies.

121

"L

 

APPENDICES

122

APPENDIX A

TABLES

123

Table 1

Annual ISE Trading Activity, 1986-1998

 

 

 

 

Year Traded Value # of Shares # of Contracts
(Billion TL) (Million $) (Million) (Thousand)

1986 9 13 3

1987 105 118 15

1988 149 115 32 112
1989 1,736 773 238 247
1990 15,313 5,854 1,537 766
1991 35,487 8,502 4,531 1,446
1992 56,339 8,567 10,285 1,681
1993 255,222 21,770 35,249 2,815
1994 650,864 23,203 100,062 5,085
1995 2,374,055 52,357 306,254 11,667
1996 3,031,186 37,737 390,917 12,446
1997 9,048,721 58,104 919,784 17,639
1998 18,029,967 70,396 2,242,531 21,571

Source: ISE Web Page
Table 2

Monthly ISE Trading Activity during the Sample Period

 

 

Traded Value # of Shares # of Contracts
Month Year (Billion TL) (Million $) (Million) (Thousand)
January 1998 1,194,321 5,669 89,653 1,481
February 1998 1,304,672 5,859 93,343 1,706
March 1998 1,311,703 5,605 101,264 1,729
April 1998 1,856,735 7,578 133,636 1,800
May 1998 2,218,081 8,853 178,274 2,044
June 1998 1,927,839 7,425 184,447 2,060
July 1998 1,990,915 7,453 240,022 2,098
August 1998 1,367,528 5,009 195,863 1,697
September 1998 1,364,690 4,973 253,912 1,890
October 1998 1,050,364 3,788 237,235 1,538
November 1998 1,374,580 4,692 291,077 1,883
December 1998 1,067,539 3,492 243,806 1,645
January 1999 939,282 2,944 182,409 1,202
February 1999 2,043,121 5,981 377,032 2,165

 

Source: ISE Web Page

124

Table 3

Comparison of Emerging Markets by Size, in Millions of Dollars

 

 

 

 

Market Traded Value Market Capitalization
1996 1997 1998 1996 1997 1998

Taiwan 470,193 1,297,474 884,698 273,608 287,813 260,015
China 256,008 369,574 284,766 113,755 206,366 231,322
So. Africa 27,202 44,893 58,444 241,571 232,069 170,252
Brazil 112,108 203,260 146,594 216,990 255,478 160,887
Korea 177,266 170,237 137,859 138,817 41,881 114,593
India 26,599 53,954 64,498 122,605 128,466 105,188
Mexico 43,040 52,646 33,841 106,540 156,595 91,746
Greece 8,283 21,146 46,999 24,178 34,164 79,992
Chile 8,460 7,445 4,419 65,940 72,046 51,866
Argentina 4,382 25,702 15,078 44,679 59,252 45,332
Israel 8,045 10,727 11,291 35,934 45,268 39,628
Philippines 25,519 19,783 9,992 80,649 31,361 35,314
Thailand 44,365 23,119 20,734 99,828 23,538 34,903
Turkey 36,831 59,105 68,646 30,020 61,090 33,646
Egypt 2,463 5,859 5,028 14,173 20,830 24,381
Indonesia 32,142 41,650 9,709 91,016 29,105 22,104
Russia 2,958 16,362 6,805 37,230 128,207 20,598
Poland 5,538 7,977 8,921 8,390 12,135 20,461
Morocco 432 1,048 1,385 8,705 12,177 15,676
Hungary 1,641 7,684 16,135 5,273 14,975 14,028
Colombia 1,360 1,894 1,539 17,137 19,529 13,357
Czech Rep. 8,431 7,055 4,741 18,077 12,786 12,045
Peru 3,805 4,033 2,776 12,291 17,586 11,645
Venezuela 1,275 3,858 1,510 10,055 14,581 7,587
Jordan 297 501 653 4,551 5,446 5,838
Pakistan 6,054 11,476 9,102 10,639 10,966 5,418
Sri Lanka 134 311 281 1,848 2,096 1,705

 

Source: International Finance Corporation, Emerging stock markets factbook

1999.

Note: Markets are sorted in descending order of market capitalization

at the end of 1998.

125

“i.

Table 4
Number of Stocks Listed on the ISE

 

 

Year National Regional New Watch-List

Market Market Companies Companies
Market Market

1986 80

1987 82

1988 79

1989 76

1990 1 10

1991 134

1992 145

1993 160

1994 176

1995 193 12

1996 213 1 1 1 3

1997 244 7 2 5

1998 262 7 1 7

 

Source: ISE Web Page

 

 

Table 5
Types of Normal Order
Type Trader Unfilled Maximum
Specifies Portion Trade Size
1 Price, Quantity Waits Expressed in lots
2 Price, Quantity Canceled Expressed in value
3 Price Canceled Expressed in value
4 Price, Maximum trade value Canceled Expressed in value

 

Source: ISE Publications, Capital Market and Securities Exchange Guide 1998

126

Table 6
Minimum Trade Size for Special Orders

 

 

Base Price Minimum Trade Size“
0’ L)
0 - 25,000 10
25,500 - 50,000 6
51,000 - 100,000 3
102,500 and above 1

 

Source: ISE Publications, Capital Market and Securities Exchange Guide 1998
Note: The size of an order expressed in lots should exceed a minimum level to
be considered a special order. This lower limit depends both on the price level
of a particular stock and its trading activity during the previous month.

a As a multiple of “maximum number of shares” limit.

Table 7
Number of ISE Members Authorized to Trade on the Stock Market,
1986-July1999

 

 

Year Individual Brokerage Commercial Investment & Total
Broker- Dealers Houses Banks Development
Banks
1 986 8 1 1 25 3 47
1 987 15 16 39 3 73
1988 18 18 40 4 80
1 989 22 20 44 8 94
1 990 17 48 43 8 1 16
1991 --- 1 10 46 9 165
1 992 --- 1 12 51 9 172
1993 -- 1 12 53 1 1 176
1994 --- 11 1 53 1 1 175
1995 --- 103 50 12 165
1996 --- 100 50 12 162
1997 --- 140 2 0 142
1998 -- 140 0 0 140
1999/7 -- 137 0 0 137

 

Source: ISE Web Page

Note: Individual broker—dealers assumed corporate status in 1991 and now
operate as brokerage houses. According to an arrangement by the Capital
Markets Board on August 15, 1996, banks that intend to operate in the equities
market must transfer their stock market operations to the brokerage firms they
control.

127

.bmucooom 550 mm 5.6% m. 9:29 9.69 3.59:9 ..< .bmucooom men.
8.. 358mb 9m QmEo .309.» 9.3.3:. 865 new 9:9: c.8202; 9: :. 80:889.. 8:268:99 $me 3955 86:9. meE .902
$9. no>> mm. ”oesom

 

 

128

 

 

80838.8 83888.8 838.48 388.8 .98 890
o: 8.8 8.. 832.88 338.88 8859.9 83v .8888 om
8.0 9.8 8.? 54.838 88.88.08 - «3.8 SE? 3:80 9
8... 8.? >3 905858 888.838 .3958 83.9 8x S
8.8 8. 3. 9: 58.98.28 48.838 8984.8.” V8.48 $522.“. 5:38, t
38 9.8 a: 8m.~E.m8 945358 825 EH8 5.59, m5 9
own 8.8 v: 8, .988 8868.88 835R 88.84..” SE? 8e 9
8... 8.8 93 88.08.88 «8.88.80 88.83 84.83 53.9 2 3
cod 8.3. 8., 8888.88 388.88 - O88 8.9m 2
o: 8.8 43 «8.88.02 8.88.88 838.9 88.2 852?. N.
8... 8. to. Be 838...: 88.94.82 288.8 8.8 58> m2. 3
8... :8 8a 8058. F 8 88.838 25.88 8 r .89 6.59 .86. a? E
8.8 8.8 28 3~8Qm8 8.88.88 8.08.8 E8 .298 m
Re 8.8 8N 8:9 to. a 8.88. 8 832. S 88.9 92 m
86 8. R Ea 288.85 388.85 839 85.8. .8me N
93 5.2 8a 888... a 8.8....va 8.43...” 818 59.8 o
8... 2.9 8N 80.83 5; 88.833 833.8 8?. 5 £59 9.8.“. m
cod 8.9 8N 08.88.83 88.48.83 - 233 828» v
8.0 S. : 88 88.88.89; 88.84.83 «8.88 84.49 55.9 8.5 m
88 m: 88 «888.83 83 8.83 88.85 :2... 53.8, :58 N
mmlsi 8... 8.4 «5.88. :8; CONEEE 832.8 83: .820 F
Emucooom .92. .9o._. .98 Emocooow bmncooom .cmEtn. Ex“. .mnEos. xcmm
9&9. 3.6.:an .050 9:3
98.... 83:3).
939... .5952
585.. fi =o_____2 I: I.

 

mam? .mnEoooQémgg 699980 o:_m> cozommcmfi .90... .3 notow 882.. 390me mm. om 3..
m 2an

60000000 0050 00 0.5000 0. 0E0.0> 00.00.. 050.900. _.< 50000000 000. 00 00580.0
00 000.0 0.0000 003.02.. 0005 0:0 .909: 0.000.052, 05 0. 000000000; 000000000: .9000. .0955 090000 b0Etau0uoz

000a. 00>> mm. ”00500

 

 

129

 

 

000.000.30.00 03.000.02.00 0000.063 30.00000 .90.. 0:05
mud 00.00 00.0 000.000. F .0 500000.000 000.0050 09.00» E_L..0> .006. 30> om
3.0 :00 :0 000. 00er «00.00103 0.300010 000.000 00.00000 0?
05.0 00.00 05.0 00.03.00 09.08.30 2.0. 50.0 00.0 E.._.0> 9E< 00.4. 0v
R... 0000 00.0 00060000.. 000.000. «00 00 F. 00.0 R r .00 E000.» .0__oc.v.m t
00.0 3.00 «00 00.00003; 00.30000; 000.00; .250 E000> ._E0o 0F
:0 00.00 00.0 000.000.00 000.0 5. F0 5010.0.“ 000.000 0:000 0?
~09 05. 0 00.0 000000.000 00000.90 000.000N 09.0000 E_._.0> 0. 3
voe 00.00 00. F 0 3.3 v.05 001090.000 000.0000 000; ...9000 0?
00.0 00.00 3.. 00.0? v.00 00.000. .00 000.09.: 000.0? 9.< 0..
_.0._. 0v.00 0.1 000005.000 30.30000 000.035 00 r .0v 600.0003 3
V04 V080 05F 000000.02 00.000600 000.000.? 0.0.9 5059.4. 0?
00.0 00.00 5.0 03.90100 000.0300 000.0F~.0~ 000.? :93. 005.2 0
00.0 00.00 00.0 000.000. .0? 30.00009 000.0900 30.: 0:5 0
:4 «0.00 00.0 09.30000 20.000000 00003.3 000.0010 E....0> 00... x.
00.? N P .2. 00.0 000.00. 0v 30000.00... 03. 80.00 30.000 00.00 0
00.0 00.05 00.0 000.0000 .0; 000000.03 000050.00 00 F. 0 E...=0> 0:00.“. 0
006 00.00 00.5 0.0.000. :0; Codenfinvé 00:05.00 0052.. .0020 v
0v.~ 00.00 00.0? 03.000000 00.000000 30.00.00 53.0.. 000.06 0
v: .L. .0 00.3 «00.39.. F 3 000.000. 00 00.00003 3.0.000 x< m
006 00.00 00.00 30.00.00 30000.50 000000.08 50600 00.000050 E000> F
.90... 00000000 B02800 .90 h b03800 b03800 308:0 E... 09:05. 0.00”.
00.0.. 00.0.. .050 000..
02.0.0an
000.... .3102
0.00; 50:03.
0:00.00 d. 00.....2

 

000w .00E000o-.c0:c0... 0000000000... 0.0000 000 0.000.055 .3 09.00 00000... 000000.05 mm. 0N nob
0 0.00...

0000 00.5 00. ”00500

 

 

 

 

 

00 3 .00 003.03 30.00 30 3 .00 000.00 300.03 000. 30 000..
000.00 003.30 000. ..0 000.00 003.00 300.00 033. 30 000.00 000.30 000. 30 000.00 30.00 000..
000. 30 030.30 03.00 000.00 300.33 300.03 030.03 000.00 000.00 300.33 300.33 000.03 0003
000.00 030. 30 003.00 030.00 030.00 300.00 000.00 000.00 303.00 003. 30 000.00 330.00 0003
000.00 000.00 000.00 303.00 030.00 300.30 003.00 000. 30 0 ..0. 30 00 .. .00 000. 30 000.03 0003
000. 30 03 3.00 50.00 030. 30 000.00 030.0_. 000.0.. 000. 3 303. 3 000.03 30.00 303. ..0 300..
300.00 030.30 000.00 003.00 030.30 300.00 000.00 000.03 033.03 000.03 000.3 300.03 0003
000.0 333.0 300.0 300.03 000.3 030.3 000.3 000.0 33.0.. 000.03 000.3 030.03 000..
300.03 030.3 000.03 000. 3 030.03 000.03 000.3 03.0.. 003.03 300.03 03.00 300. 30 300..
000.03 030.03 30.00 0 30.00 003.00 000.00 003.03 000.03 000.0 000.0 000.03 30.0.. 000..
000.0 003.3 300.3 300.3 003.0 30 3 .0 000.0 000.3 003... 000.3 300. 3 003... 0003
003.3 000.3 000.3 300... 000.3 000.3 000.3 000.0 003.0 300.0 000.0 000.0 000..
00 3 .0 n 3.3 033.0 000.3 003.0 000.3 000.0 0 30.3 000. 3 000. 3 000. 3 00 3.3 0003
000 000 300 330 030 000 000 _.00 3 30 000 000 000 000 3
.000. .32 000 .000 .0:< .02. 000.. >05. 00< 00.2 .00”. .00..
20:00 .0 mco____s_ 0. 002-002 .00. 05 S 88: 00.08500 .0 30> .9005. .98
3 0.00
.. 000a 00>> 00. ”00500

000.0 000.0 000.0 30.0 300.0 003.0 003.0 0003

000.0 000.0 000.0 000.0 030.0 000.0 000.0 000.0 300.0 000.0 000.0 03.0 0003

030.0 003.0 300.0 000.0 000.3 000.3 003.3 000.0 000.0 000.0 300.3 000.3 _ 0003

000.0 00.0 000.0 000.0 300.0 030.0 000.0 003.0 000.0 000.0 030.0 003.0 0003

000.3 0003

.000. $02 .000 .000 .000. .02. 000.. >05. 00.3 00.2 .00“. .00..

 

20.8 .o 25.0.2 0. 002-002 .2282. 00.050 .3 0050.0: 26.000 .0500

or 0.00...

130

Table 12
Some Characteristics of ISE Stocks in the Sample

 

 

Traded First Current Current Market Float
Since Price Price Price Capitalization

Stock Industry (TL) (TL) (Cents) (Million $) (°/o)
Akbnk Banking 7/26/90 13,000 10,100 2.80 3,496 15
Akcns Cement 10/6/87 1,170 7,800 2.16 411 15
Akgrt Insurance 12/5/94 10,100 17,500 4.85 273 29
Alark Conglomerates 5/24/89 1 ,900 5,200 1.44 193 23
Alctl Telecom 3/22/88 6,400 22,250 6.16 123 33
Arclk Consumer Durables 1/21/86 1,900 9,700 2.69 544 19
Bagfs Fertilizers 1/28/86 2,010 15,000 4.15 83 59
Cukel Utilities 1/7/86 3,220 455,000 126.02 630 18
Dohol Conglomerates 6/21/93 5,800 3,150 0.87 363 34
Dyhol Conglomerates 8/6/98 4,000 2,275 0.63 274 15
Efes Conglomerates 2/19/98 36,500 4,800 1.33 168 46
Enka Conglomerates 1/24/86 1 .200 57,000 15.79 373 15
Eregl Iron 8. Steel 1/13/86 800 4250 1.18 522 41
Garan Banking 6/6/90 4,500 14,500 4.02 2,007 20
Hurgz Media 2/25/92 7,900 6000 1.66 320 18
lhlas Conglomerates 3/17/94 45,000 23,250 6.44 187 25
lsctr Banking 11/16/87 325 14,250 3.95 5,007 33
Kchol Conglomerates 1/10/86 2,850 41,000 11.36 1,819 13
Migrs Retail 2/27/91 8,800 470,000 130.17 1,327 48
Nthol Conglomerates 10/5/89 3,300 1,850 0.51 51 55
Otosn Automotive 1/13/86 950 152,500 42.24 588 15
Petkm Petroleum Products 7/9/90 2,400 232,500 64.39 1,931 4
Ptofs Petroleum Products 5/30/91 4,000 69,000 19.11 1,337 7
Sahol Conglomerates 7/8/97 6,200 8,600 2.38 2,977 12
Thyao Airlines & Services 12/20/90 2,700 19,500 5.4 1,620 2
Toaso Automotive 7/1/91 20,000 3,700 1.02 181 22
Tuprs Petroleum Products 5/30/91 1,800 23500 6.51 4,865 4
Uzel Automotive 8/5/97 20,000 32,000 8.86 204 15
Vestl Consumer Durables 6/27/90 15,000 34,000 9.42 377 31
kank Banking 5/28/87 1,200 6,700 1.86 2,473 39
Mean 7,831 58,896 16.31 1,157 24
Median 7/3/90 3,650 14,750 4.09 467 20

 

Source: ISE Publications, Capital, Dividend, And Monthly Price Data 1986-1997

Note: Column 1 reports the date each stock became listed. Column 2 shows the first
trading price. Columns 3 and 4 give the share price on 3/12/99 in TL and dollars,
respectively. Columns 5 shows the market value of each firm in dollars. Column 6
shows the percentage of shares kept in custody by the ISE Settlement Bank. All figures
other than those in column 2 are as of the close of the second trading session on
3112/99.

131

Table 13

Weekdays without Trading during the Sample Period

 

 

Date Weekday Reason
1/1/98 Thu New Year
1/2/98 Fri New Year
1/29/98 Thu Religious Holiday (Ramadan)
1/30/98 Fri Religious Holiday (Ramadan)
4/6/98 Mon Religious Holiday
4/7/98 Tue Religious Holiday
4/8/98 Wed Religious Holiday
4/9/98 Thu Religious Holiday
4/10/98 Fri Religious Holiday
4/23/98 Thu National Holiday
1 5/19/98 Tue National Holiday
10/29/98 Thu National Holiday
10/30/98 Fri National Holiday
1/1/99 Fri New Year
1/18/99 Mon Religious Holiday (Ramadan)
1/19/99 Tue Religious Holiday (Ramadan)
1/20/99 Wed Religious Holiday (Ramadan)
1/21/99 Thu Religious Holiday (Ramadan)
1/22/99 Fri Religious Holiday (Ramadan)

 

132

Table 14

Number of Sessions Stocks were Traded during the Sample Period

 

 

Stock # of Sessions Date No Tradirg in

Akbnk 564

Akcns 562 08/06/1998 Both sessions

Akgrt 564

Alark 563 02/23/1998 Session 1

Alctl 564

Arclk 564

Bagfs 564

Cukel 564

Dohol 564

Dyhol 277 a 01/05/1998 - 08/05/1998 Both sessions

Efes 503 a 01/05/1998 - 02/18/1998 Both sessions

Enka 564

Eregl 562 12/22/1 998 Both sessions

Garan 563 12/24/1998 Session 2

Hurgz 564

lhlas 564

lsctr 553 b 05/04/1998 - 05/08/1998 Both sessions

Kchol 563 1 1/16/1998 Session 1

Migrs 564

Nthol 564

Otosn 564

Petkm 564

Ptofs 556 C 06/29/1998 - 07/01/1998 Both sessions
06/15/1998 Session 1
12/24/1998 Session 2

Sahol 564

Thyao 564

Toaso 564

Tuprs 564

Uzel 564

Vestl 564

kank 564

 

Note: Column 2 indicates the days on which there was no trading in at least
one session for a particular stock. Column 3 reveals the session in which there
was no trading.
a Became listed during the sample period.
b The state sold its 12.3% stake by public offering.
° The state reduced its ownership to 49% by selling 51% of the equity

through a block transaction.

133

Table 15
Monthly Volume Generated by Foreign Investors during 1998 among Stocks in the

 

 

Sample
Stock Average Average Median Maximum Minimum
($) (%) (%) (%) (%)

Akbnk 62,875,461 6.94 6.67 14.77 3.20
Akcns 10,097,018 1.00 0.88 2.70 0.22
Akgrt 17,970,050 1.98 1.95 4.08 0.93
Alark 4,870,535 0.45 0.36 1.18 0.06
Alctl 9,822,791 0.88 0.85 1.97 0.15
Arclk 22,082,140 2.06 1.76 4.19 0.59
Bagfs 5,098,539 0.45 0.27 2.03 0.04
Cukel 17,162,671 1.70 1.52 4.39 0.33
Dohol 34,134,138 3.53 3.53 5.11 2.66
Dyhol 9,022,976 1.23 1.01 2.16 0.36
Efes 16,116,866 1.48 1.34 2.30 0.64
Enka 7,306,394 0.72 0.67 1.28 0.34
Eregl 47,712,272 4.47 4.40 7.37 1.81
Otosn 7,953,310 0.88 0.85 1.87 0.40
Garan 56,572,491 5.89 5.45 9.41 3.57
Hurgz 9,869,593 1.01 0.98 1.76 0.33
lhlas 8,597,380 0.67 0.23 2.40 0.04
lsctr 81,947,593 9.14 9.55 15.83 1.45
Kchol 45,316,097 4.33 4.28 7.30 1.61
Migrs 35,379,111 3.98 3.92 8.02 1.17
Nthol 4,057,242 0.34 0.21 1.35 0.05
Petkm 4,799,491 0.46 0.30 1.38 0.04
Ptofs 13,842,697 1.35 1.00 3.14 0.37
Sahol 68,612,065 7.46 7.17 10.88 4.87
Toaso 22,186,943 2.11 2.28 4.58 0.30
Tuprs 37,177,303 3.67 3.64 6.60 1.99
Thyao 3,693,065 0.32 0.25 1.08 0.05
Uzel 4,932,315 0.45 0.32 1.70 0.04
Vestl 23,663,697 2.14 1.76 5.27 0.82
kank 89,445,803 10.55 8.13 19.65 5.03
Total 782,318,045 81.65 79.88 87.82 73.37

 

Source: ISE Web Page

Note: Foreign volume is defined as the sum of foreign purchases and sales. Columns
2-5 report statistics about the share of each firm in total foreign volume over this 12
month period. The last row shows aggregate figures for the sample firms.

134

Table 16
Stock Split Adjustments for Sample Firms during the Sample Period

 

 

 

 

TL
Stock Rights Bonus Split Rights Dividend PreSplit PreSplit PostSplit
Issue Issue Date Issue to be WAPa Adj. WAPa Tick
Price Paid
Akbnk 90.00 60.00 5/8/98 1,000 1,000 20,829.50 9,291.80 100
Akcns 0.00 400.00 4/27/98 38,683.64 7,736.73 100
Alark 149.15 1096.00 2/8/99 1,000 49,165.83 3,765.93 50
Arclk 0.00 100.00 5/29/98 300 21,835.05 11,067.53 250
Bagfs 0.00 900.00 4/21/98 6,000 283,314.31 33,731.43 500
Dohol 0.00 250.00 10/7/98 9,188.38 2,625.25 50
Efes 0.00 600.00 8/3/98 57,336.69 8,190.96 100
Enka 0.00 125.00 12/30/98 69,850.41 31,044.63 500
Eregl 200.00 400.00 2/24/99 3,000 18,318.19 3,474.03 50
Garan 0.00 25.00 11/12/98 5,315.06 4,252.05 50
Hurgz 0.00 250.00 2/25/98 4,200 48,691.21 31,579.21 500
Hurgz 0.00 50.00 8/27/98 8,910.04 5,940.03 100
lsctr 37.62 112.87 5/15/98 1,000 400.8 31,332.70 12,899.48 250
Migrs 7.94 0.00 8/18/98 260,000 241,691.06 243,037.31 2,500
Nthol 94.00 10.00 6/22/98 1,000 5,579.52 3,195.84 50
Sahol 80.00 70.00 8/24/98 1,000 14,680.23 6,192.09 100
Thyao 100.00 100.00 1/23/98 1,000 49,315.89 16,771.96 250
Toaso 0.00 40.00 12/14/98 3,924.82 2,803.44 50
Tuprs 75.00 75.00 11/9/98 1,000 34,949.30 14,279.72 250
kank 40.00 39.00 5/15/98 1,000 10,515.09 5,874.35 100
kank 0.00 24.00 11/9/98 3,986.13 3,214.62 50

 

Source: Bayindir Menkul Degerler AS, a brokerage house.
Note: The first two columns the amount of rights and bonus issues as a percentage of the number
of outstanding shares before the split. Column 4 gives the price per share to be paid to participate
in the rights issue. Column 5 shows the next dividend to be paid for stocks that exist before the

stock split. Presplit base price is adjusted for stock splits. Column 8 gives the tick size to be used

during the first session following the split.

' Weighted average price

135

Table 17
Dividend Adjustments for Sample Firms during the Sample Period

 

 

Stock Total Stock Exdiv. Dividend New Old Adj. Exdiv.
Cash Dividend' Date per Sharesb WAP° WAPc WAPc Tickcl
Dividend Share
(Million TL) (%) (TL) (%) (TL) (TL) (TL)

Akbnk 50,000,000 5/20/98 1,000 9,300 8,300 100
Akcns 4,185,673 4/1/98 1,100 32,486 31,386 500
Akgrt 1,648,125 5/21/98 293 19,016 18,723 250
Alark 249,043 5/29/98 250 69,1 13 68,863 1,000
Alctl 1 ,000,000 5/29/98 500 33,527 33,027 500
Arclk 3,037,500 5/29/98 300 stock split on the same day

Bagfs 1,200,000 5/25/98 6,000 90 31,306 24,602 24,672 250
Cukel 11,782,315 3/4/98 23,565 608,710 585,146 10,000
Dohol 2,375,100 9/3/98 200 8,836 8,636 100
Enka 525,000 5/28/98 500 135,263 134,763 2,500
Hurgz 14,217,840 300 3/31/98 4,200 75 26,368 9,683 100
lsctr 20,306,131 5/25/98 401 60 12,035 11,750 11,704 250
Kchol 3,205,1 10 5/25/98 200 57,851 57,651 1,000
Migrs 945,000 5/29/98 1 ,000 239,585 238,585 2,500
Otosn 15,317,500 4/20/98 1 1 ,000 194,730 183,730 2,500
Petkm 29,550,000 5/1/98 9,850 171,296 161,446 2,500
Ptofs 18,798,560 5/1/98 2,686 72,401 69,716 1,000
Sahol 6,500,000 5/21/98 130 16,707 16,577 250
Uzel 7,728,000 4/1/98 3,360 33,874 30,514 500
kank 24,026,587 5/15/98 400 stock split on the same day

 

Source: Bayindir Menkul Degerler AS, a brokerage house.
a As a percentage of outstanding old shares.
° The percentage of total outstanding shares that are not entitled to receive the current

dividend payment.

° Weighted average price.
dTick size used during the first session on the exdividend day.

136

Table 18

Panel A: Entire Sample

Number of Transactions for Sample Stocks

 

 

 

 

Odd Lot Round Lot
Stock All Old New Old New
Akbnk 402,768 2,155 400,613
Akcns 222,929 584 222,345
Akgrt 87,988 205 87,783
Alark 124,742 186 709 110,153 13,694
Alctl 203,625 609 203,016 ’
Arclk 181,146 854 180,292
Bagfs 224,314 277 24 204,051 19,962
Cukel 218,186 2,828 215,358
Dohol 767,309 2,548 764,761
Dyhol 273,484 15 273,469
Efes 268,107 86 268,021
Enka 136,523 2,410 134,113
Eregl 228,869 3,005 1 221,519 4,344
Garan 301,731 1,816 299,915
Hurgz 264,737 947 132 259,903 3,755
lhlas 371,123 1,262 369,861
lsctr 630,345 5,513 120 619,827 4,885
Kchol 298,753 1,634 297,119
Migrs 116,241 1,137 115,104
Nthol 160,424 1,525 158,899
Otosn 115,323 265 115,058
Petkm 283,342 1,575 281,767
Ptofs 199,345 566 198,779
Sahol 468,719 1,218 467,501
Thyao 332,909 294 332,615
Toaso 259,624 729 258,895
Tuprs 686,182 2,438 683,744
Uzel 225,331 68 225,263
Vestl 138,861 600 138,261
kank 558,960 4,266 554,694
Total 8,751,940 41,615 986 8,662,699 46,640

 

137

Table 18-continued

Panel B: New Shares

 

 

Stock All Odd Lot Round Lot Period Month
Alark 14,403 709 13,694 2/8/99 - 2/26/99 Feb.
Bagfs 8,638 10 8,628 4/21/98 - 5/22/98 Apr.
Bagfs 11,348 14 11,334 May
Eregl 4,345 1 4,344 2/24/99 - 2/26/99 Feb.
Hurgz 1,001 18 983 2/25/98 - 3/30/98 Feb.
Hurgz 2,886 114 2,772 Mar.
lsctr 5,005 120 4,885 5/15/98 - 5/22/98 May

 

Note: When there is an increase in capital before dividend is paid in a given
year, newly issued shares are not entitled to receive the next dividend

payment. Until the payment of the next dividend, there is a separate market
for “old” and “new" shares.

138

.80. cow was E:E_me on. 22:; .2 .Exumd new .895. .6330 new .90.
80.: mm; mum mum: .525me m5 523 .8 2:9; cam .maE. .owmoh .655 .652 .5sz ._m9m .6200 €92 2:92. 9m 552?. “oz ”902

 

 

139

8m 8m 8m 8m 8m 8m 8m 8m 28 83 8m 8m 8m :85
8m 8m 28 8m 8m 8m 8m 8m o8 8m 8m 8m 8m .8:
83 83 83 28 83 83 28 28 8m. 28 28 28 SN 82.:
8w 2: 8m 8: 8m 8m 8m 8m 8m 8m 8m 8m 28 8a
SN 2: 2: 2: 2: 2: 2: 2: 2: 2: 2: 2: 2: :86
83 83 83 com 28 28 08 28 28 28 8m 8m 8m .89.
83 83 83 83 83 83 83 83 83 8m 8m 8m 8m 8m.
8m 28 28 28 28 88. 28 83 83 83 8m 8m 28 82.
83 83 83 83 83 83 83 83 28 28 8m 8m 8m 85:
8m 8m 8m 2: 2: 2: 2: 2: 2: 2: 2: 2: 2: saw
83 83 83 83 83 83 8m 8m 8m 8m 8m 8m - 8m
83 83 83 83 83 83 - - - - - - - .85
28 28 28 8m 8m 8m 8m 8m 8m 2: 2: 2: 2: 28m
83 83 83 83 83 83 83 83 8m. 28 8m O8 28 x62
83 . 28 83 08 8m 8m 8m 8m 08 8m 28 08 28 82
o8 8m 8m 28 8m 8m 8m 8m 8m 8m SN 8m 8m 382
83 83 83 83 83 83 83 83 83 8m 8m 8m 8m 892
8.8”. 8.8., 8.2.5 8.32 8.80 8.8m 8.22. 8.5.. 8.8.. 852 822 88.2 8.8“. 3.85
£82

 

30.. E .622ch 29:3 9: mczan 9.005 anmm .0. 5:22 con 3E3 855 :0 .mnEsz E:E_xm_2
2. 20m...

Table 20
Maximum Trade Size in Lots for Normal Orders (Type 1)

 

 

Average Trade Size in Lots Maximum Trade Size in Lots
>= 61 1,000
31 - 60 500
16 - 30 250
0 - 15 100

 

Source: ISE Publications, Capital Market and Securities Exchange Guide 1998

 

 

Table 21

Identification of Special Orders

Stock Eligible Special
Akbnk 7 3
Akcns 3 0
Akgrt 7 0
Alark 1 0
Arclk 6 1
Bagfs 671 2
Cukel 887 0
Efes 10 0
Enka 716 1
Garan 6 2
lhlas 7 4
lsctr 6 1
Kchol 1 1
Migrs 759 1
Nthol 4 0
Otosn 284 0
Petkm 2,970 0
Sahol 6 0
Toaso 1 0
Vestl 1 1 5
kank 17 2
Total 6,380 23

 

Note: Among eligible transactions, some can be identified as special orders,
so the quantities in the first column include the quantities in the second
column.

140

Table 22
The Tick Rule on the ISE and Six Other Exchanges that Use Step Functions

 

Stock Price Absolute Relative Regime

 

 

Exchange Regime Low High Tick Tick Width
(TL) (%) (%)

Istanbul
1 10 1,000 10 1.98
2 1,025 2,500 25 1.42 144
3 2,550 5,000 50 1.32 96
4 5,100 10,000 100 1.32 96
5 10,250 25,000 250 1.42 144
6 25,500 50,000 500 1.32 96
7 51,000 100,000 1,000 1.32 96
8 102,500 250,000 2,500 1.42 144
9 255,000 500,000 5,000 1.32 96
10 510,000 and up 10,000

Helsinki
1 0.01 9.99 0.01 0.20
2 10 99.9 0.10 0.18 899
3 100 999 1.00 0.18 899
4 1,000 and up 10.00

Hong Kong
1 0.001 0.249 0.001 0.80
2 0.25 0.495 0.005 1.34 98
3 0.5 1.99 0.010 0.80 298
4 2 4.975 0.025 0.72 149
5 5 29.95 0.050 0.29 499
6 30 49.9 0.100 0.25 66
7 50 99.75 0.250 0.33 100
8 100 0.500

Paris
1 0.01 4.99 0.01 0.40
2 5 99.95 0.05 0.10 1,899
3 100 499.9 0.10 0.03 400
4 500 4,999 1 .00 0.04 900
5 5,000 and up 10.00

 

141

Table 22-continued

 

Stock Price Absolute Relative Regime
Exchange Regime Low High Tick Tick Width

(TL) (%) (%)

 

Singapore
1 0.005 1 0.005 1.00
2 1.01 3 0.010 0.50 197
3 3.02 5 0.020 0.50 66
4 5.05 10 0.050 0.66 98
5 10.1 25 0.100 0.57 148
6 25.5 100 0.500 0.80 292
7 101 and up 1.000

Tokyo
1 1 999 1 0.20
2 1,000 9,990 10 0.18 899
3 10,000 99,900 100 0.18 899
4 100,000 999,000 1,000 0.18 899
5 1,000,000 and up 10,000

Toronto

1 0.005 0.495 0.005
2 0.5 4.995 0.010 0.40 899
3 5 and up 0.050

 

Sources: Bacidore (1997), Booth et al. (2000), Chan and Hwang (1998),
Hamao and Hasbrouck (1995), Hameed and Terry (1998), Harris(1996), and
ISE Publications, Capital Market and Securities Exchange Guide 1998.

Note: Columns 2 and 3 report the lowest and highest prices within each tick
regime. Column 4 gives the absolute tick size. The relative tick size is defined
as the ratio of absolute tick size to the average of lowest and highest prices for
a given regime. The last column shows the width of each tick regime
expressed by the ratio of the difference between the highest and lowest prices
to the lowest price in that regime.

142

 

 

88 888.888

8.3 8.8-8.8

82 88-8: .2”. 3.8.8 8.888.. 8888 3.828 8.8.8.8.. 888:8 .8 £88

88 8.8.88

R8 88.88

88 88:3

Bum co. Toad am 3628 cozommcm: 9.088me B omcmcoxm 3.8on 88: Pm... new nmoEmI

88 ...o 83 v5

5;. mow mmmcmams

83 at... 8:2.

88 888 8:80 8888.. mum... 388: .8 .m 52.30

88 888 comm-E

mmd 8922 comma-E

88 85.2 858-5. 8888.. wk: .88: .8 8 8.5.0

88 {ms-8.8

omé ERA-5:28 030:0 63.5.2 omcmzoxm c990“.

eve Boa matmm—o «Ox-5.2 200 c095.—

ond xooum DEODU wmcmnoxw xoouw concou-

83 3.8.8 8.88 o<om<z

88 3.8.8 8.88 xm..,.<m.m>z :8: .8 8 8.88.0

88 88.8.88 8. 3.8.8 8888.. 888m. 3.8.8 8.888.... .88: .8 .m 882

«FM {mE-.m__ov 82038 63.5.). mmcmcoxw c980”. 28: 0.230 new 2.850000

88 8.32 .82.."- 8888: 38.200 :8: .8 .8 28.8
L88 8.8 8.8.8 .98.). 8.80 888.. .88: .8 .8 :8

35¢ _m>o._
Em 8...... .82 83 8E .88.... 28.8

 

8.98.2 328.8 a. 8.8.8.0 .o .85 9:
8. 8.8.:

143

8.82.85 .mco_.oom-mmo.o cco mo_.om-oE_. mo.o..oc mowh cco moron-won... mo.o:oc wk 0
85.2828 ,6 .co.xo o... :0 Soto >mc..o.oE_. mo.ocoo ..
..omwo o5 masom :05 .0 £5.38 on. cam co£ono .83
o... .o 3.32 .0 8.2.80 .0 b___no__m>o o... 8:2 86.30... m m. co~>_m:o .83 o... .8... >EE:c on. E co_xo.a m. Eommo .o cocoa? .o omow .

 

 

womb wk momh mom-r womb mom-r womb mowh 9. wk m_m>_oc<
29.3.: 29.6 o_8_._:E 99.3.: 99.3.: 29.3.: 29.3.: 29.32. 29.6 99...» .omm<
no.8 cox... no.8 89m cox... cox: cox... cox: cox: cox... 3.0:. 22082
o: m: o: m: o: + m: m: me o... .oxfis. .o_ooo
o: - o: m: m: m: m: o... o: o: 30.0
o: + me o: m: m: o: m: o: o: :80
o: 8. - m: o: o: m: o: m: o: ouw 3.2.8 92038.

- 8.88-85

- mossntmcozao

- o cash. ccw-omoa
o: m: m: 0 me me o: .co_.o:_m> ho omom
o: o: o: o o o - - m: o: oum .EE
8. oc - m: m: m: m: 8. m... o: oE:.o>
- + o: - m: - - m: o: o: >ocoaaoi ococ.
o: m: o: + o: m... m: m: o: m: b....o.o> .9822
+ + o + + + + + + + b__=m_o> ammm<
- - o: + o... 8. o: o: - o: o~_m ocoF
+ o + + + + + + m: + _o>o.. 02.5

.88. .8 .88: .8 .88. .8 .88. .8 .88. :8: :8: .8 .88. .8

.o 588 .o 52.5 .8 88.8: .m 882 82.88 88 885 I 8.8: II .o :88 .8 :8 0.88.8

 

actofiao .o m.cmc_E..o.oo
vm oBmP

144

 

 

«.888 8888 R38 588
88.8: 88.8 28.8 820
8.88 3.8 :88 k 8.8 83 .8.2
:88: 838 88.8 28.5.
8888 88.8 88.2: 85.8 .88.
88.88 88.5 838 28.88 88.
838 8.8.8: :88: 8....
28.88 888 88.8 88.8 838 8 :88 N85:
538 88.8 :88: 88.8 88o
8888 888: 4888 88.8 .85
8888 88.: 88.8 8.8 9.8.
2:88 88.8 88. : 8 2.8 8888 88.8 8m
.8888 58.8 8888 .85
838 8888 «3.88 88.8: 88.88 .88
88.8..- 838 88.8 .88
83.8. 888 48.8 :88 2.3: 8.88
88:8 :3: 88.8 28.2: .8 3:88.
8888 83.: :88 ..o.<
838 8.8.8 .88 83 88.8 .82
88.3 88.8 88.8 :92
8888 v8.8 83 88.2: 88.8 892
88.8.- 8888 88.88 48.2: .89....

.58. 88.8 88.8 888 83 88 88 2: 8 8 8 .86

.0:

 

mm 03m..-

.8.m 8.88m .8 8.8 .0.» .o 8.3885

mcozoomcoc. __< “.4 can.

145

 

mood 0mm bow.

 

88... o8 85:
88... 8 .85
838 com 8mm
8% 8m 88
8.}: 8 .82
.98 .08 .86

 

8:85 262 @5202: 20.63%? ”m tma

 

ovmémfim mhmévr mood? www.mmv 0:..an @0955... oomdomd wvudwh... Nmmdmmé www.mvm mnoé _m~o._.

 

98.88 08.8 83: 838 08.8 v.5;
88.88 ~88 858 :85
8.8m v8.8 8 8 .8 8 8...: .85
«8.8... :38 :88 9...:
.888 83: 8.18 08.8 0.8.... 083
838 Ram 8&8 838 8...:
82.8 :38 «3.8 838 .88
88.88 «8.8 88.88 ash.

.98 .88 83 83 83 8m 8m 8 8 8 8 .85

.0:

 

contacooém £an

146

Table 26
Actual Frequencies

Part A: Tick Sizes 100, 1,000,10,000

 

 

 

 

 

Category 100 1,000 10,000 Combined
0 10.43 10.36 11.39 10.48
100 9.84 11.32 11.07 10.20
200 10.97 11.30 9.91 10.97
300 11.00 10.81 10.28 10.92
400 9.95 10.72 9.97 10.10
500 10.19 9.63 9.96 10.07
600 9.57 8.85 9.27 9.41
700 9.52 8.64 8.57 9.30
800 9.72 9.51 9.74 9.68
900 8.80 8.87 9.84 8.88
Total 1,725,748 437,110 141,979 2,304,837
772(9) 0.000 0.000 0.000 0.000
Even 50.65 50.73 50.28 50.64
Odd 49.35 49.27 49.72 49.36
772(1) 0.000 0.000 0.036 0.000
Part B: Tick Sizes 25, 250, 2,500
Category 25 250 2,500 Combined
0 26.79 27.43 26.88 27.24
25 24.73 24.55 23.71 24.45
50 25.61 25.35 25.84 25.47
75 22.87 22.68 23.57 22.84
Total 542,317 2,268,860 489,522 3,300,699
12(3) 0.000 0.000 0.000 0.000
Even 52.40 52.78 52.72 52.71
Odd 47.60 47.22 47.28 47.29
x2(1) 0.000 0.000 0.000 0.000

 

147

Part C: Tick Sizes 50, 500, 5,000

Table 26-continued

 

 

Category 50 500 5,000 Combined Expecteda
0 6.82 6.15 5.13 6.38 6.00
50 5.14 4.36 4.53 4.69 4.00
100 5.17 4.52 5.34 4.82 4.00
150 4.30 3.77 6.09 4.09 4.00
200 4.44 4.31 5.01 4.39 4.00
250 4.40 3.56 5.12 3.98 4.00
300 4.51 3.77 5.55 4.15 4.00
350 4.09 3.42 4.86 3.76 4.00
400 5.01 4.04 4.65 4.46 4.00
450 4.58 3.48 5.40 4.01 4.00
500 5.04 4.53 5.21 4.77 4.00
550 4.98 4.42 6.46 4.74 6.00
600 5.36 5.57 6.48 5.52 6.00
650 4.97 5.97 4.31 5.49 6.00
700 5.22 7.33 4.00 6.32 6.00
750 5.23 6.83 4.22 6.06 6.00
800 5.92 6.42 4.16 6.12 6.00
850 4.87 5.71 4.17 5.30 6.00
900 5.24 6.11 4.83 5.70 6.00
950 4.69 5.73 4.47 5.25 6.00
Total 1,293,952 1,717,568 133,809 3,145,329
x2(19) 0.000 0.000 0.000 0.000
Even 52.74 52.74 50.38 52.64
Odd 47.26 47.26 49.62 47.36
78(1) 0.000 0.000 0.006 0.000

 

148

Table 26-continued

Note: Tick size 10 has only 1,075 observations and is excluded from the table.
Depending on the tick regime, the frequency of the last two to five digits of
transaction prices are shown. For example, Part A uses the last three, four,
and five digits of transaction prices for regimes 100, 1,000, and 10,000,
respectively. The last column reports combined results for the three regimes.
All transactions in the sample were used to prepare the table. The total row
gives the total number of transactions in each tick regime. 772(9), 772(3), and
76(19) report the p value for the hypothesis that each final-digits category has
equal probability of occurrence in parts A, B, and C, respectively. Even and
odd rows report the frequencies of even and odd final-digit categories. 772(1)
reports the p value for the hypothesis that even and odd final digits have an
equal probability of occurrence.

3 By definition expected frequencies vary across price categories for tick
regimes 50, 500, and 5000.

149

Table 27

Frequencies Adjusted for Serial Dependence in Price

Part A: Tick Sizes 100, 1,000,10,000

 

 

 

 

 

Category 1 00 1 ,000 10,000 Combined
0 10.56 10.71 10.71 10.62
100 9.59 10.15 10.85 9.78
200 10.53 10.23 10.36 10.45
300 10.39 10.52 10.30 10.39
400 9.96 10.44 9.56 10.04
500 10.33 9.73 9.28 10.17
600 9.71 9.51 8.85 9.63
700 9.71 9.44 8.85 9.60
800 9.88 9.83 10.79 9.90
900 9.33 9.43 10.46 9.42
Total 1,725,748 437,110 141,979 2,304,837
772(9) 0.000 0.000 0.000 0.000
Even 50.65 50.73 50.27 50.64
Odd 49.35 49.27 49.73 49.36
772(1) 0.000 0.000 0.045 0.000
Part B : Tick Sizes 25, 250, 2,500
Category 25 250 2,500 Combined
0 26.90 27.24 27.84 27.25
25 24.10 23.72 23.40 23.72
50 25.51 25.54 24.89 25.47
75 23.49 23.49 23.87 23.56
Total 542,317 2,268,860 489,522 3,300,699
76(3) 0.000 0.000 0.000 0.000
Even 52.41 52.79 52.73 52.72
Odd 47.59 47.21 47.27 47.28
772(1) 0.000 0.000 0.000 0.000

 

150

sf

Part C: Tick Sizes 50, 500, 5,000

Table 27-continued

 

 

Category 50 500 5,000 Combined Expecteda
0 7.20 6.60 6.49 6.84 6.00
50 3.50 3.62 4.07 3.61 4.00
100 4.03 4.33 4.72 4.25 4.00
150 3.67 3.57 5.51 3.71 4.00
200 4.10 4.33 4.51 4.26 4.00
250 3.87 3.90 4.35 3.94 4.00
300 3.91 4.13 4.01 4.07 4.00
350 3.48 3.53 3.46 3.54 4.00
400 4.36 4.01 3.92 4.18 4.00
450 3.76 3.35 4.60 3.59 4.00
500 3.90 4.15 3.62 4.05 4.00
550 5.83 5.77 5.42 5.81 6.00
600 6.23 6.30 5.51 6.25 6.00
650 5.84 5.98 5.12 5.86 6.00
700 6.13 6.40 5.38 6.18 6.00
750 5.86 5.80 5.79 5.75 6.00
800 6.64 6.28 6.01 6.38 6.00
850 5.76 5.97 5.55 5.84 6.00
900 6.24 6.22 6.21 6.19 6.00
950 5.69 5.76 5.75 5.70 6.00
Total 1,293,952 1,717,568 133,809 3,145,329
x2(19) 0.000 0.000 0.000 0.000
Even 52.74 52.75 50.38 52.65
Odd 47.26 47.25 49.62 47.35
772(1) 0.000 0.000 0.005 0.000

 

151

Table 27-continued

Note: f,,,,,.,,,,,ed = factuaI + funm, — fdomam, where funm shows the expected frequency
given uniform distribution, and fame," denotes the frequency of domain events.
A domain event over a price category occurs when prices change and the price
path passes over or arrives in a different price category. Depending on tick
regime, the frequency of the last two to five digits of transaction prices are
shown. For example, Part A uses the last three, four, and five digits of
transaction prices for tick regimes 100, 1,000, and 10,000, respectively. The
last column reports combined results for the three tick regimes. All
transactions in the sample were used to prepare the table. The total row gives
the total number of transactions in each tick regime. 772(9), 772(3), and 772(19)
report the p value for the hypothesis that each final-digits category has an
equal probability of occurrence in parts A, B, and C, respectively. Even and
odd rows report the frequencies of even and odd final-digit categories. 772(1)
reports the p value for the hypothesis that even and odd final digits have an
equal probability of occurrence.

3 By definition expected frequencies vary across price categories for tick
regimes 50, 500, and 5000.

152

Table 28
Distribution of Bid-Ask Spread as a Multiple of Tick Size for Sample Stocks

 

 

Tick 1 2 3 4 6 8 Total
Akbnk 258 9 0 0 0 0 267
Akcns 259 1 0 0 0 0 0 269
Akgrt 262 4 1 1 0 0 268
Alark 262 5 0 1 0 0 268
Alctl 265 3 0 0 0 0 268
Arclk 265 4 0 0 0 0 269
Bagfs 245 1 1 4 0 0 1 261
Cukel 265 2 0 0 0 0 267
Dohol 265 2 0 0 0 0 267
Dyhol 124 3 0 0 0 0 127
Efes 21 5 4 0 0 0 0 21 9
Enka 254 6 2 1 0 0 263
Eregl 262 4 0 0 0 0 266
Garan 263 6 1 0 0 0 270
Hurgz 262 6 0 0 0 0 268
lhlas 260 5 1 0 0 0 266
lsctr 262 2 0 0 0 0 264
Kchol 266 4 1 0 0 0 271
Mig rs 262 3 1 1 0 0 267
Nthol 260 8 0 0 0 0 268
Otosn 264 5 0 0 0 0 269
Petkm 177 2 0 0 0 0 179
Ptofs 176 2 0 0 0 0 178
Sahol 266 1 1 0 0 0 268
Thyao 266 2 0 0 0 0 268
Toaso 264 3 1 0 0 0 268
Tuprs 263 4 0 0 0 0 267
Uzel 263 2 0 0 1 0 266
Vestl 256 9 0 0 0 0 265
kank 255 6 0 1 0 0 262
Total 7,486 137 13 5 1 1 7,643

 

 

Source: Dunya, a daily Turkish newspaper.
Note: Bid and ask prices at the daily close of the market are used to calculate
spread size.

153

 

Table 29
Price Change in Consecutive Transactions

 

 

Change Frequency Percent
-12 1 0.00
-11 1 0.00
-10 1 0.00

-8 3 0.00
-7 9 0.00
-6 18 0.00
-5 29 0.00
-4 73 0.01
-3 176 0.02
-2 1,039 0.15
-1 355,552 49.75
1 356,569 49.89
2 1,027 0.14
3 143 0.02
4 58 0.01
5 24 0.00
6 16 0.00
7 2 0.00
8 5 0.00
11 1 0.00

 

Note: Transactions of old shares only (there were 47,626 transactions

in which new shares were traded). After removing the 16,546 first
transaction of each trading session, the remaining 8,687,768 transactions
were examined. Price did not change in 7,973,021 transactions. For the
remaining 714,747, the frequencies are expressed in multiples of tick size.

154

Table 30
Variation in Clustering within Tick Regimes

Part A: Classification Rule

 

 

Price Range
Tick Low High Low High Medium‘1
25 1 ,025-1,100 2,450—2,525 4 4 53
50 2,550-3,500 4,100-5,050 20 20 1 1
100 5,100-6,000 9,300-10,200 10 10 32
250 10,250-11,000 24,500-25,250 4 4 53

500 25,500-35,000 41 ,000-50,500 20 20 1 1

1,000 51 ,000-60,000 93,000-102,000 10 10 32
2,500 110,000-112,500 245,000-252,500 4 4 53
5,000 255,000-350,000 410,000-505,000 20 20 1 1

 

Part B: Tick Sizes 100, and 1,000 Combineda

 

 

Expected
Category Low Medium High MediumCl
0 9.99 9.11 11.23 12.50
100 7.96 12.04 6.52 12.50
200 9.88 13.00 4.78 9.38
300 10.90 10.79 12.41 9.38
400 10.91 9.95 10.48 9.38
500 11.19 9.51 10.09 9.38
600 9.71 9.55 9.77 9.38
700 9.96 8.97 11.88 9.38
800 10.16 9.12 12.48 9.38
900 9.34 7.96 10.37 9.38
Total 497,883 1,387,592 193,105
772(9) 0.000 0.000 0.000
Even 50.65 50.73 48.73
Odd 49.35 49.27 51.27
772(1) 0.000 0.000 0.000
772(2) 0.000

 

155

Table 30-continued

Part C: Tick Sizes 25, 250, and 2,500 Combinedb

 

 

Expected

Category Low Medium High Mediumd

0 39.53 26.28 25.82 26.42

25 13.00 25.94 17.44 24.53

50 21.93 25.20 31.39 24.53

75 25.54 22.58 25.34 24.53
Total 192,182 2,880,770 128,256
772(3) 0.000 0.000 0.000
Even 61.45 51.48 57.22
Odd 38.55 48.52 42.78
772(1) 0.000 0.000 0.000

772(2) 0.000

 

156

Table 30-continued

Part D: Tick Sizes 50, 500, and 5,000 Combinedc

 

 

Expected
Category Low Medium High Medium‘1
0 5.76 11.00 3.88 9.09
50 4.54 7.94 2.47 9.09
100 4.92 7.79
150 4.71 5.99
200 4.86 6.95
250 4.55 6.24
300 5.40 5.90
350 4.94 5.19
400 5.64 6.17
450 4.89 5.35
500 5.55 5.38
550 2.88 8.35 4.98 9.09
600 4.14 9.32 4.91 9.09
650 4.51 8.57 4.92 9.09
700 5.38 10.18 5.06 9.09
750 5.85 9.45 4.04 9.09
800 5.49 10.20 4.10 9.09
850 5.36 7.96 3.42 9.09
900 5.69 8.39 3.97 9.09
950 4.93 8.64 3.27 9.09
Total 1,369,268 809,330 845,289
772719) 0.000 0.000 0.000
Even 52.84 49.08 54.12
Odd 47.16 50.92 45.88
772(1) 0.000 0.000 0.000
772(2) 0.000

 

157

 

 

Table 30-continued

Note: Transactions within each tick regime are sorted into three price
classifications, shown in Part A. Only one set of final-digit categories is assigned
to the low and high price classification. Tick sizes 10 and 10,000 are not
included; in the former, all transactions fall into the high price classification, and
in the latter the high price classification is undefined. The remaining 8,608,886
transactions are included in the table.

Panels B-D report actual frequencies, analogous to Table 26. Due to the
small number of observations in the low and high price classifications, the table
reports combined frequencies. Depending on tick regime, the frequency of the
last two to five digits of transaction prices are shown. For example, Part B of the
table uses the last three, and four digits of transaction prices for tick regimes
100, and 1,000, respectively. The total row gives the total number of
transactions in each price classification. 772(9), 772(3), and 772(19) report the p
value for the hypothesis that each final digits category has an equal probability
of occurrence in parts B, C, and D of the table respectively. Even and odd rows
report the frequencies of even and odd final digit categories. 772(1) and 772(1)
report the p values for the hypotheses that even and odd final digits have an
equal probability of occurrence and that the occurrence of even final digits does
not depend on the price level, respectively.

a 84,278 transactions had prices too low(high) to be included in the low(high)
price classification, and these were excluded from the analysis.

” 99,491 transactions had prices too low(high) to be included in the low(high)
price classification, and these were excluded from the analysis.

° 121,442 transactions had prices too low(high) to be included in the low(high)
price classification, and these were excluded from the analysis.

d Expected frequencies vary across price categories for the medium price range.

158

 

 

Table 31
Parameter Values and the Number Of Stock—Windows Used in the Experiment

 

 

 

Trading Investment # of \Nil'ldOWS
Test Window Vlfindow Before Filtering After Filtering
(Days) (Days)
0.5 1 3 2,100 1 ,546
1 .5 1 3 2,100 1 .543
2.5 2 3 1 ,500 1 ,181
3.5 3 3 1,200 953

 

Note: The limit order test gives the discount from the equilibrium price expressed
in ticks used to calculate the limit buy price in the experiment.

 

 

 

Table 32
The Number of Stock-Windows Conditional on Execution and Nonexecution of Limit
Orders
Subperiod
Test Status Observation 1 2 3 4 5 6 7 8 9 10 Overall
0.5 Unexecuted Stocks 2 26 16 23 12 14 2 24 12 3 134

Windows 3 43 22 27 16 17 2 46 12 3 191

Executed Stocks 28 28 28 29 29 30 28 28 28 30 286
VWndows 183141 .58111 183 164 156 124124111 1,355

1.5 Unexecuted Stocks 20 28 24 29 29 29 12 28 26 25 250
VWndows 34 91 49 50 84 53 13 77 28 40 519

Executed Stocks 28 28 22 29 29 30 28 28 28 29 279
Windows 152 93 31 88 114 128 143 93108 74 1,024

2.5 Unexecuted Stocks 28 26 21 25 29 2 20 28 25 30 234
Windows 73 60 29 46 74 2 24 66 42 80 496

Executed Stocks 28 27 29 28 23 30 28 28 28 28 277
Windows 60 72 49 58 44 124 99 63 81 35 685

3.5 Unexecuted Stocks 20 27 24 29 28 21 24 27 13 30 243
Windows 32 73 36 43 66 22 37 48 16 80 453

Executed Stocks 28 22 13 25 28 30 27 28 27 28 256
Vlfindows 71 33 14 35 44 88 59 53 66 37 500

 

Note: These data are used to compute the subperiod averages in the following tables.
The limit order test gives the discount from the equilibrium price expressed in ticks used
to calculate the limit buy price in the experiment.

159

 

Table 33

Average Standardized Execution Price of Limit Orders

Part A: Unconditional

 

 

 

 

 

Test
Subperiod 0.5 1.5 2.5 3.5
1 99.362 3 98.684 3 99.918 97.025 3
2 101.448 3 101.795 3 99.311 b 102.089 3
3 101.188 3 102.401 3 99.632 105.621 3
4 100.416 b 100.045 99.685 102.436 “
5 99.565 3 99.625 ” 100.512 100.435
6 99.750 99.222 3 96.298 3 96.107 3
7 99.410 3 98.383 3 98.243 3 100.664
8 101.565 3 101.606 3 102.416 8 101.777 "
9 99.595 3 98.678 3 98.198 3 96.320 3
10 99.446 3 99.746 102.681 3 103.011 3
Overall 100.173 b 100.024 99.689 b 100.558 °
Part B: Conditional on Nonexecution
Test
Subperiod 0.5 1.5 2.5 3.5
1 109.428 103.922 3 103.203 3 102.447 °
2 109.187 3 105.981 3 103.215 3 105.925 3
3 106.713 a 105.643 3 105.825 3 110.690 3
4 105.744 3 104.600 3 104.830 3 109.710 3
5 103.518 3 102.281 3 102.935 3 104.713 3
6 103.957 3 102.610 3 100.043 102.390
7 113.446c 105.402“I 106.113a 110.198a
8 107.944 3 106.307 3 108.236 3 109.378 3
9 103.377 3 102.578 3 102.188 3 103.066 3
10 108.133 ° 103.481 3 105.770 3 106.872 3
Overall 106.547 3 104.208 a 104.625 3 106.886 “

 

Note: Execution prices are standardized by using the equilibrium price at the
close of the day preceding the order submission day. Part A reports these
prices for all limit orders, and Part B reports these prices for limit orders that
are not naturally executed during the trading window.

3 Significant at the 1 percent level.

b Significant at the 5 percent level.
“ Significant at the 10 percent level.

160

Table 34
The Difference between Execution and Triggering Prices for Executed Limit

 

 

Orders
Test
Subperiod 0.5 1.5 2.5 3.5

1 0.208 a 0.049 b 0.022 0.000

2 0.667 2.197 2.745 3.245

3 0.800 0.746 1.008 2.058

4 0.161 a 0.000 0.008 0.000

5 0.029 b 0.000 0.014 0.169

6 0.278 a 0.059 ° 0.154 a 0.008

7 0.827 a 0.395 a 0.185 b 0.104

8 0.630 a 0.196 a 1.371 a 0.000
9 2.159 a 1.545 a 0.292 a 0.843 a

10 0.268 c 0.159 0.131 0.145
Overall 0.595 a 0.522 b 0.594 b 0.519 °

 

3 Significant at the 1 percent level.
" Significant at the 5 percent level.
° Significant at the 10 percent level.

 

 

Table 35
Average Time to Execution in Hours for Naturally Executed Limit Orders
Test
Subperiod 0.5 1.5 2.5 3.5

1 0.24 0.62 1.96 5.80

2 0.32 0.92 2.79 6.95

3 0.14 0.61 1.10 8.88

4 0.71 1.38 4.89 7.69

5 0.52 2.29 4.29 5.42

6 0.75 1.07 1.90 3.16

7 0.25 0.34 1.83 2.07

8 0.14 0.46 1.55 4.49

9 0.26 0.32 3.93 2.18

10 0.32 1.40 4.71 6.14

Overall 0.37 0.96 2.86 5.01

 

Note: Four hours is equal to a trading day.

161

Table 36

Unconditional and Conditional Returns for Market and Limit Orders

 

 

Test
0.5 1.5 2.5 3.5

Discount 0.779 2.310 3.861 5.423
Fraction Executed 87.65 66.36 58.00 52.47
Limit Orders

All 0841 a -0.696 a -0.731 b -1.803 ’

Executed -1.176 a -0.853 a -0.314 -1.495 "

Unexecuted 1.191 b -0.495 -0.763 ° -2.807 a
Market Orders

All -1.104a 4.1493 -2.111‘ -2.0713

 

Note: The first row reports average percentage discount from the equilibrium
price at the close of the day preceding the order submission day used to
determine limit buy order prices. The second row reports the fraction of limit
buy orders that are naturally executed during the trading window. Rows 3-5
report unconditional, conditional on natural execution, and conditional on
forced execution overall returns for limit buy orders. Row 6 reports the
unconditional overall returns for market orders.

3 Significant at the 1 percent level.

b Significant at the 5 percent level.
° Significant at the 10 percent level.

162

Table 37

Differential Returns

Part A: Unconditional

 

 

 

 

 

Test
Subperiod 0.5 1.5 2.5 3.5
1 0.531 ° 1.242 a -1.856 a 2.348 a
2 0.397 0.080 3.774 a 0.876
3 -1.636 a -2.789 a 2.323 b -7.749 a
4 1.430 a 1.829 a 0.985 -0.597
5 1.584 a 1.545 a 0.109 0.876
6 -0.058 0.491 1.044 b 0.042
7 -4.035 a -2.865 a 1.036 -2.857 a
8 -0.121 -0.132 -1.896 a 2.235 a
9 -0.609 a 0.330 5.666 a 3.889 a
10 4.816 a 4.552 a 2.574 a 3.679 3
Overall 0.263 0.454 a 1.380 a 0.267
Wilcoxon p 0.061 0.001 <.0001 0.004
Part B: Conditional on Execution
Test
Subperiod 0.5 1.5 2.5 3.5
1 0.319 1.515 a -2.728 3.959 a
2 0.025 0.447 3.291 a 5.447 a
3 -1.091 ° -1.711 c 5.668 a -0.803
4 0.283 0.976 3.757 a -1.238
5 1.837 a 2.880 a 2.856 a 4.195 a
6 -0.385 1.618 a 1.221 a 0.835
7 -4.118 a -3.407 a 0.106 -3.298 a
8 -1.594 b -3.236 a 1.196 0.327
9 -0.944 a -0.024 4.585 a 5.571 a
10 4.750 a 4.539 a -0.913 -1.072
Overall 0052 0.439 ° 1.890 a 1.454 ‘
Wilcoxon p 0.822 0.021 <.0001 <.0001

 

163

Table 37-continued

Part C: Conditional on Nonexecution

 

Test

Subperiod 0.5 1.5 2.5 3.5

5.393 -1.449 -2.186 -1.680
1.820 b -0.415 5.042 a -1.277 °
-3.704 b -3.479 a -4.078 a -10.753 3

6.466 a 4.096 a -3.443 a 0.627

-1.751 -0.415 -1.015 c -1.153
0.866 -2.148 c -10.227 -3.542 a

-1.827 -0.390 2.977 b -3.313
2.386 b 3.494 a -4.396 a 2.521 a
2.439 b 2.338 b 7.492 a -4.029 a
10 1.244 3.594 a 3.845 a 5.134 3
Overall 1.681 a 0.657 ° 0.373 -1.335 a

 

(OCDVODU'l-wa—t

Wilcoxon p 0.001 0.048 0.455 0.009

Note: For each firm the difference between subperiod average firm limit buy
order return and unconditional market order return forms the basic observation.

3 Significant at the 1 percent level.

b Significant at the 5 percent level.
c Significant at the 10 percent level.

164

.88. E808: or 85 am. EmoEcgw o
.88. 2008: m 85 8 200.28% a
.88. E888: _. 05 um E00535 a

.0>mu m 0.03.08 38:? E95095 8:... .:o_00_En:m .820 85 05.0809:
>0: 05 :0 800.0 9: “m 80:: Eztnzsg 85 96: >9 00520390 99s 950%? 0.00. 0:0 Em .:o_00_E::m 820 8:. 936:2
m>mu :9 85 acts: >0: :008 8 000.0 05 :0 95000:: 00900 0.0.0.2: “08:28:90 :0 00888 =90>o 9: 9m 09.003: ”902

 

 

«Nomi: .3 No.9: a 802: a 05.2: a 0:9: 5 mums? m 09.8? m N89: .5 mmmhor a wonder “09388:: m6
5 Bfmow a mmobor a 50mg 5 moméor a mmnéor a Name? 5 >363 a mmméor a 8de a SENS 85098:: m.~
amoméor a mmvéov o 85.9 a www.moF a 852:. a mood? a 836— a 28.2: a mmméor m www.mor umSomxmc: m...
1899: a Nmmsor a 9.2.2: 5 www.moF a ommsor a mmvdov a evade. a 80mm: 5 $98? a $62: “093098:: md
«09.8 a 03.3 a comma a www.mm a 308 a P560 5 203 a $58 a omvém a. «3.3 umSomxw 0m
«805 a 8me a 303 a 28.8 «90.8 a 958 a www.mm m omvém a 03.8 a 98.3 “0930me mN
www.mm 858 «38.8 dem a 25mm 5 mmndm a 25.8 a. 803 a mafimm a $08 “3930me m...
.8de 96.2: 05.09 mnvdow 868 33$ 5 www.mm 85mm mmrdow a ommdm 09:0me md
or m w n o m v m N F 0306 :08...
>00

 

>0: :o_mm_En:m .820 9: 9:50:00 825 umuEmucmuw
mm 2an

165

Table 39

Sensitivity of Costs to the Choice of Investment Window Length

 

 

 

Investment Window

Test Cost 3 Days 5 Days 7 Days
0.5 Bagging 0.052 1.181 a -1.358 a
Nonexecution -1.681 a -0.355 3.638 a
Total -0.263 0.916 a -0.590 b
1.5 Bagging -0.439 ° 0.793 b -3.604 a
Nonexecution -0.657 ° 1.119 b 3.490 a
Total -0.454 a 0.706 a -0.745 a
- 2.5 Bagging -1.890 a 1.494 a -0.951 °

Nonexecution -0.373 -0.564 0.088
Total -1.380 a 0.411 -0.506 °
3.5 Bagging -1.454 a -1.689 a -2.415 a
Nonexecution 1.335 a 0.739 -3.583 a
Total -0.267 -0.887 ° -2.340 a

 

3 Significant at the 1 percent level.
b Significant at the 5 percent level.
c Significant at the 10 percent level.

166

Table 40
Market-Adjusted Differential Limit Buy Order Returns

 

Test
0.5 1.5 2.5 3.5

Subperiod m 02 m 112 m 02 m 02
5.017 -5.211 -1.111 1.279 -1.647° 2.706c -2.670“ 3.906“
-0.944 1.258 4.039“ 1.993“ -0.319 0.818 -1.234“ 4.421“
-4.016“ 5.570“ -2.480“ 6.334“ -4.356" 7.007“ -2.840“ 8.707“
-0.854 1.308 -0.240 0.745 -0.946 1.653“ -0.271 1.420
-2.105“ 2.332“ 4.093“ 1.973“ -1.896“ 5.076“ -0.013 0.542
-0.481 0.745 0.19 -0.006 0.552 -0.467 -1.159 1.337
40.498" 10.189“ -6.766“ 6.964“ -3.962“ 4.718“ -3.876“ 5.756“
-1.747“ 2.351“ -0.639 1.047 -1.152 2.536“ 4.896“ 3.182“
-1.827 1.855 -0.893 1.013 -0.858 1.663c -5.476“ 6.975“
10 -0.152 0.413 -1.773“ 3.123“ 0.167 -0.038 -0.235 1.376
Overall -1.573“ 1.793“ -1.123“ 1.693“ 4.297“ 2.249“ -1.461“ 2.800“

mmumm-tsz-A

 

Note: For each firm the difference between window limit order return and average
market order return during the corresponding subperiod was regressed on a constant
and the difference between portfolio window limit order return and portfolio average
market order return during the corresponding subperiod. The residuals from these 30
regressions were stacked to form a vector and were then regressed on a constant and a
dummy that takes on the value 1 for natural execution, 0 otherwise. 111 is the intercept,
and 02 is the coefficient of execution dummy. Subperiod and overall results are
reported for each limit order test.

a Significant at the 1 percent level.

° Significant at the 5 percent level.
° Significant at the 10 percent level.

167

Table 41
Sensitivity of Market-Adjusted Differential Limit Buy Order Returns to the
Choice of Investment Window Length

 

 

 

Investment Window

Test 3 Days 5 Days 7 Days
0.5 17, -1.573 a -2.050 a -1.725 b
112 1.793 a 2.385 a 1.964 b

1.5 n, -1.123a -1.2768 -0.812
112 1.693 a 1.887 a 1.287 b

2.5 17, -1.297 a -0.998 a -0.639
112 2.249 a 1.848 a 1.359 b
3.5 111 -1.461 a -1.010 b -1.300 b
112 2.800 a 1.929 a 2.442 “

 

3 Significant at the 1 percent level.
” Significant at the 5 percent level.
6 Significant at the 10 percent level.

168

 

 

 

.88. E808: or 85 8 8:805:85 u
.88. 5:80.88 8 85 88 5:805:85 ..
.88. 8:80:88 P 858 8:805:88 .

.8088 ho 08.88.35

88 Eek. 8:8 .owz .Om 8:8 0889:8088 :. 8800998 88 0: 8:8 220.». 80.88 8:.885 85 n8 8:8 85 8 80...: 85 >8 08:_8> 8... 85
08~.888:90 >8 0859.. 8E80 85 08088. m 888 JP :. mo: 8:8 .598 dwz .0m .00 .z_<0km 08008.98 < .88 .8889: 8,500.0 85 85820::
0:58:80: ”.0 88:5: .99 85 >8 85.88085 8500.0 505 ES: 85 8:88.05. 59: .fi .99 85 8:.8.>.8 >8 8:98 0. 80.55 888.80 888 5.5..
0. mo: .3888 85 EB. :88 .99 0. EB: .8000 8.509880: 85 0. 8.2 .800 858.888 85 0. 0m .0998 88.9.88 :. 8800998 80.88 8:.885
8 8:8 85 8 8058885. >:9:8>:. 0. ms: 8808.0 0. >988. 85 c855 8088 Am:.0m:0.:8.m:.._80 85 8:8 80.88 85885 85 8:888 88088053
8.00 00.090 :9 8.88. 882809 80:: 88988 85 :88388 80888.8 850. o: .8088: 8:898 85 8:888 88:0 :5: .0 5.8098 85 >8 88:08:00
0. 808.885. >988. 85 08:5 .0 88:5: 0. @2um 80:9, 80.> 8:8 0.090 85 8:...80 8:8 m:.>:8 So... :88 88988 85 08>.m 220.5. .882

 

 

 

 

 

 

00 . 33- a 005- a 08.0 a 02.0- a £00 83 a 0.4.2 . 00: 0
00 30.8. .9. F- m 02.0 a :60- a 23 18.0 m 0.38 b 50.0 N
2 0o.. 59: 082 00 ms: 3.808 08.58 30.2288 .03
8 :8
00 00.83- ”00.80.02- .032.me .3839- 10.0 8.88.0- .082 .8028 m
00 BEN- 5.30.9- ”00.05.09. 10.30.90- 39.0 3:08- 805.88 ..0809: N
z 00.. .89: 082 00 ms: 08 0:85 229.: .089
_ “<5:

80.5 5:88:55 85 ES. 00.2.? v. H 8 088.0 __8w 8:8 >:m 5.5... 858.585 2088:9555 n.o 5:85.896 85
N8 8.83

169

 

 

900.909. 099.00 0 0 0 N0 00 909 9.90 N099 000.N 000.99 009.0N 69.00
099.009 NN9..9 9 - 9 0 99 0N 00 90N 900 N900 000.09 099:
909.90N 009.09 9 9 N 0 0N 00 009 004 900.9 90N.0 09.9.9 95.98:
000.09 9 990.0 9 - - - 0 99 N0 NN9 0N0 009.N 990.0 0090
000.09 000.09 0 9 N 0 99 09. 00 9N0 900 099... N90.09 .652
39.099 00N.0 9 - 9 9 0 09 90 009 N3. 04N.N 0.0.0 0.0.5.
09 9.90N 9.00.0N 0 N .9 N9 0N 09 N09 0N0 090.9 9N9_0 999.49 .28.
9N0.090 900.00 09 0 0 90 099 N0N 0N0 0N0.N 9N0... 98.09 900.00 .900.
900.000 009.0N 0 9. 9 09 00 099 90N 009 909.9 00N.0 900.99 0.0....
000.00N 000.NN N 9 9 v 9 99. 0N9 v0.9 N9N.9 000.0 090.: N05:
9008 9.09.9N 0 0 9. 09 0N 09 09N 000 900.9 090.9 000.09 00.08
090. 9NN 990.99 NN 0 0 N9 0N 00 009 009. 000.9 090... N009 .098
09 9.009 900.09 9 - - - N 9 90 009 0N0 9Nv.N 000.0 9.9.8
9N0.00N 0N0.0N 9 - - 0 N9 3 009 0N0 090.9 900.0 000.N9 0890.
00.9.09N 000.99 0 N 0 0 99 N0 0N9 30 N00 000... 89.99 .298
909.9909 90.9.N0 ON 9 0 90 9.09 90N 090 000.9 99N.9. 000.09 09.9.00 .96:
000.9N 099.9 9 - - N 0 09 00 SN 0N9 90N.0 090.0 .930
900.08 80.09 0 9 N 0 99 0N .99 $0 900 000.0 090.99 09000
039 009.99 9 9 9 0 9 9N .90 N0N 000 900... 090.99 3.82
090.08 000.9 N 9 N 09 «N 9.0 0.9 909. 009.9 N000 8:9 .902
009.099 .000 9 - - - 9 0 N0 9N9 0N0 09 9.N 990.0 0.902
009.90 009.9 9 - - 9 9 0 09 90 00N 000.9 99.0 9.032
000.NNN 099.NN 9 - 0 99 0N 00 009 90.. 90N.9 30.0 900.09 0:92
090.00.. 099.00 9 v 0 09 0.. 0N9 90N 0N0 N0N.N 9N0.09 000.9N 9.9.92
08.900000: .069 09 0 0 9. 0 N 9 0 3.8.0
_.< 03

 

>_8>.S080:oo 888:9..0 80.5 80.:>> :. 080:8:88w 888; .90 8:89.889“.
9. 8.889

170

88:98 .0: 8.8 80.... 9.0.9.2, :. 0:0.9080c85 .0 .8858: 85 02.080 0:53.00 02: 908. 85 882,988 80:99.8 8......
.>.8>.88808. 08808880 .888. 8:8 0088. 85 09:. 88:8.008 8.8 000.0 8:8 09.0.0 080:8 8 88....000 85 0:0.9080:8.._. .8088 .oo~.m .8098 .8989
80:88:80 9.30:0. 85 82200 8.8598 .0... 080.... 09.0.8.8 .0 5.8.88 85 :0 88088 08808880 09:. 88:8.008 0. :0.9080:8.9 :08m .902

 

171

 

000.N00.0 909.0N9 009 00 N99 .900 N00 09N.N 0.9.9.0 90.99 0N9.0.9 000.00N 00990.9 .069
900.900 0N0. 999 N N .9 N9 00 909 09N 900 000.N 900.99 000.0N 9.09;
90N.009 090.N9 99 N N N 0 ON 00 0NN 000 9N9.0 009.0 08>
00N.0NN 009.0N 9 - - 9 09 N99 .999 N999 099.9 090.0 900.09 .89:
399.000 900.990 N0 09 NN 00 909 09N .00 .9099 900.0 900.09 090.90 0.9:
000.00N 000.9.N 9 9 .9 09 0N 09 009 000 0.90.9 N.90.0 000.99 0089
090.90 000.0N N 0 0 9 0N N0 099 090 000.9 900.9 900.09 60.39
08.900000: .069 9 0 0 9 0 0 .9 0 N 9 0 0.8.0
__< 00.

 

09:...08-0.9 0.009

Table 44
Distribution of Intervals Based on Price Change and No Price Change

 

 

 

Classification
15 min 30 min 60 min

Stock No Change Change No Change Change No Change Chang_e_
Akbnk 2,284 2,216 942 1,308 364 762
Akcns 2,131 2,353 893 1,349 316 806
Akgrt 2,533 1,967 1,021 1,229 380 746
Alark 2,427 2,065 961 1,285 363 761
Alctl V 2,373 2,127 1,016 1,234 364 762
Arclk 2,318 2,182 955 1,295 376 750
Bagfs 2,439 2,061 970 1 ,280 389 737
Cukel 2,551 1,949 1,075 1,175 387 739
Dohol 2,169 2,331 900 1,350 361 765
Dyhol 1 ,002 1 ,202 415 687 143 409
Efes 2,136 1,876 888 1,1 18 309 695
Enka 2,569 1,931 1,072 1,178 419 707
Eregl 2,378 2,106 988 1,254 359 763
Garan 2,203 2,289 916 1,330 352 772
Hurgz 2,216 2,284 909 1,341 346 780
lhlas 2,311 2,189 974 1,276 382 744
lsctr 2,279 2,141 970 1,240 330 776
Kchol 2,290 2,202 965 1,281 364 760
Migrs 2,561 1,939 1,045 1,205 422 704
Nthol 2,276 2,224 967 1,283 393 733
Otosn 2,542 1,958 1,046 1,204 395 731
Petkm 2,414 2,086 ‘ 982 1 ,268 399 727
Ptofs 2,147 2,289 842 1,376 301 809
Sahol 2,217 2,283 946 1,304 359 767
Thyao 2,348 2,152 998 1 ,252 360 766
Toaso 2,263 2,237 950 1 ,300 367 759
Tuprs 2,240 2,260 915 1,335 333 793
Uzel 2,234 2,266 948 1,302 387 739
Vestl 2,334 2,166 966 1,284 341 785
kank 2,134 2,366 867 1,383 311 815
Total 68,319 63,697 28,302 37,706 10,672 22,362

 

172

Table 45
Stationarity of Price and Volume Series

 

 

Dickey-Fuller Phillips-Perron
Tau Pr < Tau Tau Pr < Tau

Price

Zero Mean 0.07 0.7060 0.07 0.7050

Single Mean -1.01 0.7519 -1.07 0.7310

Trend -0.84 0.9605 -0.94 0.9510
Return

Zero Mean -31.26 0.0000 -66.26 0.0010

Single Mean -31.26 0.0001 -66.26 0.0010

Trend -31.27 0.0001 -66.25 0.0010
Turnover

Zero Mean -12.59 0.0000 -29.19 0.0010

Single Mean -22.20 0.0001 -45.14 0.0010

Trend -22.54 0.0001 -45.35 0.0010

 

173

Table 46

Average Turnover during 15 Minute Intervals by Weekday, in Percentage

 

 

Interval Monday Tuesday Wednesday Thursday Friday All Fdfl
10:00-10:15 1.018 0.876 0.986 0.970 0.911 0.952 0.40
10215-10230 0.534 0.536 0.636 0.565 0.535 0.561 0.57
10:30-10:45 0.371 0.425 0.414 0.484 0.415 0.422 0.74
10:45-11:00 0.304 0.350 0.368 0.351 0.371 0.349 0.38
11:00-11:15 0.286 0.308 0.338 0.310 0.350 0.318 0.49
11:15-11:30 0.278 0.321 0.311 0.285 0.278 0.295 0.27
11230-1 1 :45 0.242 0.225 0.379 0.300 0.333 0.295 2.30 °
11:45-12:00 0.298 0.365 0.351 0.353 0.334 0.340 0.34
14:00-14:15 0.566 0.667 0.634 0.644 0.662 0.634 0.20
14:15-14:30 0.465 0.528 0.469 0.508 0.443 0.483 0.30
14:30-14:45 0.337 0.439 0.376 0.448 0.388 0.398 0.64
14:45-15:00 0.313 0.408 0.367 0.373 0.410 0.374 0.64
15:00-15:15 0.311 0.362 0.441 0.410 0.439 0.392 1.03
15:15-15:30 0.364 0.452 0.389 0.474 0.409 0.418 0.67
15:30-15:45 0.370 0.455 0.444 0.433 0.450 0.430 0.55
15:45-16:00 0.700 0.846 0.862 0.785 0.850 0.809 1.29
All 0.422 0.473 0.485 0.481 0.474 0.467
F,,,, 17.02 3 8.02 a 9.94 a 11.43 ° 11.60 ‘

Fm, 187.68 ‘ 61.77 ‘ 85.89 a 101.70 a 96.01 ‘
F,,,,,,,, 22.44 ' 16.45 ‘ 14.74 ‘ 19.59 a 24.82 ‘
F,,,,,,,,,,,, 66.69 ‘ 55.88 “ 61.33 ‘ 54.33 a 81.64 ‘

 

Note: Turnover per stock is calculated by dividing the cumulative volume during an
interval by the number of floating shares (number of outstanding shares*fioat). The
reported results are the equal weighted averages of individual stock mean turnovers.
Fm, tests the hypothesis of equality of mean turnover during all intervals in a given
weekday. Fm, ,F,,,,,,,,, and F300...“ test the hypotheses that mean turnover in interval 1,
9, and 16 are not different from the mean turnover in the remaining intervals,
respectively (excluding intervals 1, 9, and 16). Fday tests the hypothesis that there is
no interday difference in mean turnover during a given interval. F,,,, has degrees of
freedom of (15,880), (15,896), (15,864), (15,848), and (15,864) for Monday-Friday,
respectively. Fm, , F,,,,,,,,, and F,.,,,,,,,,,, have degrees of freedom of (1,782), (1,796),
(1,768), (1,754), and (1,768) for Monday-Friday, respectively. Fclay has degrees of
freedom of (4,272).

‘ Significant at the 1 percent level.

" Significant at the 5 percent level.
° Significant at the 10 percent level.

174

Table 47
Average Return Volatility during 15 Minute Intervals by Weekday (x104)

 

Monday Tuesday Wednesday Thursday Friday All Fcm

 

Interval
10:00-10:15 2.737 1.770 2.925 2.155 3.671 2.647 0.67
10:15-10:30 0.982 0.763 0.768 0.798 1.130 0.888 0.33
10230-10245 0.541 0.816 0.443 1.070 1.004 0.773 0.69
10:45-11:00 0.420 0.241 0.360 0.382 0.375 0.355 0.52
11:00-11:15 0.420 0.279 0.541 0.311 0.584 0.426 0.61
11:15-11:30 0.535 0.288 0.306 0.239 0.461 0.366 0.66
11:30-11:45 0.332 0.170 0.282 0.284 0.254 0.264 0.54
11:45-12:00 0.356 0.270 0.261 0.297 0.221 0.281 0.58
14:00-14:15 0.922 1.126 0.978 1.358 1.083 1.092 0.31
14215-14230 0.385 0.760 0.470 0.409 0.443 0.495 1.13
14:30-14245 0.373 0.210 0.310 0.620 0.606 0.421 1.67
14:45-15:00 ' 0.183 0.229 0.534 0.204 0.327 0.295 3.33 "
15200-15115 0.279 0.410 0.489 0.400 0.441 0.403 0.42
15:15-15:30 0.396 0.417 0.358 0.357 0.737 0.453 0.75
15:30-15:45 0.502 0.278 0.365 0.406 0.730 0.455 1.05
15245-16200 0.334 0.623 0.414 0.295 0.526 0.440 1.50
All 0.606 0.541 0.613 0.599 0.787 0.628
F,,,, 5.58 a 5.57 a 4.11 ' 4.86 a 4.41 ‘
F,,,, 70.09 a 63.38 “ 52.55 “‘ 51.77 a 53.99 a
F,,,,,,,, 9.41 a 18.20 a 19.15 ‘ 21.92 ‘ 3.48 °
5mm" 0.51 2.27 0.00 0.88 0.02

 

Note: The reported results are the equal weighted averages of individual stock
volatilities, proxied by return squared. Fm, tests the hypothesis of equality of mean
volatility during all intervals in a given weekday. Fm F,,,,,,,,, and F3000". test the
hypotheses that mean volatility in interval 1, 9, and 16 are not different than the mean
volatility in the remaining intervals, respectively (excluding intervals 1, 9, and 16). Pm
tests the hypothesis that there is no interday difference in mean volatility during a
given interval. F,m has degrees of freedom of (15,880), (15,896), (15,864), (15,848),
and (15,864) for Monday-Friday, respectively. Fm, , PM“, and F,,,,,,,,,,,, have degrees of
freedom of (1,782), (1,796), (1,768), (1,754), and (1,768) for Monday-Friday,
respectively. F,my has degrees of freedom of (4,272).

° Significant at the 1 percent level.

” Significant at the 5 percent level.
° Significant at the 10 percent level.

175

Table 48

Average Return during 15 Minute Intervals by Weekday

 

 

 

Interval Mondayj Tuesday Wednesday Thursday Friday All £42,,
10:00-10:15 -0.245 -0.020 0.017 0.301 0.287 0.065 1.10
10215-10230 0006 -0.112 0.081 0.139 -0.010 0.017 0.57
10:30-10:45 -0.092 -0.272 0.059 -0.235 -0.137 -0.136 1.25
10:45-11:00 0.027 .-0.066 -0.041 -0.108 -0.123 -0.062 0.56
11:00-11z15 0.012 -0.104 -0.170 -0.118 0.002 -0.075 0.82
11:15-11:30 -0.072 -0.010 -0.146 -0.021 -0.119 -0.073 0.54
11:30-11245 -0.055 -0.025 -0.088 0.055 -0.108 -0.044 0.85
11:45-12:00 0.153 0.066 0.112 -0.052 0.229 0.102 223°
14:00-14z15 -0.040 0.021 -0.026 -0.052 0.113 0.003 0.22
14215-14230 0.016 -0.021 -0.098 0.055 0.065 0.003 0.48
14:30-14245 0.039 0.040 0.003 -0.050 0.034 0.014 0.19
14:45-15:00 -0.095 0.005 -0.142 -0.125 0.072 -0.057 1.60
15:00-15:15 -0.244 0.018 -0.062 -0.043 0.063 -0.054 1.94
15:15-15:30 -0.029 0.051 -0.122 -0.049 0.176 0.006 1.58
15:30-15:45 -0.052 -0.037 0.075 -0.001 0.085 0.014 0.48
15:45-16200 0.139 0.473 0.369 0.301 0.455 0.348 3.32b
All -0.034 0.001 -0.011 0.000 0.068 0.004
F,,,, 1.09 2.43 “ 1.64 ° 1.92 b 1.87 "

Fm, 3.98 b 0.03 0.29 10.55 a 4.74 b
F,,,,,,,, 0.01 0.39 0.03 0.01 0.78
F300"... 3.49 ° 34.78 “ 20.93 “ 13.88 a 17.89 '

 

Note: Return per stock is calculated as the difference of log prices at the end and at
the beginning of an interval. The reported results are the equal weighted averages of
individual stock mean returns. F,,,t tests the hypothesis of equality of mean return
during all intervals in a given weekday. Fm F,,,,,,,,, and F5000. test the hypotheses that
mean return in interval 1, 9, and 16 are not different than the mean return in the
remaining intervals, respectively (excluding intervals 1, 9, and 16). Fday tests the
hypothesis that there is no interday difference in mean return during a given interval.
F,,,, has degrees of freedom of (15,880), (15,896), (15,864), (15,848), and (15,864) for
Monday-Friday, respectively. Ffrst , F,,,,,,,,, and F,,,,,,,,,,,,, have degrees of freedom of
(1,782), (1,796), (1,768), (1,754), and (1,768) for Monday-Friday, respectively. Eda,
has degrees of freedom of (4,272).

3 Significant at the 1 percent level.

° Significant at the 5 percent level.
° Significant at the 10 percent level.

176

Table 49

Regression Results for the Test of a Contemporaneous Relationship between

Volume and Return

 

Specification 3
p Value Estimate p Value

 

Specification 1 Specification 2

Estimate p Value Estimate
Intercept -0.3302 <.0001 -0.1962 0.0137
|R| 0.4910 <.0001 0.5394 <.0001
|R| D -0.0622 0.0720 -0.0688 0.0398
Dum(Tue) -0.0048 0.9426
Dum(Wed) —0.01 15 0.8691
Dum(Thu) 0.0079 0.9102
Dum(Fri) 0.0216 0.7571
Dum(10130) -0.0088 0.8750
Dum(10:45) 0.0151 0.8238
Dum(11200) -0.0153 0.8329
Dum(11215) 0.0121 0.8728
Dum(11230) 0.0178 0.8238
Dum(11z45) -0.0081 0.9176
Dum(12:00) -0.0194 0.8106
Dum(14:15) -0.0044 0.9567
Dum(14230) -0.0267 0.7432
Dum(14z45) -0.0106 0.8941
Dum(15200) -0.0150 0.8544
Dum(15:15) -0.0023 0.9758
Dum(15230) -0.0086 0.9059
Dum(15z45) -0.0089 0.9002
Dum(16200) -0.0188 0.7630
Aftaug -0.3361 <.0001
|R| Dum(Tue)
|R| D Dum(Tue)
|R| Dum(Wed)
|Rl D Dum(Wed)
|R| Dum(Thu)
|R| D Dum(Thu)
|R| Dum(Fri)
|R| D Dum(Fri)

|R| Dum(10230)
|R| D Dum(10:30)
|R| Dum(10:45)
|R| D Dum(10:45)
|R| Dum(11:00)
|R| D Dum(11200)
5| Dum(11215)

177

-0.0087
0.3741
-0.2782
-0.0429
-0.0668
0.0232
0.0176
-0.1335
-0. 1452
-0.1867
-0.2309
-0.2044
-0.3351
-0.1584
-0.2171
-0.1572
-0.2925
-0.1063
-0.3119
-0.2265
-0.2143
0.0681
-0.3298
-0.0770
0.2576
0.0608
0.0262
-0.0452
0.0504
-0.0615
0.1352
0.2315
-0.1063
0.1035
0.2445
0.2182
0.0509
0.2658

0.9291
0.0045
0.0510
0.5829
0.3945
0.7777
0.8267
0.1991
0.1339
0.0707
0.0310
0.0537
0.0013
0.1506
0.0536
0.1546
0.0081
0.3552
0.0025
0.0403
0.0496
0.4971
<.0001
0.4519
0.0217
0.5753
0.7984
0.6374
0.6099
0.5072
0.1843
0.2079
0.5492
0.5144
0.1791
0.1698
0.7699
0.1146

Table 49-continued

 

 

Specification 1 Specification 2 Specification 3

Estimate p Value Estimate p Value Estimate p Value
|R| D Dum(11215) 0.1659 0.3409
|R| Dum(11230) 0.2578 0.1006
|R| D Dum(1 1 :30) 0.1203 0.4788
|R| Dum(11z45) 0.4521 0.0040
|R| D Dum(11z45) 0.0108 0.9525
|R| Dum(12:00) 0.1435 0.3714
|R| D Dum(12200) 0.1146 0.4989
|R| Dum(14:15) 0.2944 0.0975
|R| D Dum(14:15) 0.0124 0.9510
|R| Dum(14z30) 0.1146 0.4618
|R| D Dum(14230) 0.1538 0.4035
|R| Dum(14:45) 0.3020 0.0791
|R| D Dum(14:45) 0.1866 0.3280
|R| Dum(15100) -0.0635 0.6643
|R| D Dum(15200) 0.3975 0.0184
|R| Dum(15:15) 0.2993 0.0877
|R| D Dum(15:15) 0.2829 0.1386
|R| Dum(15230) 0.3000 0.0222
|R| D Dum(15230) 0.0144 0.9253
|R| Dum(15z45) 0.2162 0.1163
|R| D Dum(15245) 0.1438 0.4479
|R| Dum(16:00) -0.1217 0.4355
lR| D Dum(16200) 0.0368 0.8236

 

Note: The three specifications are versions of the model in equation (10).

178

 

Table 50
Granger Causality: Relationship between Return and Volume

 

 

 

Dependent Return, Volume,
Coefficient p Value Coefficient p Value
Intercept 0.0000 1 .0000 0.0000 1 .0000
Return,1 0.0290 0.3194 0.0394 0.0389
Return,2 0.0793 0.0014 0.0694 <.0001
Return,3 0.0562 0.0558 0.0548 0.0018
Return, 4 0.0287 0.2745 0.0202 0.3038
Volume,1 -0.0374 0.1519 0.2634 <.0001
Volume,2 -0.0107 0.6931 0.0895 0.0021
Volume,3 0.0364 0.2134 0.1460 <.0001
VolumeM -0.0092 0.7494 0.1705 <.0001
Wald 3.99 0.4075 30.86 <.0001
Table 51

Granger Causality: Relationship between Return Volatility and Volume

 

Volume,
Coefficient P Value

Dependent |Return,|

Coefficient P Value

 

Intercept
|Retu rn,_1 |

|Return,,2|
|Return,_3|
|ReturnM|
Volume,1
Volume,2
Volume”
Volume“
0..

t-2

1-3

DUO

t-4

Wald

0.1615
0.1135

-0.0434
-0.0716

0.0198
-0.0659

0.0066

0.0544
-0.0254
-0.1460
-0.1144
-0.0354
-0.0661

8.59

0.0043
0.0204

0.3285
0.1325
0.6135
0.0124
0.8173
0.0811
0.3949
0.0004
0.0041
0.4169
0.1243

0.0723

179

0.2350
0.0376

-0.0677
-0.0709
-0.0066
0.2528
0.1102
0.1596
0.1609
-0.0641
-0.1193
-0.0654
-0.0808

40.34

<.0001
0.2394

0.0177
0.0086
0.8370
<.0001
0.0005
<.0001
<.0001
0.0770
0.0006
0.0632
0.0298

<.0001

 

APPENDIX B

FIGURES

180

A

 

 

 

 

v...
Q-
o
3......” .
0..
'5
- 9.
I ~ I
0‘
- ‘0
r1 '?
"5
o
..o ' '
a...
.‘I'.
“:0
..- l ,..
..-¢--" "
_.!.l
5""
o. ‘
III
II
’ a.
-—" 9&-
.‘ .‘..
a
:0 ,-.
. ..
I.
E-
. .
O . . i
:_ .-
1 5 ‘m'
oopdpozipupuas

181

5 5551/20/92
f 6661/20/91
’ 5651/20/50
; 5551/10/92
5551/10/80
i 8551/21/52

9661/le8l.

5 8551/21/50
S 8551/11/05
8551/11/51
2 8651/11/01
E 5551/01/82
8551/01/51
i 8551/01/80
2 8551/60/52
-’ 8551/50/81
i 8661/60/60
2 8551/80/15

9661/80/02

'966l/901ll

8661/10/18

. 9551/20/22
I 8551/20/51
, 8661/L0/ZO

866 lBO/EZ

: 8551/90/21
1 8551/90/50
f 8551/50/52
. 8551/90/51

866le

: 5551/18/22

866lN0/Cl

: 8551/50/92
. 9551/50/21
: 8551/50/90
_' 8661/20/53
8551/20/91

966l/ZOI90

I. 5551/10/52
; 8661/10!”
8551/10/90

day

l

i—f—SAMPLE nah-155108

Figure 1
Standardized Price Level of ISE100 Index and Equal Weighted Portfolio of Sample Firms

 

 

 

 

 

5\\\\

TL250 -

TL100

TL50

TL25

regime

 

 

3.0%

2.5% </

2.0%

%
5.

1

on? so: 2529.

1.0%

0.5% l

0.0%

. 1.
2.500 Wamdmn 5000 0000

1,000

Figure 2
Tick Rule on the ISE

 

 

 

 

 

 

 

4

3
9.83 3 i=5:

2

men IN
QON:

amt. {NP
093 2 P w

022:

029.20 P

0280..
cm: 2m
QOB
02 F S
out. to
$290
QQNNR
was»:
oQNNE
we to
cm: In
oQon
manna
oQo Sn
029.95.“
oQQN
020:

Figure 3
Distribution of Nontrivial Spread over Time

182

trading window=2 days investment window=3 days

a) limit order gets natural execution b) limit order gets forced execution

limit order
trading window

limit order

trading window investment window investment window

 

 

 

 

 

 

9‘1 1 1 1’1 (I 1 ’1‘1 1 >1

0 l 2 3 4 5 day 0 l 2 3 4 5 day
( ) ( >

investment window investment window

market order

market order

Figure 4
Timing of Events in the Experiment

102

 

pflco

 

 

 

 

,¥9;5“-:13_3-7_5__£33
Figure 5
Standardized Market Price during 10 Days Following the Order Submission for
Executed Limit Orders

183

 

1

 

 

 

 

 

4H» ,_,,,,,,,, - _, "7

 

150 L_._,.-._--_m
120 ~—~~A +--«~

150
140
40 --

20

nwdmmPBENo
«.mmmpatflo
«$988
Wmmmtntwo
0.308308?
vwéoott‘up
vwgpg—
N68218:.
0.82: 2:
mpg—33v
$2va—
org—.38..
mgwawao
nwgwkgo
Tamwag
$5358

38:58 W

npémmwgmo
70820250
0&8 FB§
mémm SN :8
9.89298
wémmtom‘mo
0.00m 2.830
mémm wag
232N910
708294.30
mg to {no
mémmtmwao
n Tag 2N 2N0
«damp‘No‘No
9.1083320
Pémmtno‘wo

 

 

 

Figure 6
Index Price Level over Time

oESSEo
2, 9.89328
.. 385:8
38528
meow—SEN.
_ 3852a.
. 93859.
. 5385::
. $853.
. 9.8935.
“ 9.85:8?
L. 95853“.
1.0. 538580
.5... . 3853..
385.3
1! 38528 M
NESSQB
£58528
. ~-o85§
3855c
3853..
2.89%
. - . 3. 358.33
.lltltl.-. .. .. a 385%
A 3853“.
. “85528
W 9.89528
. L 285598
m 9.85550

      

 

 

 

 

   

. _ 38.3
382an

“

 

008
006 -

007

7%.ng

’
m m m m m o
O o

835:. 58:09.3

Figure 7
Percentage Turnover over Time
1 84

 

 

 

 

 

0.10

 

 

 

 

 

 

 

E32 not...

mow—A-» _7

-0.15

9ng «No
mpmmawaoao
Tmmmwgpo
9mg '85
mémmtnmap
5.0mm «Btu?
908 (BE:
80829;:
mémm (nos:
wig—.3 {or
908 28.6..
aroma {ago
Yomm-B two
EFBNBO
90mm FR too
opémmtvgo
«Toma—gun.
vpéomtmotb
mpémmpamao
«108.328
Toma-Bog
gtwgo
$2.330
opémmsg
Nwémmwgvo
vwémmtmmsmo
09.08.2330
«.08 :waNo
Yong)” :No
gtwoao
owémmtmtwo
Némm Smoto

Figure 8
Index Return over Time

 

 

 

I I

 

 

0016

0.014

0.012

m m.

0 0
$225.53! .39..

0010

0004
0002*

 

 

0000

. owdgtuwao

mpg—$330

IV Tmmmwgpo

némm 28:0
mémm wgp
508 wBZN F
9.08 :53:
Nam 26::
mémm who: r
70¢me So.
@108 283..
«.mmm Sang

- 185.8
$8.88

0.08 2h Coo

opémmwgo m
u

Nwémmrgso
:émmpgho
opmoawamswo
Némm 20:8

a 38528

wig—=30
mémmwkosmo
opémmw?§

t. «38593
. 388580
. 9.8898
‘ $8.88

Twain—No
nggo
wwbomZmao
”$82950

Figure 9
Index Return squared over Time

185

 

 

 

867me—

 

 

m w i .. _ 9mg»?
”1

on.m_..m—Hm—

U mew—3006..
w H _ ,
0 u 823.
a , i
i 7

mv.v_..0m v—

 

 

 

 

i r

_ i _ _ 83$ 3

1

  

mfvwnoomvw

 

 

, W i
, “ ,1

a e H 8S6???
, ,— W
, .
, , w

a ‘ mvé...0m.§

,
V
l
7 a

 

8.2..mvop

 

 

,n I 99-82
7 . i i
m _ 1

  

Ono—3m— 0—

 

 

web-3009

 

 

1.200

.8352 35:09.3

umoofday

Figure 10
Average lntraday Turnover

a
dey
mammw
Mm mm”
D'UDI1

 

 

 

 

 

 

4000

 

unison ESP.

8nmrfivnmw

mvnmwénnmw

. 89....th

mwnmwéoum—

. 89.9.!

mvdeénnvw

2 : 0031.3“:

9 . 3:693

San—$3:

x 32.8.:
8.73”:
a , 32.8.:

., . 8.79.?

a . 9.989

0067.35:

., £9.89

timofdly

Figure 11
Average lntraday Volatility

186

 

 

 

 

 

 

 

0.600

 

€22 58:09.2.

-0 400

oowwfivmp

awn—.9,” mp

ommpfi— mp

2meng

oomeVvF

mv v70”:

Chi—$3.:

mfvwéo 3

DON—Ame 3 .

mvwwpdmep

RUN—7m? :

mF :éo PF

ooaTmVOw

mv‘owémop

ononwC—

mw oTOoop

time 0! day

Figure 12
Average lntraday Return

187

 

APPENDIX C

DETAIL TABLES

188

Table 52

Unconditional and Conditional Returns for Market and Limit Orders, Investment

 

 

Window of 5 Days
Test
0.5 1.5 2.5 3.5

Discount 0.788 2.335 3.845 5.378
Fraction 85.92 67.49 53.95 52.67
Executed
Limit Orders

All -2.827 a -2.898 a -2.955 a -3.230 a

Executed -3.111 a -3.153 a -4.351 a -4.314 a

Unexecuted 0.285 -2.223 a -0.841 -3.106 a
Market Orders

All -1.9113 -2.1933 -2.545“‘ -4.117“‘

 

Note: The first row reports average percentage discount from the equilibrium
price at the close of the day preceding the order submission day used to
determine limit buy order prices. The second row reports the fraction of limit buy
orders that are naturally executed during the trading window. Rows 3-5 report
unconditional, conditional on natural execution, and conditional on forced
execution overall returns for limit buy orders. Row 6 reports the unconditional

overall returns for market orders.
3 Significant at the 1 percent level.

b Significant at the 5 percent level.
° Significant at the 10 percent level.

189

 

 

Table 53

Differential Returns, Investment Wlndow of 5 Days

Part A: Unconditional

 

 

 

 

 

Test
Subperiod 0.5 1.5 2.5 3.5
1 0.925 “ 2.124 “ -0422 -0.954
2 -0.401 -2381“ 3.459“ 5.479 “
3 0.327 2.092“ 3.131“ 1.336
4 -0.160 0.198 1.041“ 4.912“
5 0.205 0.688b -2525“ 2.024“
6 -3740“ -2764“ 3.089“ 3.686“
7 -6451“ 4741“ -6177“ -8131“
8 1.171 1.737b -7215“ 41.710“
9 -0516“ 1.088“ 4.796“ 3.936“
10 -0.463 -0792“ 2.731“ 7.842“
Overall -0916“ -0.706“ -0411 0.887°
VWcoxonp <.0001 0.000 0.459 0.001
Part B: Conditional on Execution
Test
Subperiod 0.5 1.5 2.5 3.5
1 0.654 1.875 ° 2.527 -9021 “
2 -O.629 -3831“ 3.297 “ 7.285“
3 0.239 0.038 2.321b 1.336
4 -2223“ -0974 1.500“ 4.300“
5 0.258 1.279b -0.047 2.874
6 -3.650“‘ -2407“ 2.661“ 3.512“
7 -6.451“ 4014“ 40.767“ -2.822°
8 2.335“ 3.335“ 41.989“ 7.045b
9 0516“ 1.088“ 3025“ 7.922“
10 4.588“ -3.927“ 4.418 0717
Overall 4.181“ -0793“ 4.494“ 1.689“
Wllcoxonp <.0001 0.018 0.061 <.0001

 

 

190

Table 53-continued

Part C: Conditional on Nonexecution

 

 

Test

Subperiod 0.5 1.5 2.5 3.5
1 1.035 1.115 4.749b 5.007“
2 -0.396 -2000“ 4.006“ 0.619

3 4.511 -3577“ 4.127b
4 4.400“ 2.546“ 0.052 4.702“
5 0.517 0.575 -3722“ 2.247“
6 -5173“ 4118“ 10184" 11.835
7 -26.375 ° 2.780 ° 48.414 “
8 3210“ -2.041 -5433“ 43.603“
9 0.126 2.357c
10 4.107b 1.510 6.838“ 15.043“
Overall 0.355 4.119b 0.564 -0.739
Wllcoxonp 0.516 0.037 0.496 0.766

 

Note: For each firm the difference between subperiod average firm limit buy
order return and unconditional market order return forms the basic observation.

3 Significant at the 1 percent level.

b Significant at the 5 percent level.
° Significant at the 10 percent level.

191

 

Table 54
Unconditional and Conditional Returns for Market and Limit Orders, Investment

 

 

Wlndow of 7 Days
Test

0.5 1.5 2.5 3.5
Discount 0.773 2.302 3.859 5.347
Fraction 87.92 63.37 47.61 53.77
Executed
Limit Orders
All 0710 -0.555 -0.930 4.667"

Executed -0.402 0.699 -2.7493 -3.749a
Unexecuted -1.109 -4.736a 0.968 3.0563

Market Orders
All 4.300“ 4.300“ 4.436“ 4007“

 

Note: The first row reports average percentage discount from the equilibrium
price at the close of the day preceding the order submission day used to
determine limit buy order prices. The second row reports the fraction of limit buy
orders that are naturally executed during the trading window. Rows 3-5 report
unconditional, conditional on natural execution, and conditional on forced
execution overall returns for limit buy orders. Row 6 reports the unconditional
overall returns for market orders. The investment window is equal to 7 days in
this table.

3 Significant at the 1 percent level.

b Significant at the 5 percent level.
c Significant at the 10 percent level.

192

Table 55

Differential Returns, Investment Wlndow of 7 Days

Part A: Unconditional

 

 

 

 

 

Test
Subperiod 0.5 1.5 2.5 3.5
1 0.130 0.655 0.621 -3597“
2 1.957“ 2.262“ -2113“ 0.912
3 2.528“ 2.178“ 0.478 9.683“
4 2.442“ 2.450“ -0.076 -2.452“
5 2.911“ 2.256“ 3.706“ 2.332“
6 0.340 1.188“ 2.290“ 1.742“
7 4.936“ -3.810“ -0357 3812“
8 4.184“ -5111“ 3.772“ 5.904“
9 1.844“ 3.154“ -0.038 7.782“
10 2.879“ 2.174“ -3309“ 4.843“
Overall 0.590“ 0.745“ 0.506“ 2.340“
Vlfilcoxonp 0.001 <.0001 0.389 0.005
Part B: Conditional on Execution
Test
Subperiod 0.5 1.5 2.5 3.5
1 0.214 5.549“ 4.925 -2240
2 0.918 1.067 4.528“ 8.598“
3 2.705“ 5.363“ 9.898“ 7.191“
4 2.401“ 4.050“ -2500 -5.586°
5 4.300“ 7.179“ 0.867 0.647
6 0.340 4.033“ 1.546 1.711“
7 4.049“ -2.620“‘ 5.938“ -0.728
8 1.295 5.763“ 11.912“ -0.085
9 1.844“ 3.251“ -2606“ 11.372“
10 3.626“ 5.062“ -8511“ 0.926
Overall 1.358“ 3.604“ 0.951“ 2.415“
Wllcoxonp <.0001 <.0001 0.021 <.0001

 

193

 

Table 55-continued

Part C: Conditional on Nonexecution

 

 

Test
Subperiod 0.5 1.5 2.5 3.5
1 -3.639 40.551 “ 2.024 “ 4.474 “
2 6.521 “ 5.519 “ 4 .412 4.672 “
3 1.598 -0324 -3.987 “ 12.175 “
4 0.416 0.538 3.508 2.546
5 -3427 “ 4 .568 6.229 “ 4.204 “
6 -8.368 “ 10.361 0.300
7 21.023“ 47.904“ 49.922“ 46.764“
8 -8.343 “ -8.792 “ 0.381 11.290 “
9 0.220 1.613 “ -5271 “
10 46.518 -0.936 4.297 9.338“
Overall -3.638 “ 3490 “ -0.088 3.583 “
Wllcoxon p 0.000 0.000 0.001 <.0001

 

Note: For each firm the difference between subperiod average firm limit buy
order return and unconditional market order return forms the basic observation.

3 Significant at the 1 percent level.

b Significant at the 5 percent level.
° Significant at the 10 percent level.

194

Table 56
Market-Adjusted Differential Limit Buy Order Returns, Investment Wlndow of 5 and 7

 

 

 

 

 

 

 

 

 

Days
PartA:
Test
0.5 1.5 2.5 3.5
Subperiod m 112 Th 02 m 112 m 112
1 3.430 -3.418 -0.867 1.233 2.519“ 4.384“ 4.359 1.649
2 2.803“ 3.616“ -0.955 2.531“ -0.283 0.806 2.441“ 4.217“
3 -5.411“ 6.501“ 2.843“ 6.701“ 2.841“ 4.572“
4 -0574 1.241 -0.610 1.910“ 0.050 -0.006 -0.538 0.617
5 -0.879 1.337 -0.556 1.323 -0.920“ 2.910“ -0.079 4.421
6 -3.505 4.003 4.682 2.166 1.762 4.081 16.392“46.678“
7 41.543“ 11.07“ -0.762 0.363 2.620 3.535“
8 -3.217“ 3.734“ 4.444 1.668 4.438“ 3.012“ -0.877 7.299“
9 -0.887 1.783 2.169“ 6.835“
10 4.506 1.865 4.195 2.355 -0040 0.754 0.107 2.189
Overall 2.050“ 2.385“ 4.276“ 1.887“ 0998“ 1.848“ 4.010“ 1.929“
Part 8:
Test
0.5 1.5 2.5 3.5
Subperiod nr 02 nr 02 nr 112 m 112
1 -3001 2.272 2.698 2.860 4.176 0.393 2.733“ 2.583
2 -0.318 0.662 -0.272 0.579 4.165 1.848 2.658“ 7.292“
3 -0.149 0.665 2.585“ 6.905“4.316 2.903 1.389 -3.417
4 1.716 4.195 2.156 2.591 4.279 2.371 4.720 3.751
5 4.636 2.154 4.456“ 3.904“ 0.363 0.202 -0.814 1.436
6 1.670 4.862 3.877 -3.845 2.090 2.148
7 44.812“ 15.439“40.127“ 11.075“ 4.581“ 7.326“43.456“ 15.643“
8 4.897 2.455 0.038 2.864 0.350 1.584 -0.146 4.247
9 -6.095“ 6.573“-0.555 1.315 -5.419“ 7.115“
10 48.464“ 19.681“ 2.858“ 6.422“-0.161 0.588 1.800 4.255
Overall 4.725“ 1.964“ -0.812 1.287“-0.639 1.359“ 4.300“ 2.442“

 

 

Note: For each firm the difference between window limit order return and average
market order return during the corresponding subperiod was regressed on a constant
and the difference between portfolio window limit order return and portfolio average
market order return during the corresponding subperiod. The residuals from these 30
regressions were stacked to form a vector and were then regressed on a constant and
a dummy that takes on the value 1 for natural execution, 0 otherwise. n1 is the
intercept, and '12 is the coefficient of execution dummy. Subperiod and overall results
are reported for each limit order test.

3 Significant at the 1 percent level.

b Significant at the 5 percent level.

° Significant at the 10 percent level.

195

Table 57

Details for the Experiment of Simultaneously Submitting Limit Buy and Sell
Orders at :l: k Ticks from the Equilibrium Price

 

 

 

 

2 3
Transaction Inventory Transaction Inventory
Stock Subperiod Sell Buy Min Max Sell Buy Min Max
Akbnk 1 198 201 -10 6 63 65 -6 4
2 315 345 -10 31 86 107 -6 21
3 328 276 -55 1 1 13 80 -35 1
Akcns 1 234 222 -18 12 77 68 -12 7
2 285 316 -7 32 103 126 -5 24
3 289 247 -43 1 96 68 -28 -1
Akgrt 1 147 145 -15 9 45 43 -10 6
2 166 188 -8 27 65 81 -5 18
3 206 161 -45 0 91 62 -29 0
Alark 1 185 173 -16 12 75 64 -14 7
2 187 207 -5 24 73 90 -3 17
3 237 201 -38 1 95 72 -24 -1
Alctl 1 218 217 -9 1 1 66 64 -7 7
2 188 218 -6 32 61 82 -4 22
3 154 132 -24 3 55 41 -15 2
Arclk 1 216 210 ~21 15 63 60 -13 10
2 243 261 -6 24 83 95 -4 17
3 238 215 -24 -1 84 68 -17 -1
Bagfs 1 270 250 -26 9 107 91 -19 6
2 257 281 -16 25 97 113 -11 17
3 185 178 -8 3 63 57 -6 2
Cukel 1 190 186 -10 14 67 64 -7 8
2 175 201 -7 27 70 86 -5 17
3 133 116 -18 -1 50 40 -11 -1
Dohol 1 318 305 -17 21 79 70 -11 14
2 325 336 -6 23 92 101 -4 15
3 353 331 -24 1 97 82 -16 1
Efes 2 280 318 -3 42 94 120 -2 29
3 301 262 -40 1 88 65 -24 -1
Enka 1 155 151 -15 9 47 43 -11 5
2 177 186 -5 31 78 86 -3 22
3 225 190 -36 7 95 70 -25 4
Eregl 1 249 244 -17 13 78 74 -1 1 9
2 250 279 ~10 29 79 97 -6 18
3 195 187 -10 8 67 61 -8 5
Garan 1 248 248 -7 18 68 67 -6 12
2 259 287 -6 30 78 96 -4 20
3 355 291 -65 0 125 81 -44 -1

 

196

 

Table 57-continued

 

 

 

 

2 3
Transaction Inventory Transaction Inventory
Stock Subperiod Sell Buy Min Max Sell Buy Min Max
Hurgz 1 278 256 ~27 7 1 12 93 ~22 5
2 288 305 ~14 18 99 112 ~10 13
3 305 274 ~32 0 97 75 ~22 0
lhlas 1 249 249 ~12 20 89 89 ~8 14
2 246 273 ~7 29 78 93 ~4 18
3 208 202 ~8 4 55 50 -6 2
lsctr 1 232 242 ~2 12 70 78 -1 8
2 297 314 ~5 19 93 105 ~3 13
3 310 279 -35 1 96 76 ~22 1
Kchol 1 217 216 ~8 18 68 68 -4 12
2 233 253 -8 28 77 91 -4 19
3 306 274 ~28 6 81 61 ~20 -1
Migrs 1 143 132 ~17 9 46 38 ~11 6
2 187 196 ~8 18 72 77 ~7 1 1
3 208 165 ~44 1 83 54 ~30 0
Nthol 1 211 207 ~10 15 71 69 -5 9
2 233 264 -1 36 67 88 0 24
3 162 143 ~20 0 60 49 ~12 0
Otosn 1 1 18 126 ~2 18 40 44 -2 1 1
2 144 173 -3 32 51 70 ~2 21
3 184 153 ~32 2 70 50 ~20 1
Petkm 1 252 242 -17 3 89 83 -1 1 2
2 188 214 -1 27 66 83 0 17
3 239 191 ~48 0 100 69 ~31 0
Ptofs 1 295 286 ~18 1 1 95 89 -1 1 7
2 244 249 -5 15 89 94 -3 10
3 298 267 ~32 10 123 98 ~25 6
Sahol 1 186 178 ~16 8 47 42 ~10 5
2 305 325 -8 21 82 96 -5 15
3 350 296 ~56 1 1 13 80 ~34 ~1
Thyao 1 345 326 ~24 5 109 97 ~16 3
2 241 257 ~7 19 86 96 -5 1 1
3 167 153 ~15 6 52 45 -8 4
Toaso 1 239 239 —9 18 79 79 ~5 12
2 262 286 ~3 27 102 120 -2 20
3 224 208 ~21 1 85 75 ~13 1
Tuprs 1 386 370 ~22 14 120 108 ~15 8
2 327 336 ~1 1 19 98 104 -7 12
3 201 173 ~29 0 59 41 ~18 0

 

197

Table 57~continued

 

2 3
Transaction Inventory Transaction Inventory
Stock Subperiod Sell Buy Min Max Sell Buy Min Max

 

 

 

Uzel 1 214 203 ~12 1 1 64 58 -7 8
2 269 284 ~18 24 97 110 ~11 18
3 229 212 ~17 4 84 71 ~13 2
Vestl 1 232 206 -34 10 81 63 ~22 7
2 269 270 -14 13 105 108 -9 10
3 243 208 ~35 0 100 80 -20 0
kank 1 206 201 ~14 10 69 66 -9 6
2 424 446 ~24 22 1 51 167 ~17 16
3 409 355 ~57 O 134 99 ~37 0

 

 

198

REFERENCES

199

 

References

Admati, Anat R., and Paul Pfleiderer, 1988, A theory of intraday patterns:
Volume and price variability, Review of Financial Studies 3, 593-624.

Ahn, Hee-Joon, Charles Q. Cao, and Hyuk Choe, 1996, Tick size, spread, and
volume, Joumal of Financial lntennediation 5, 2-22.

Aitken, Michael, Philip Brown, Christine Buckland, H.Y. lzan, and Terry Walter,
1996, Price clustering on the Australian Stock Exchange, Pacific-Basin Finance
Journal 4, 297-314.

Angel, James J., 1997, Tick size, share prices, and stock Splits, Joumal of
Finance 52, 655-81.

Anshuman, Ravi, and Avner Kalay, 1998, Market making with discrete prices,
Review of Financial Studies 1 1, 81 -1 09.

Arnold, Tom M., and Mark Lipson, 1998, Tick size and limit order execution: An
examination of stock splits, working paper, University of Georgia.

Bacidore, Jeffrey M., 1997, The impact of decimalization on market qualityzan
empirical examination of the Toronto Stock Exchange, Journal of Financial
Intennediation 6, 92-120.

Bagehot, W.,1971, The only game in town, Financial Analyst Journal 27, 12-14.

Ball, Clifford, Walter Torous, and Adrian Tschoegl, 1985, The degree of price
resolution: The case of the gold market, Journal of Futures Markets 5, 29-43.

Bessembinder, Hendrik, 1997, The degree of price resolution and equity trading
costs, Journal of Financial Economics 45, 9-34.

Bhattacharya, Utpal, and Matthew Spiegel, 1990, Anatomy of a market failure:
NYSE trading suspensions (1974-1988), working paper, University of Iowa.

Biais, Bruno, Pierre Hillion, and Chester Spatt, 1995, An empirical analysis of the
limit order book and the order flow in the Paris Bourse, Joumal of Finance 50,
1655-89.

Blume, Lawrence, David Easley, and Maureen O’Hara, 1994, Market statistics
and technical analysis: the role of volume, Joumal of Finance 49, 153-81.

200

 

Booth, G. Geoffrey, Juha-Pekka Kallunki, Ji-Chai Lin, and Teppo Martikainen,
2000, lntemalization and stock price clustering: Finnish evidence, Journal of
lntemational Money and Finance, Forthcoming.

Brown, Sharon, Paul Laux, and Barry Schachter, 1991, On the existence of an
optimal tick size, Review of Futures Markets 10, 50-72.

Chan, K. C., and Chuan-Yang Hwang, 1998, The impact of tick size on market
quality: an empirical investigation of the stock exchange of Hong Kong, working
paper, Hong Kong University of Science and Technology.

Cheung, Yan-Leung, Richard Yan-Ki Ho, Peter Pope, and Paul Draper, 1994,
lntraday stock return volatility: The Hong Kong evidence, Pasific-Basin Finance
Journal 2, 261-76.

Chordia, Tarun, and Avanidhar Subrahmanyam, 1995, Market making, the tick
size, and payment-for-order flow: theory and evidence, Journal of Business 68,
543-75.

Christiano, Lawrence J., and Lars Ljunngist, 1988, Money does Granger-cause
output in the bivariate money-output relation, Journal of Monetary Economics 22,
217-35.

Christie, William G., and Paul H. Schultz, 1994, Why do NASDAQ market
makers avoid odd-eighth qoutes? Jouma/ of Finance 49, 1813-40.

Christie, Wllliam G., Jeffrey H. Harris, and Paul H. Schultz, 1994, Why did
NASDAQ market makers stop avoiding odd-eighth qoutes? Journal of Finance
49, 1841-60.

Clark Peter K., 1973, A subordinated stochastic process model with finite
variance for speculative prices, Econometrica 41, 135-55.

Cohen, Kalman, Steven F. Maier, Robert A. Schwartz, and David K. Whitcomb,
1981, Transaction costs, order placement strategy, and existence of the bid-ask
spread, Joumal of Political Economy 89, 287-305.

Copeland, Thomas E., 1976, A model of asset trading under the assumption of
sequential information arrival, Journal of Finance 31, 1149-68.

Domowitz, lan, 1993a, A taxonomy of automated trade execution systems,
Journal of lntemational Money and Finance 12, 607-31.

201

 

, 1993b, Automating the price discovery processzsome
international comparisons and regulatory implications, Joumal of Financial
Services Research 6, 305-26.

, 1995, Financial market automation and the investment
services directive, in R. A. Schwartz, ed.: Global Equity Markets: Technological,
Competitive and Regulatory Challenges, (Irwin Professional Publishing, Burr
Ridge, lll.).

 

Dutta, Prajit K., and Ananth Madhavan, 1995, Price continuity rules and insider
trading, Jouma/ of Financial and Quantitative Analysis 30, 199-221.

Easley, D., M. O’Hara, 1987, Price, trade size, and information in securities
markets, Journal of Financial Economics 19, 69-90.

Epps, Thomas W., 1975, Security price changes and transaction volumes: theory
and evidence, American Economic Review 65, 586-97.

Epps, Thomas W., and Mary L. Epps, 1976, The stochastic dependence of
security price changes and transaction volumes: implications for the mixtures of
distributions hypothesis, Econometrica 44, 305-21.

Feige, Edgar L., and Douglas K. Pearce, 1979, The causal relationship between
money and income: some caveats for time series analysis, Review of Economics
and Statistics 61, 521-33.

Foster, Douglas F., and S. Viswanathan, 1990, A theory of the interday
variations in volume, variance, and trading costs in securities markets, Review of
Financial Studies 3, 593-624.

, 1993a, The effect of public information and competition on
trading volume and price volatility, Review of Financial Studies 6, 23-56.

 

, 1993b, Variations in trading volume, return volatility, and
trading costs: evidence on recent price formation models, Journal of Finance 48,
187-21 1 .

Foster, Andrew J., 1995, Volume-volatility relationships for crude oil futures
markets, Journal of Futures Markets 15, 929-51.

Gallant, A. Ronald, Peter E. Rossi, and George Tauchen, 1992, Stock prices and
volume, Review of Financial Studies 5, 199-242.

202

Geweke, John, Richard Meese, and Warren Dent, 1983, Comparing alternative
tests of causality in temporal systems: analytic results and experimental
evidence, Journal of Econometrics 21 , 161 ~94.

Glosten, L., and P. Milgrom, 1985, Bid, ask, and transaction prices in a specialist
market with heterogenously informed traders, Jouma/ of Financial Economics 13,
71-100.

Glosten, Lawrence R.,1995, Is the electronic open limit order book inevitable?
Journal of Finance 49, 1127-61.

Godek, Paul E., 1996, Why NASDAQ market makers avoid odd-eighth quotes,
Journal of Financial Economics 41 , 465-74.

Goodhardt, C., and R. Curcio, 1991, The clustering of bid/ask prices and the
spread in the foreign exchange market, discussion paper 110, London School of
Economics.

Granger, C. W. J. and O. Morgenstern, 1963, Spectral analysis of New York
Stock Market prices, Kyklos 16, 1-27.

Granger, C. W. J., 1969, Investigating causal relations by econometric models
and cross-spectral methods, Econometrica 37, 424-38.

Grossman, Sanford, Merton Miller, Kenneth Cone, and Daniel Fischel, 1997,
Clustering and competition in asset markets, Journal of Law and Economics 40,
23-60.

Gwilym, Owain, Andrew Clare, and Stephen Thomas, 1998a, Extreme price
clustering in the London equity index futures and options markets, Journal of
Banking and Finance 22, 1193-1206.

,1998b, Price clustering and bid ask spreads in international
bund futures, lntemational Financial Markets, Institutions & Money 8, 377-91.

 

Gwilym, Owain, David McMillan, and Alan Speight, 1999, The intraday
relationship between volume and volatility in LlFFE futures markets, Applied
Financial Economics 9, 593-604.

Hamao, Y., and J. Hasbrouck, 1995, Securities trading in the absence of
dealers: trades and quotes on the Tokyo Stock Exchange, Review of Financial
Studies 8, 849-78.

Hamon, Jacques, Puneet Handa, Bertrand Jacquillat, and Robert A. Schwartz,
1995, Market structure and the supply of liquidity, in R. A. Schwartz, ed.: Global

203

Equity Markets: Technological, Competitive and Regulatory Challenges, (Irwin
Professional Publishing, Burr Ridge, lll.).

Hamon, Jacques, Bertrand Jacquillat, and Robert A. Schwartz, 1995, A program
to increase the liquidity of shares in the French equity market, in R. A. Schwartz,
ed.: Global Equity Markets: Technological, Competitive and Regulatory
Challenges, (Irwin Professional Publishing, Burr Ridge, lll.).

Hameed, Allaudeen, and Eric Terry, 1998, The effect of tick size on price
clustering and trading volume, Journal of Business Finance and Accounting 25,
849-67.

Hamon, Jacques, Puneet Handa, Bertrand Jacquillat, and Robert A. Schwartz,
1993, The profitability of limit order trading on Paris Stock Exchange, working
paper, New York University.

Handa, Puneet, and Robert A. Schwartz,1994, Limit order trading, Journal of
Finance 51, 1835-61.

, 1996, How best to supply liquidity in an electronic
environment: is disintermediation a desirable route? Journal of Portfolio
Management 22, 44-51.

 

Handa, Puneet, Robert A. Schwartz, and Ashish Tiwari, 1994, Quote setting and
price formation in an order driven market, working paper, University of Iowa and
City University of New York.

, 1998, The ecology of an order-driven market, Journal of
Portfolio Management 24, 47-55.

 

Hanna, Mark, 1978, Security price changes and transaction volumes: additional
. evidence, American Economic Review 68, 692-95.

Harris, Lawrence, 1986, A transaction data study of weekly and intradaily
patterns in stock returns, Joumal of Financial Economics 16, 99-117.

, 1991, Stock Price clustering and discreteness, Review of
Financial Studies 4, 389-415.

 

, 1994, Minimum price variations, discrete bid ask spreads, and
quotation sizes, Review of Financial Studies 7, 149-78.

 

, 1997, Decimalization: a review of the arguments and
evidence, working paper, University of Southern California.

204

 

Harris, Lawrence, and Joel Hasbrouck, 1996, Market vs. limit orders: the
superdot evidence on order submission strategy. Journal of Financial and
Quantitative Analysis 31 , 213-31.

Hasbrouck, Joel, 1991, Measuring the information content of stock trades,
Journal of Finance 46, 179-207.

Haugh, Larry D., 1976, Checking the independence of two covariance stationary
time series: a univariate residual cross-correlation approach, Joumal of the
American Statistical Association 71, 378-85.

Hedvall, Kaj, Jonas Niemeyer, and Gunnar Rosenqvist, 1997, Do buyers and
sellers behave similarly in a limit order book? A high-frequency data examination
of the Finnish stock exchange, Journal of Empirical Finance 4, 279-93.

Ho, Richard Yan-ki, Yan-Leung Cheung, and Daniel W.W. Cheung, 1993,
lntraday prices and trading volume relationship in an emerging Asian market—
Hong Kong, Pasific-Basin Finance Journal 1, 203-14.

Hollifield, Burton, Robert A. Miller, and Patrik Sandas, 1999, An empirical
analysis of a limit order market, working paper, Carnegie Mellon University and
University of Pennsylvania.

Hopewell, Michael H., and Arthur L. Schwartz, 1978, Temporary trading
suspensions in individual NYSE securities, Journal of Finance 33, 1355-73.

Huang, Roger D., and Hans R. Stoll, 1991, Major world equity markets: current
structure and prospects for change (New York University Salamon Center, New
York, NY).

International Finance Corporation, Emerging stock markets factbook 1999,
(International Finance Corporation, Washington, DC).

Jain, Prem C., and Gun Ho Joh, 1988, The dependence between hourly prices
and trading volume, Journal of Financial and Quantitative Analysis 23, 269-83.

Jennings, Robert H., Laura T. Starks, and John C. Fellingham, 1981, An
equilibrium model of asset trading with sequential information arrival, Journal of
Finance 36, 143-61.

Karpoff, Jonathan M.,1986, A theory of trading volume, Journal of Finance 41 ,
1 069-87.

, 1987, The relation between price changes and trading
volume: a survey, Journal of Financial and Quantitative Analysis 22, 109-26.

205

 

Lamoureux, Christopher G., and Wllliam D. Lastrapes, 1990, Heteroskedasticity
in stock return data: volume versus GARCH effects, Joumal of Finance 45, 221-
29.

Lau, Sie Ting, and Thomas H. Mclnish, 1995, Reducing tick size on the stock
exchange of Singapore, Pacific-Basin Finance Journal 3, 485-96.

Lee, Charles M., Mark J. Ready, and Paul J. Seguin,1994, Volume, volatility, and
New York Stock Exchange trading halts, Journal of Finance 49, 183-214.

Mclnish, Thomas H., and Robert A. Wood, 1995, Hidden limit orders on the
NYSE, Journal of Portfolio Management 21, 19-26.

Najand, Mohammad, and Kenneth Yung, 1991, A GARCH examination of the
relationship between volume and price variability in futures markets, Journal of
Futures Markets 11, 613-21.

Niederhoffer, Victor, 1965, Clustering in stock prices, Operations Research 13,
258-65.

, 1966, A new look at clustering in stock prices, Journal of
Business 39, 309-13.

 

O’Hara, Maureen, 1994, Market Microstructure Theory (Blackwell Publishers,
Malden, MA).

Osborne, M. F. M., 1962, Periodic structure in the brownian motion of stock
prices, Operations Research 10, 345-79.

Parlour, Christine A.,1998, Price dynamics in limit order markets, Review of
Financial Studies 11, 789-816.

Pierce, David A., and Larry D. Haugh, 1977, Causality in temporal systems:
characterizations and a survey, Journal of Econometrics 5, 265-93.

Porter, David, and Daniel Weaver, 1997, Tick size and market quality, Financial
Management 26, 5-26.

Sandas, Patrik, 1999, Adverse selection and competitive market making:
empirical evidence from a pure limit order market, working paper, University of
Pennsylvania.

Sims, Christopher A., 1972, Money income and causality, American Economic
Review 62, 540-52.

206

 

Smirlock, Michael, and Laura Starks, 1985, A further examination of stock price
changes and transaction volume, Joumal of Financial Research 8, 217-25.

Stickel, Scott E., and Robert E. Verrecchia, 1994, Evidence that trading volume
sustains stock price changes, Financial Analysts Journal 50, 57-67.

Stock, James H., and Mark W. Watson, 1989, Interpreting the evidence on
money-income causality, Journal of Econometrics 40, 161-81.

Tauchen, George E., and Mark Pitts, 1983, The price variability-volume
relationship on speculative markets, Econometrica 51, 485-505.

Wang, Jiang, 1994, A model of competitive stock trading volume, Journal of
Political Economy 102, 127-68.

Wood, Robert A., Thomas H. Mclnish, and J. Keith Ord, 1985, An investigation of
transactions data for NYSE stocks, Journal of Finance 40, 723-41.

207