View of the Literature on Circuit Breaker Systems and Provide Suggestions for Future Studies on This Issue

UNIVERSITY OF CINCINNATI

______, 20 _____

I,______, hereby submit this as part of the requirements for the degree of:

______in: ______It is entitled: ______

Approved by: ______

A MARKET STABILIZATION MECHANISM – CIRCUIT BREAKER: THEORY AND EVIDENCE

A dissertation submitted to the

Division of Research and Advanced Studies of the University of Cincinnati

in partial fulfillment of the requirements for the degree of

DOCTORATE OF PHILOSOPHY (Ph.D.)

in the Department of Finance of the College of Business Administration

2003

Jr-ming Jimmy Yang

B.A., National Chung Hsing University, 1993 M.B.A., Saint Louis University, 1997

Committee Chair: Professor Yong H. Kim

A MARKET STABILIZATION MECHANISM – CIRCUIT BREAKER: THEORY AND EVIDENCE

ABSTRACT

The term “circuit breaker” originates in electrical engineering to describe a pre-set

switch that shuts down electrical activity in excess of a system’s design capacity. Since

late 1988, the New York Stock Exchange has been imposing circuit breaker systems,

which mandate trading halts for a stipulated period of time if the Dow Jones Industrial

Average moves by more than a certain amount compared to the previous day’s close.

Besides the U.S., many countries in the world have also imposed circuit breaker systems

in an attempt to reduce market volatility. The purpose of this dissertation is to examine

the effectiveness of circuit breaker systems in financial markets.

In the first chapter, I conduct a thorough review of the literature on circuit breaker systems and provide suggestions for future studies on this issue. The review covers theoretical background, empirical evidence from both stock markets and futures markets, and the related research methodology. The results of an in-depth analysis of current circuit breaker systems in the world are presented in this chapter.

There are two different types of circuit breakers: trading halts and price limits. My

second chapter is designed to test the performance of price limits empirically using initial

public offering (IPO) data. I compare IPOs with their industry-and-size matched seasoned

equities to test three hypothesis raised by price-limit opponents. My results represent the

performance of price limits for IPOs and can be used to predict the performance of price limits during periods with high information asymmetry. The most popular rationale for imposing price limits is to reduce market overreaction and volatility. To date, the empirical literature does not give a clear answer on whether price limits reduce or induce overreaction. Therefore, I examine trade-to- trade data in an effort to provide insight to the ongoing debate over the relation between price limits and overreaction in chapter three. I test two hypotheses to investigate whether price limits reduce or induce overreaction. Overall, I conclude that price limits induce overreaction when the price is approaching the limit, but they also reduce overreaction when prices hit the limit consecutively.

This study has benefited from the advice and support of many people. First of all,

I gratefully acknowledge Professor Yong H. Kim, my dissertation committee chair, for his guidance, support, and encouragement. He merits thanks words cannot convey. I would also like to express my appreciation to the other members of my dissertation committee, Professors Brian C. Hatch and Martin S. Levy, for their insightful comments and suggestions.

In addition, I have greatly benefited from the discussion with Professors Michael

F. Ferguson and Norman T. Bruvold. Comments from Gina Nicolosi, my colleague at

UC, have improved my dissertation in a very positive way. I also thank my friends,

Vincent Chen and Vincent Lin, for providing data.

Most of all, I wish to thank my parents and my wife, Ting-yu, for their support and patience through the entire period of my study. This study would not have been possible without the understanding and cooperation of my family.

TABLE OF CONTENTS

List of Tables 4

List of Figures 6

Chapter I: A Critical Analysis of the Circuit Breaker Systems: A Worldwide Investigation

I.1. Introduction 7

I.2. Price limits 8 I.2.A. Brief history and current rules 8 I.2.B. Theoretical background 9 I.2.B.1. Rationales for price limits 9 I.2.B.2. Costs of price limits 11 I.2.B.3. Optimal price limits 13 I.2.C. Empirical evidence 14 I.2.C.1. Futures markets 15 I.2.C.2. Stock markets 19 I.2.D. Research methodology 25

I.3. Firm-specific trading halts 29 I.3.A. Different types of trading halts 29 I.3.B. Benefits and costs 30 I.3.C. Theoretical models and prediction 32 I.3.D. Empirical evidence 34 I.3.D.1. SEC-initiated trading halts 35 I.3.D.2. Exchange trading halts 36

I.4. Market-wide circuit breakers 39 I.4.A. NYSE circuit breakers 39 I.4.B. Theoretical model and prediction 40 I.4.C. Empirical evidence 41

I.5. Comparison between different circuit breakers 45

I.6. Conclusion 47

References I 49

Chapter II: The Performance of Price Limits under Asymmetric Information: Evidence from Taiwanese IPOs

II.1. Introduction 60

II.2. Institutional background 64 II.2.A. Taiwan Stock Exchange 64 II.2.B. IPOs in Taiwan 65

II.3. Literature review 67

II.4. Hypotheses 70

II.5. Data and methodology 72 II.5.A. Data description 72 II.5.B. Methodology 77 II.5.B.1. Information asymmetry hypothesis test 78 II.5.B.2. Trading interference hypothesis test 79 II.5.B.3. Delayed price discovery hypothesis test 80

II.6. Results 82 II.6.A. Information asymmetry hypothesis 82 II.6.B. Trading interference hypothesis 84 II.6.C. Delayed price discovery hypothesis 86 II.6.D. Summary results 88

II.7. Conclusion 89

Appendix 92

References II 94

Chapter III: Price Limits and Overreaction

III.1. Introduction 109

III.2. Institutional Background 112

III.3. Literature Review 114

III.4. Hypotheses 118 III.4.A. Cooling-off hypothesis (reduce overreaction) 118

2 III.4.B. Magnet hypothesis (induce overreaction) 119

III.5. Data and Methodology 120 III.5.A. Data description 120 III.5.B. Methodology 121 III.5.B.1. Cooling-off hypothesis 122 III.5.B.2. Magnet hypothesis 127

III.6. Results 129 III.6.A. Summary Statistics 129 III.6.B. Cooling-off hypothesis 131 III.6.B.1. Closing limit hits 131 III.6.B.2. Single limit hits 132 III.6.B.3. Consecutive limit hits 135 III.6.B.4. Regression analysis 136 III.6.C. Magnet hypothesis 138 III.6.C.1. Return autocorrelations 138 III.6.C.2. Trading volume 141 III.6.C.3. Relative spreads 142

III.7. Conclusion 143

References III 145

3 LIST OF TABLES

Chapter I: A Critical Analysis of the Circuit Breaker Systems: A Worldwide Investigation

I Types of Circuit Breakers 54

II Price limits in futures markets 55

III Price limits in stock markets 56

IV Circuit Breaker System on the NYSE 57

V Classification of Circuit-Breaker Studies 58

Chapter II: The Performance of Price Limits under Asymmetric Information: Evidence from Taiwanese IPOs

I Summary Statistics of IPOs 99

II Number of Delayed Days 100

III Market Capitalization 101

IV Volatility Analysis 102

V Trading Activity Analysis 104

VI Price Discovery Analysis 106

VII Abnormal Return Analysis 107

VIII Summary Table 108

Chapter III: Price Limits and Overreaction

I Price-Limit Rates and Tick Size 147

II Summary Statistics 148

4 III Cooling-off hypothesis: Closing Limit Hits 149

IV Cooling-off hypothesis: Single Limit Hits 150

V Cooling-off hypothesis: Consecutive Limit Hits 152

VI Regression Analysis 154

VII Magnet hypothesis: Return Autocorrelations 155

VIII Magnet hypothesis: Trading Volume 157

IX Magnet hypothesis: Relative spreads 158

5 LIST OF FIGURES

Chapter II: The Performance of Price Limits under Asymmetric Information: Evidence from Taiwanese IPOs

Figure 1: Timeline 97

Figure 2: Distribution of Delayed Days for a sample of 294 IPOs from October 11, 1989 to July 31, 2000 98

6 Chapter I: A Critical Analysis of the Circuit Breaker Systems: A Worldwide Investigation

I.1. Introduction

The term “circuit breaker” originates in electrical engineering to describe a pre-set switch that shuts down electrical activity in excess of a system’s design capacity. After the stock market crash of October 1987, the Brady Report (1988) and several academic researchers suggested the imposition of “circuit breakers” to prevent the market from fluctuating excessively. Many countries in the world have also imposed the circuit breaker systems in an attempt to reduce excessive market volatility. The activation levels of the circuit breakers, which vary from country to country and from market to market, reflect an ex ante decision on the capacity of the financial markets as perceived by securities regulators. It has been more than a decade since the 1987 crash, but whether the circuit breaker systems are effective or not is still an ongoing debate. Harris (1998) contends that little is known about how circuit breakers affect markets and market participants’ behavior. The purpose of this paper is to conduct a thorough review of the literature on circuit breaker systems and provide suggestions for future studies on this issue.

Exchanges around the world have implemented a variety of circuit breakers for halting trades under certain pre-specified situations. Basically, circuit breakers can be categorized into three different types: 1) price limits 2) firm-specific trading halts and 3) market-wide circuit breakers. Price limits are implemented by futures and stock markets.

By definition, price limits are boundaries set by market regulators to restrict daily security price changes in an effort to curb excess volatility. That is, trading is permitted

7 only at prices within a pre-specified range. If traders are unwilling to negotiate prices within the limited range, trading will stop. Trading can resume anytime traders are willing to trade at a price within the range. Firm-specific trading halts stop trading on individual company securities. These trading halts are usually called by exchanges or security regulators. Trading resumes after some predetermined time interval or based on the discretion of the initiators. Market-wide circuit breakers halt trading on the whole market for a pre-specified duration when the designated index reaches the pre-specified level. The well-known example is the NYSE circuit breakers. Table I provides the organization chart of each type of circuit breakers.

Like any other regulations, circuit breakers have proponents and opponents. In this survey, we analyze the benefits and costs of each type of circuit breaker based on their proponents’ and opponents’ point of view, provide existing theoretical models and predictions related to each type of circuit breakers, and present findings from empirical studies that address issues related to circuit breakers. More importantly, we synthesize the existing studies and offer directions for future research in this area.

In the next section, we start with price limits. Section II focuses on firm-specific trading halts. In Section III, market-wide circuit breakers are the theme. Section IV presents studies that compare different types of circuit breakers. Conclusions appear in the final section.

I.2. Price limits

I.2.A. Brief history and current rules

According to Moser (1990), price limits were first used in rice futures trading at the Dojima exchange in Japan during the early eighteenth century in response to excess volatility. The imposition of the rule came during a time when rice markets were deteriorating. The first instance of a price limit rule in the U.S. came during the First

World War. The U.S. government requested a price limit on the cotton futures contract in August 1917 due to the excess volatility caused by the War. The Chicago Board of

Trade formally adopted price limit rules in 1925. In 1982, futures contracts on stock indexes were introduced with price limits. Table II provides the current price limit rules of the futures markets in the U.S.

Besides the futures markets, price limits also exist in stock markets in many countries in the world. Table III provides the current price limit rules in stock markets.

The levels of price limits vary from country to country, being as tight as 2% and as wide as 60%. Some price limits are based on a fix percentage, while others are based on the price level. Some countries impose different price limits on different groups of stocks.

Price limit rules are also changing over time in many countries. For example, Thailand and Korea widened their price limits, while Taiwan tightened its price limits several times during periods with special events.

I.2.B. Theoretical background

I.2.B.1. Rationales for price limits

9 Several informal rationales have been provided for the existence of price limits in futures markets. First, price limits exist to prevent large movements in prices due to panic and speculation. Second, the daily price limit rule limits the daily liability of market participants and their consequent costs of portfolio adjustment. This rationale is particularly related to the practice of daily settlement in futures markets. Third, price limits give brokers time to consult with their clients at times of market turbulence. Two more formal explanations have been provided by Brennan (1986) and Kodres and

O’Brien (1994).

Brennan (1986) develops a theory of price limits which is consistent with the hypothesis of efficient contract design in futures markets. He shows that price limits may act as a partial substitute for margin requirements in ensuring contract performance without resorting to costly litigation. This explains why the limit is set on a daily basis and why it is based on the price change since the close of the previous day. Given the belief that margin requirements are costly for some market participants, it may be optimal to run the risk of a trading interruption by the imposition of price limits to reduce the margin requirement. However, the implementation of a daily price limit imposes clear costs on market participants by prohibiting trades at prices outside the limits. He predicts that price limits will eventually disappear in most markets. The fact that price limits still exist casts some doubt on the predictive power of the model.

Kodres and O’Brien (1994) develop a model to examine the welfare effects of price limits using a four-stage futures market with risk-averse investors. Investors wish to trade futures contracts either to hedge their endowments of an underlying asset or for purely speculative reasons. They distinguish two types of implementation risk, namely,

10 initiation risk and transactional risk. Initiation risk concerns shocks to the underlying value of the asset that lead to price adjustments between the time an investor decides to place an order and the time the order is submitted. Transactional risk concerns shocks that lead to price adjustments between the time orders are placed and the time they are executed. Initiation risk exists because monitoring the market is costly and current technology does not allow orders to be placed instantaneously. Technological limitations give rise to transactional risk as well as initiation risk. Based on those risks, the model shows that price limits may promote better risk sharing than unconstrained trades when price fluctuations are driven by news about fundamentals.

I.2.B.2. Costs of price limits

Fama (1989) argues that rational prices are not necessarily less volatile prices, and less volatile prices are not necessarily better than more volatile prices. That is, as long as the price volatility comes from a rational response to changes in fundamental values, high volatility per se is not necessarily a bad thing for the economy. Since price limts are imposed to reduce noise or unnecessary volatility in prices during periods of high volatility, the fundamental question is how to identify the “unnecessary volatility”. He also points out that price limits serve no purpose other than to delay price discovery.

Even though price limits can stop the price of a share from falling or rising at the limit on a given trading day, the price will continue to move in the direction towards equilibrium as new trading limits are established in subsequent trading day(s).

11 Lehmann (1989) also suggests that imbalances in supply and demand in trading induce prices to reach their limits, which implies a transfer of transactions to subsequent days. Therefore, price limits may cause volatility to spread out over a longer period of time because limits prevent large one-day price changes and immediate price corrections.

This is often referred to the volatility spillover hypothesis, which implies that price limits cause higher volatility levels on subsequent day(s). Besides, he also argues that blocking trading induces prices to reach their limits and impatient investors will trade at unfavorable prices and patient investors will wait till prices reach the equilibrium level.

In either situation, the trading volume will be higher on days following limit hits. This is consistent with the trading interference hypothesis.

Contrary to Brennan (1986), Telser (1981) argues that price limits should have no effect on margin levels. Price limits merely delay the equilibrium prices from prevailing in the market. Such delay do not lower the default risk of individual traders, so exchanges and brokers will still demand a margin large enough to protect themselves from default risk as if price limits were not imposed.

Chou, Lin, and Yu (2000) extend Brennan’s (1986) model to take into account the spillover of unrealized residual shocks due to price limits. Under the assumption that price limits do not affect the underlying generating process for the equilibrium futures prices, they show that trades following a limit move will reflect unrealized shocks carried over from previous trading periods. Therefore, the losing party of the futures contract will have more incentive to renege. Brennan ignores the additional liquidity costs and reneging costs exist in the following periods, so the effect of price limits may have been overstated. In fact, Chou et al. (2000) show that when traders receive additional

12 information about the equilibrium price, the optimal margin remains unchanged with or without the imposition of price limits. Apparently, this contradicts with Brennan’s result that price limits may help reduce the default risk and lower the margin requirements.

Following the introduction of informed traders in Anshuman and Subrahmanyam

(1999), Kim and Sweeney (2001) model the effects of price limits on the behavior of informed investors and focus on the consequences of limits for price revelation and resource allocation. Their model shows how price limits may induce an informed investor to shift part or all of her profit-motivated trades until the next day and thus delay the spread of information. The informed investor’s strategy depends on the distance from the current price to the limit price. If the current price is far from the price limit, the investor initiates her trading program today. In contrast, if the current price is close to the price limit and the equilibrium price is substantially beyond the limit, the model predicts that the trader may delay her trading activities to tomorrow. However, since the model assumes that information leaks to the market over time even if the informed investor does nothing, it is not clear why the informed investor would forgo the profit opportunity today.

I.2.B.3. Optimal price limits

Ackert and Hunter (1994) test a simple descriptive model of optimal futures price limits under the assumption that futures exchanges minimize the long-run average cost associated with market making. Costs associated with price limits include increased operating costs when limits are set liberally, and illiquidity costs incurred during

13 excessive limit moves. Benefits of price limits include the ability of price limits to prevent excessive speculation, give traders time to arrange for financing or to acquire additional margin, and limit overreaction to news. The optimal daily price limits capture the tradeoff between the costs and benefits. Although empirical evidence from the

Chicago Board of Trade supports the idea, no formal models are developed in this study.

Both Brennan (1986) and Kodres and O’Brien (1994) consider scenarios in which market participants are symmetrically informed. However, it is reasonable to suppose that some market participants are better informed than others. Anshuman and

Subrahmanyam (1999) use a setting with asymmetric information to examine the impact of price limits on information acquisition, price efficiency, and market liquidity. This is the first paper to consider the effects of price limits in a theoretical framework where agents are heterogeneously informed. With price limits, privately informed traders realize that extreme values of information have limited profit potential because no trading can take place outside the limit bounds. This serves to discourage information acquisition. Because informed traders trade less aggressively as the limit is tightened, the bid-ask spread decreases in the tightness of the limit. With the trade-offs between the bid-ask spread and the quality of information acquired, an objective function is set and an optimal limit is determined. They find that price limits are negatively related to the profit potential of agents with superior information.

I.2.C. Empirical evidence

14 Since price limits are imposed in both the futures markets and stock markets, we present the empirical findings based on the markets examined. Results from futures markets cannot be generalized to the stock markets because of the different characteristics between them. For example, the difference in the degree of margin requirement, the mark-to-market feature in futures markets, and the market makers’ obligation to provide liquidity on NYSE may affect investors’ trading behavior between the futures markets and the stock market.

I.2.C.1. Futures markets

Ma, Rao, and Sears (1989) examine the impact of hitting price limits on the return and volume behavior of four futures contracts. They find that following a limit move, prices tend to stabilize or even reverse direction, volatility of prices decreases, and volume of trade remains unchanged. Therefore, they conclude that price limits cool off market reaction without imposing any substantial cost. However, it is argued that volatility is likely to be lower after a day of unusually high volatility. The reduced volatility may simply reflect the mean reversion. Besides, Miller (1989) points out the selection bias in this study. Price limit moves can occur for several days in a row. He argues that there appears to be no solution as to whether keeps multiple-move cases or throws them out of the sample. The best method would be to construct the unconditional distribution of prices, volume and volatility over intervals longer than one trading day, e.g., a week or a month.

15 Kao and Ma (1992) examine the time series property of four actively traded currency futures: the British pound, Canadian dollar, German mark, and Swiss franc.

Opponents of price limits suggest that price limits delay the price adjustment process and thus create artificial price dependence. However, they find significant short-term, and not long-term, price dependence regardless whether or not price limit rules are in effect.

This seems to indicate that price limits are not the cause of short-term dependence and rejects the delayed price discovery hypothesis of price limits.

Arak and Cook (1997) examine the price behavior in the U.S. Treasury bond futures market from 1980 to 1987 in an effort to test whether the behavior is affected by proximity to a price limit. In other words, they try to test whether price limits have a magnet effect or a calming effect. By analyzing the morning behavior after large overnight price moves, they find that the proximity to the limit tends to cause a small price reversal, which supports the calming effect. Besides, their analysis also shows that the observed price reversal is caused by the price limit per se, not a correction of market overreaction.

Berkman and Steenbeek (1998) examine the price formation of the Nikkei 225 stock index futures contract traded on both the Osaka Securities Exchange (OSE), a market with strict price limits, and the more lenient Singapore International Monetary

Exchange (SIMEX) to test the gravitation effect of price limits. Using the Nikkei futures contract on SIMEX as a benchmark, they find that Nikkei futures on the OSE do not trade at a relatively low (high) price near the lower (upper) price limit. That is, no gravitation effect is found. However, they find that when the actual price moves closer to the price limit, the trading volume and price volatility on the OSE are relatively smaller

16 than those on the SIMEX, which is consistent with the “satellite market” model presented by Subrahmanyam (1994).

Chen (1998) investigates 19 futures contracts to test the overreaction hypothesis.

The author finds little evidence to support the hypothesis. The direction of price movements on the next day after a big price swing is generally unpredictable so that overreaction is not a good rationale for imposing price limit rules in futures markets. By arguing that futures prices are extremely noisy in the opening and closing minute, Chen uses the difference between the closing price on the event day and the average of the opening, closing, daily high and low prices on the next day to measure overreaction.

However, Chen points out that transaction data appears to be superior to the average daily price in measuring overreaction.

Park (2000) studies four agricultural futures contracts from January 2, 1986 to

July 20, 1998 to investigate whether price limits can moderate volatility. He shows that price limits have a significant impact on price changes and volatility on the trading day after a limit-lock day. The data also indicate price continuations after an up-limit day for corn and soybean futures. In addition, price limits appear to influence price volatility for some but not all of the four contracts. However, since the findings vary across different futures contracts, it is likely that limits do not directly impact price volatility.

Hall and Kofman (2001) make an attempt to identify the impact of futures price limits on traders’ expectations and price discovery. Basically, they are testing the gravitation effect versus the stabilization hypothesis of price limits. Using five agricultural futures contracts traded at the Chicago Board of Trade in 1988, they find evidence consistent with the stabilizing hypothesis from corn futures. However,

17 conclusive evidence for price stabilization is unable to be found from soybean and wheat futures. The gravitation hypothesis is rejected. They conclude that there is either no impact of price limits on traders’ expectations or there is some stabilization impact.

Chen (2002) tests whether price limits have any effect on margin requirement by examining 5 futures contracts during the period from 1980 to 1994. Based on Brennan

(1986), the null hypothesis states that the presence of price limits reduces margin levels in futures markets. A negative relation between margin and price limits is observed. It is shown that the margin is smaller when price limits are imposed than when they are not present. This finding supports the theoretical prediction by Brennan (1986).

In sum, the existing empirical studies on futures markets try to answer four questions: 1) do price limits moderate volatility 2) do price limits counter overreaction 3) do price limits act as substitutes for margin requirements and 4) do price limits have the magnet effect? For the first question, the results are mixed and inconsistent across different contracts. Some find evidence that price limits moderate volatility, but the reduced volatility after limit moves may simply reflect the mean reversion. For the second question, no direct evidence is observed. Even though the answer seems to be yes to the third question, the result is based on only one study. As to the magnet effect, two studies reject this hypothesis.

Given the mixed and inconsistent empirical evidence on the performance of price limits, future studies are necessary to answer these four questions. Since there is only one existing study examining the relation between price limits and margin requirements, more evidence is required from future research. Besides, most studies are based on the

U.S. markets. International evidence is also encouraged.

I.2.C.2. Stock markets

Chung (1991) examines the price limit system in the Korean stock market and finds that there is no evidence that restrictive price limits decrease the volatility of stock prices. Even though the observed volatility in the Korean stock market is much higher than that in the NYSE, attributing the difference to the ineffectiveness of price limit system appears to be far from conclusive. Besides, the problem of measuring volatility when limit hits exist was not addressed in Chung’s study.

Chen (1993) examines the effect of varying daily price limits on stock price volatility in the Taiwan Stock Exchange. By comparing the stock volatility over three different price limit regimes, he finds that price limits do not provide a cooling-off effect on stock volatility. It is shown that price limits prolong price discovery as evidenced in longer serial correlations.

George and Hwang (1995) compare the volatility of open-to-open returns and close-to-close returns of Tokyo Stock Exchange stocks and find that volatility at the open is greater than volatility at the close only for the most actively traded stocks. Since price limit rules exist in the Tokyo Stock Exchange, they conclude that the price limit rules have a significant impact on the dynamics of security prices. More specifically, the inability of prices to adjust to satisfy a large imbalance would cause the closing price to reflect only partially the security’s true value. This implies that return continuations at the close should be associated with large order imbalances.

19 Lee and Kim (1995) investigate the effect of price limits on stock price volatility using the daily stock price data of Korea Stock Exchange from 1980 to 1989 and find that price limits serve to reduce stock price volatility. They compare the return volatilities between high price limit portfolio (HPLP) and low price limit portfolio (LPLP) using both the original return data and the residual return data after controlling for price limit rates. The difference of volatilities between those two portfolios is significant using the original return data, but it is not significant using the residual return data. Thus, it is concluded that the difference is caused by price limits. Because the volatility of HPLP is significantly higher than that of LPLP using the original return data, the price limits were believed to serve to reduce stock price volatility. However, the multiple limit-hits issue is ignored in their study.

In Japan, the opening price of an IPO was subject to the price limits imposed on all stock listed prior to May 1988. The IPO shares were not traded until the price was determined to be within the range of allowed prices. Share prices were allowed to rise by limited increases each day until demand equaled supply. Due to the severe underpricing, shares of the IPO issuing firm would not be allowed to trade for a few days after the offer date. However, price limits that delayed opening of trading were removed after May

1988. Pettway and Kaneko (1996) investigate whether this change has impact on the level of initial returns. They find that the changes that removed price limits and introduced public auctions reduced the level of initial returns significantly and conclude that public policy can reduce, but not eliminate underpricing. However, the net impact of the removal of price limits is not clear.

20 Kim and Rhee (1997) examine the daily stock price data of the Tokyo Stock

Exchange from 1989 to 1992 to investigate the performance of price limits. Based on the

belief that stocks that reach their price limit are prevented from correcting their order

imbalance, but stocks that almost hit their limit are not, they divide stocks into three

different groups. The first group includes stocks that reach their daily price limit. The

second group includes stocks that experience a price change of at least 90% of the limit, but do not reach the price limit. The third group includes stocks that experience a price change between 80% and 90% of the limit. With the comparison of those three groups, they find evidence that supports the arguments from opponents of price limits, i.e. volatility spillover, trading interference, and delayed price discovery. They then conclude that the price limit system of the Tokyo Stock Exchange may be ineffective.

However, we argue that the formation of the comparison groups in their study is arbitrary given the fact that the price limits range from 5% to 50% or higher in terms of percentage. More importantly, using the daily high or low price, instead of the closing price, as the criterion for selecting limit-hit stocks cannot really capture stocks with actual order imbalance problem. If closing price is less than the daily high or higher than the daily low, the order imbalance problem has already been resolved by the market itself.

The inclusion of those stocks in the first group would exaggerate the degree of order imbalance.

In a market with price limits, the estimated serial correlation and volume relationship may be spurious if price limits are ignored. Shen and Wang (1998) study the daily Taiwanese stock returns for the sample periods from November 1988 to December

1995. They use dummy variables to capture the impact of price limits on autocorrelation.

21 It is found that price limits have a stronger impact on the correlation than the trading volume. The price following a limit move is more likely to be trended, which strengthens the autocorrelation.

Phylaktis, Kavussanos, and Manalis (1999) assess the impact of price limits on stock volatility in the Athens Stock Exchange where price limits were imposed in August

1992. They use the ARCH/GARCH methodology to model the time-series of stock market volatility and a dummy variable to indicate the time period during which price limits are employed. Overall, they find that volatility did not increase or decrease after price limits were adopted. However, like Lee and Kim (1995), this study is subject to the time-varying volatility critique. Results from the before-and-after analysis over a period of several years need to be interpreted cautiously. It is difficult to associate a change in long-run volatility to a specific change in policy.

Kim and Limpaphayom (2000) examine the Taiwan Stock Exchange and the

Stock Exchange of Thailand in an attempt to identify the characteristics of stocks that frequently hit price limits. They find that volatile stocks, actively traded stocks, and small market capitalization stocks hit price limits more often than others. The first and third characteristics are not surprising, but the second one is worth further examination.

Intuitively, actively traded stocks are on average larger in size and thus less volatile. A thorough correlation analysis between these three characteristics would be helpful in identifying the true firm characteristics that are associated with frequent limit hits.

Chan, Kim, and Rhee (2001) use the trade-to-trade data from the Kuala Lumpur

Stock Exchange to examine the impact of a wide price limit on price discovery process.

Because the price limit is set at 30%, they are able to find only 98 limit-hit observations.

22 By comparing the limit-hit sample with a control sample of stocks that also experience a large price change but did not hit their limits, they find that price limits do not improve information asymmetry, delay the arrival of information, and cause order imbalance both prior to and after a limit-hit.

Choi and Lee (2001) examine both the inter-day and intra-day data from the

Korean market to investigate the transitory and asymmetric properties of price limits.

Using variance ratio tests and the modified Kim & Rhee (1997) method, they provide evidence of delayed price discovery due to price limits. They further show that the delayed price discovery and trading interference are transitory because they are resolved once the constraint of price limits is removed at the open on the next day following the limit days. More importantly, they identify the asymmetric feature of price limits by showing that price limits act differently on the upper and lower limit activities. They find that criticisms of price limits are partially supported by upper limit moves while price limits are found effective in the case of lower limit moves. Because of this asymmetric effect, they suggest a price-limit system with an upper limit wider than the lower limit to enhance market efficiency and reduce market volatility. However, the appropriate magnitude of limit rates is left for future studies. While this paper uses intra-day data, its major focus is on the first five transactions after the opening.

Kim (2001) examines the relationship between price limits and stock market volatility. Using data from the Taiwan Stock Exchange during the period from 1975 to

1996, he is able to test whether price limits can moderate volatility because there were six different price limit ranges during this period. The result shows that the stock market is not less volatile when price limits are more restrictive. That is, restrictive price limits do

23 not moderate volatility. However, it should be noted that price limits are set by the regulators. In Taiwan, price limits are adjusted based on market conditions. Usually, they are adjusted downward during periods of important events that are likely to cause dramatic moves in the market. The higher volatility observed during periods with restrictive price limits is thus understandable.

Kim and Sweeney (2001) test the predictions of their model using Taiwan Stock

Exchange data from 1991 to 1994. Results are consistent with the view that informed investors are likely to put off their trading programs if the current price is near the price limit. However, the test is limited to the today-versus-tomorrow and ignores the intraday sequence of trades. Informed investors’ trading behavior is easier to pick up from the intra-day data.

Cho et al. (2003) use the intraday data from Taiwan Stock Exchange to test the magnet effect of price limits. It is the first attempt to test the magnet effect using intraday data. They find a statistically and economically significant tendency for stock prices to accelerate toward the upper bound and weak evidence of acceleration toward the lower bound as the price approaches the limits. That is, the magnet effect is supported.

However, their study is limited to the return generating process and has nothing to say about the informed investors’ behavior. Besides, similar results from different thresholds used for the proximity to the limits cloud the magnet effect. If different levels of proximity to the limits generate similar results, there is no evidence of magnet effect.

24 the magnet effect is supported only if significant results are observed when the price is very close to the limit. Besides, the discarding of limit-hit observations may underestimate the means, standard deviations and correlation coefficients, as pointed out by Chiang, Wei, and Wu (1990).

In sum, the existing empirical studies on stock markets try to answer four questions: 1) do price limits moderate volatility 2) do price limits promote price discovery 3) do price limits affect informed investors’ trading behavior and 4) do price limits have the magnet effect? Many papers examine the impact of price limits on volatility, but no consistent evidence is found. Besides, results from long-term volatility studies are difficult to interpret because the time-varying volatility makes it impossible to associate a change in volatility to a specific change in policy. As to the price discovery, all existing studies find that price limits delay the price discovery process. Only one paper investigates the impact of price limits on informed investors’ behavior. The result seems to suggest that informed investors are likely to put off their trading if the current price is near the price limit. The answer to the last question is yes, but it is based on one study.

Given the lack of consistent results from the stock markets, future studies are necessary to answer these questions. The main focus is on volatility, but the magnet effect and informed investors’ behavior are important issues to explore. Besides, current studies seem to focus on only a small number of countries. It is interesting to see results from countries imposing different price-limit levels and rules.

I.2.D. Research methodology

25 With price limits, the observed prices may differ from the true or equilibrium prices. Chiang, Wei, and Wu (1990) examine the effect of price limits on the risk-return estimations using Taiwanese data during the period of September 1, 1986 to September

10, 1987. They compare the unbiased method with the discarding method in the estimation of means, standard deviations, and correlations. The unbiased method is developed along the lines of a true price-time interval. The true security prices are assumed to be lognormally distributed. Each time interval where consecutive true prices are observed is used to measure the return variance and covariance. The discarding method discards the quoted prices from limit days because they are not equilibrium prices. They show that the mean return estimates from the discarding method are downward biased. The estimates of standard deviations, correlation coefficients, and covariances are also uniformly underestimated for the discarding method. However, since both the covariance and the variance of the market return are biased downward, the biasness of beta is unclear.

Lee and Kim (1997) examine the impacts of price limits on the estimation of the market model parameters. They use open-to-open returns as a proxy for true returns and compare the estimates of the market model parameters based on open-to-open returns with those based on close-to-close returns. They find that price limits affect the stochastic properties of stock returns considerably. The estimates of beta from close-to- close returns are biased and inconsistent. However, it is noted that only betas calculated from returns over short intervals suffered from this bias and inconsistency. The result suggests that the impact of price limits on delaying the adjustment of a security’s price does not last long.

26 Chou (1997) proposes a Baysian approach using the Gibbs sampler to estimate linear regression models in which the dependent variable is subject to price limits. An advantage of this approach over the GMM and maximum likelihood approach is that the statistical properties do not depend on the sample size. Unlike the MLE or GMM estimator that requires the availability of large samples to ensure convergence and to obtain the asymptotic normality of the estimator, this approach deals with the posterior distributions of the parameters directly. However, like most classical maximum likelihood methods, the Gibbs sampling approach relies on the assumption that the distributions of the model errors are normally distributed, an assumption that is common and crucial in Bayesian analysis.

Chung and Gan (2000) demonstrate how price limits can affect a return series on limit-hit days. They distinguish two different kinds of effects on the price behavior, the ceiling effect and cooling-off or heating-up effect (C-H effect). Whether price limits have any effect on stock prices depends on the presence of a C-H effect. The return series is assumed to mixture normal instead of simple iid normal. By examining five randomly selected stocks in the Taiwan Stock Exchange, they conclude that price limits have some cooling-off effect on means of stock returns. However, the effect is less in magnitude when mixture normal is used. Although a simple normal density suggests that price limits reduce variance, the mixture normal density does not find significant results.

That is, the result is sensitive to the density of the return series they assumed.

Wei (2002) proposes a censored-GARCH model for return process of assets subject to price limits and develops a Bayesian approach to this model. Using Treasury bill futures contracts, Wei shows that price limits distort the tail behavior of the return

27 distributions, which causes negative kurtosis of the observed return sample. Price limits should be seriously taken into account in modeling such asset returns. This model indicates that the conditional equilibrium returns of T-bill futures are fat-tailed.

The extreme-value statistical techniques have recently been applied to margin setting in futures markets. However, the question is would the existence of price limits in futures markets affect the application of this technique? Broussard (2001) evaluates the robustness of the extreme-value statistical estimation procedure in the presence of price limit data within a context of setting appropriate futures margin levels. The results indicate that price limits impact the estimation by generating more conservative probability estimates for margin violations. By applying the extreme-value method on two futures contracts, he shows that the technique provides useful information on the probabilities of observing large price changes. That is, the existence of price limits affects the performance of the extreme-value method, but they do not remove its usefulness.

Wei and Chiang (2001) derive a generalized method of moments (GMM) estimator for variances in markets with price limits to overcome the measurement problem of deleting prices on limit days. They compare the GMM estimator with the maximum likelihood (ML) estimator and find that the ML estimator is difficult to use but the GMM estimator is simple and easy to implement. Furthermore, when the normality assumption is violated, the ML estimator becomes biased. Given the evidence that daily returns are non-normal for most financial assets, it is recommended that the GMM estimator be used in real applications.

28 I.3. Firm-specific trading halts

I.3.A. Different types of trading halts

The 1934 Securities Exchange Act granted the SEC the authority to suspend trading in any or all securities in all organized markets. SEC-initiated suspensions are used selectively to stop trading in individual company securities. However, occasionally, trading is suspended in the securities of two or more related companies (e.g., firms involved in mergers and takeovers). Originally, the suspension power applied only to the exchange-traded securities, but the SEC’s power to suspend was extended to include the over-the-counter (OTC) markets in 1964. Since then, the majority of suspensions are

OTC securities.

There are two primary categories of firm-specific trading halts on the NYSE: news halts and order imbalance halts. News-related trading halts are also implemented on the NASDAQ. Exchange officials initiate news halts when an information release is expected to have a significant impact on prices. Order imbalance halts are initiated by the exchange specialist when a large imbalance exists between buy and sell orders.

Based on Howe and Schlarbaum (1986), SEC-initiated suspensions differ from the NYSE trading halts in two ways. First, SEC suspensions are intended to protect investors, while NYSE trading halts are designed to protect the specialists. Second,

NYSE trading halts occur more often than SEC suspensions. SEC suspensions averaged about 78 per year in the period 1964-1978, roughly one-tenth the frequency of NYSE trading halts.

29 I.3.B. Benefits and costs

Trading halts can be used to help protect traders from incurring heavy losses during periods with extreme illiquidity. Another argument in favor of trading halts is that when traders are given an opportunity to cancel orders during extreme market changes, they may be more willing to supply liquidity during normal conditions. One argument against trading halts is that they impede price formation because trading aggregates information that is distributed across market participants. An apparent cost of trading halts is the explicit welfare losses to investors from an inability to trade. Liquidity traders may not be able to buy or sell stocks as needed. Further, managers following hedging strategies may be unable to rebalance portfolios when trading is halted. However, the fact that a variety of impediments to trade persist on many exchanges suggests that the benefits of market closures outweigh the welfare losses of having such impediments to trade in place.

Moser (1990) examines the effects of circuit breakers on the stability of the financial markets. Although no formal models are developed, costs and benefits associated with circuit breakers are thoroughly presented. Circuit breakers are classified into three types based on capacity issues: order-imbalance circuit breakers halt trading when orders to buy or sell threaten the viability of the specialist; volume-triggered circuit breakers halt trading when volume exceeds order-processing capacity; and price-limit circuit breakers halt trading when price changes are regarded as excessive. The purpose of the order-imbalance circuit breakers is to protect specialists from large losses incurred by purchases in declining markets. Because the cost-effectiveness of order processing

30 depends on the level of trading volume, the volume-triggered circuit breakers are designed to protect exchanges from losses arising from additional costs incurred by heavy volume. The price-limit circuit breakers are intended to bring excessive volatility under control. Moser argues that circuit breakers reduce access to markets and thus reduces the ability of markets to resolve needs for liquidity. Price limits extend credit to loss positions in futures and options markets. If the credit balances become too large, the clearinghouse’s guarantees of contract performance may be threatened.

Kyle (1988) points out and evaluates four different arguments for trading halts.

The first argument is that trading halts reduce price volatility by giving traders an opportunity to cool off and think before they act. The problem is that there is no obvious reason to believe that a mandatory trading halt makes stampeding traders to calm down.

A mandatory trading halt might make wary traders even more wary and thus might accelerate a price collapse or increase market volatility. Besides, if the price quickly reaches the level it would have reached in the absence of the trading halt after reopening, the halt has simply suppressed volatility, not reduced it. The second argument is that trading halts allow time for margin and credit mechanisms to work because it takes time to issue margin calls, renegotiate loans, and collect money. This argument suggests that an effective trading halt rule would consist of a price limit proportionate with the minimum margin requirement. Besides, since margin and credit problems are likely to have spillovers into the banking system, the Fed is likely to be a good candidate for the regulatory agency. The third argument is that trading halts prevent trading with incentives distorted by financial distress. Because of the lack of instantaneous clearing, the clearing corporation is unable to monitor the intraday trading by all firms perfectly.

31 When a floor trader faces significant losses that his career is at stake, he has an incentive to take on very large intraday positions in an attempt to speculate and survive from the market. If those distorted incentives exist, the clearing corporation of an exchange has incentives to declare a trading halt on any of its floor-trader customers who might have financial problem. However, for this purpose, it might be more effective to simply impose limits on the sizes of the intraday positions that firms and their customers could bear. The last argument is that trading halts protect customers from illiquid and disorderly markets. During a period of extreme volatility, liquidity may dry up because some market makers are bankrupt and those that remain are trading smaller quantities than usual. In addition, the arbitrage mechanism connecting different markets may not be working well during this period. However, financial libertarian would argue that trading halts should not be called for this reason because a warning issued by exchanges should be sufficient to convey the information that trading conditions are unsafe to potential traders.

I.3.C. Theoretical models and prediction

Greenwald and Stein (1988) provide reasons for recommendations made by the

Presidential Task Force on Market Mechanisms after the October 1987 market crash.

They argue that the Crash was a fundamentally unique event, so analysis should focus on the speed and disorder of the crash rather than on causes of the overall drop in the level of stock prices. During the Crash, price quotes gave investors virtually no information about either the levels at which their trades would actually be executed or the general state of the market demand. Therefore, one important goal of the circuit breaker system

32 should be to stimulate information flows during trading halts. There is nothing wrong with large price movements as long as they are fair. That is, as long as there are no tremendous asymmetries of information between specialists and their customers. Thus, the primary function of a circuit breaker should be to reinform market participants when information transmission breaks down.

Fama (1989) argues that mandatory trading halts increase rather than reduce volatility by inciting trading in anticipation of halts. Besides, he also points out that personal trading halts are available to any investor who wants time to analyze market events. That is, if the rationale for imposing trading halts is to provide investors a

“cooling-off” period to evaluate information, the “home-made” trading halts are readily available.

Greenwald and Stein (1991) develop a pair of models to show that properly designed and implemented circuit breakers may help overcome some of the informational problems and improve the market’s ability to absorb large volume shocks. This paper clarifies the informal arguments made in Greenwald and Stein (1988). It is argued that the breakdown of market making was one of the key characteristics of the 1987 market crash. If the market-making system is failing, value buyer who sees attractive opportunities may be deterred from placing orders because they are unsure as to the prices at which their orders will be executed. This so-called “transactional risk” may justify the use of circuit breakers. If a circuit-breaker mechanism can be designed to improve the information available to market participants at the time they submit their orders, the transactional risk caused by informational imperfection can be reduced and value buyers will be more willing to bring their demands to market. However, in their

33 models, traders use only market orders so limit orders are not considered. If transactional risk is what prevents value buyers from submitting their orders, limit orders, by definition, should mitigate the risk by setting a desirable execution price. Besides the common argument that limit orders carry a risk of no execution, Greenwald and Stein provide a different drawback of limit orders to justify the use of only market orders in their models. They argue that a limit buy order involves giving away a put option on fundamentals because a buy order is more likely to be executed at the limit price when bad news arrives. In other words, the buyer would be better off by submitting a market order rather than a limit order in this case. In reality, major institutional investors, who constitute much of the economy’s risk absorption capacity, do not appear to make much use of limit orders on a regular basis.

Subrahmanyam (1997) explores the ex ante response of strategic informed traders to closure boundaries and thus the ex ante effect of closures on market liquidity. Prior to this paper, most theoretical models assume that all traders are competitive and do not react to the closure strategically. In his model, an informed trader knows that trading large quantities will cause the closure bound to be crossed and make him lose his profit potential. The strategic action would be to scale back his trading in response to the closure. The effect of this strategy is the increasing spread for small quantities and decreasing the spread for large quantities. He suggests that the problem can be mitigated if the closure is discretionary. That is, the closure boundary is not perfectly predictable from the perspective of the informed trader.

I.3.D. Empirical evidence

34 I.3.D.1. SEC-initiated trading halts

Howe and Schlarbaum (1986) examine the price behavior of a sample of SEC- initiated suspensions during 1959-1979. The sample covers 1186 trading suspensions of

1038 different companies. They find that these suspensions are associated with substantial devaluations and nearly 80 percent of the securities experienced negative abnormal returns over the suspension period. On average, suspensions are related to bad news. In addition, longer suspensions are significantly correlated with greater devaluations. More importantly, significant and prolonged negative abnormal returns are observed in the post-suspension period, an apparent violation of semi-strong form of market efficiency.

Ferris, Kumar, and Wolfe (1992) analyze the effect of SEC-ordered trading suspensions during 1963-1987 and find that volatility and volume are higher prior to and after suspensions, but they return to their prior levels at a later date. Basically, they examine the pattern of daily abnormal returns surrounding the suspension for a portfolio of suspended stocks and then compare the variances of suspended securities’ returns across several time periods to determine if a trading suspension is associated with changes in stock volatility and trading volume. The SEC’s definition of unusual trading activity as higher than normal return variance or trading activity is adopted. Based on their analysis, they conclude that trading suspensions are not associated with the immediate elimination of unusual market activity. Whether unusually high trading volumes and return volatility are desirable is not addressed.

35 I.3.D.2. Exchange trading halts

Lee, Ready, and Seguin (1994) examine the New York Stock Exchange and find that trading halts increase, rather than reduce, both volume and volatility. Basically, their results are from the comparison between trading halts and pseudo halts. A pseudo halt is a control period of continuous trading for the same firm, matched on time of day and duration. Higher post-halt volume is observed into the third day while higher post-halt volatility decays within hours. Even though the extent of media coverage is a partial determinant of volume and volatility, the halt effect remains significant after controlling for the media effect. That is, trading halts do not have the calming effect.

Corwin and Lipson (2000) examine the nonpublic NYSE data on all SuperDOT

(electronic) orders to analyze the effects of trading halts on order flow, spreads, and the limit order book. They argue that information on order flow is an important missing element in the ongoing debate on the usefulness of trading halts. Submissions and cancellations of orders during a halt may provide an alternative source of information that may assist with price formation. They find that limit order submissions and cancellations are extremely high during halts and remain high for several hours after the halt. Even though the cancellations are expected, the large increase in submissions suggests that traders are willing to reenter the market before trading resumes. They also find that limit order book depth near the quotes is unusually low before, during, and after trading halts, which suggests that investors are unwilling to supply liquidity during these times. As to the bid-ask spreads, they find that the quoted spread and several measures of limit order book spread are unusually high at the reopen. Consistent with Lee, Ready, and Seguin

(1994), they also find that both volume and volatility increase significantly after NYSE

36 trading halts. The increased volatility observed after trading halts can be partially explained by the decreased depth in the limit order book. The problem is that we do not know what would have occurred in the absence of a halt. They also tried to use the duration of the halt as an additional explanatory variable for the abnormal volatility after the halt, but no evidence was found.

Christie, Corwin, and Harris (2002) study a sample of 714 news-related trading halts on NASDAQ between September 1997 and December 1998 to examine the effects of trading halts and reopening procedures on prices, transaction costs, and trading activity. Even though a similar study has been done on the NYSE by Corwin and Lipson

(2000), this NASDAQ-based research provides further evidence on the effects of trading halts from a market with different structures and halt mechanisms. Unlike specialist markets such as the NYSE, NASDAQ does not use a series of central indicator quotes to convey information and attract order flows during a trading halt. Price discovery for

NASDAQ halts that reopen during normal trading hours consists of a five-minute quotation period prior to the resumption of trading. This quotation period allows dealers to signal their information to other market makers and engage in price discovery through non-binding quotes. However, if a trading halt is lifted after 4:00 p.m., trading reopens the following morning with a 90-minute quotation period. Continuous trading is resumed after the quotation period and does not involve the use of a call auction. The difference in reopening procedures allows authors to directly test whether the extent of information dissemination during the halts affects post-halt market conditions. Their major finding is that inside quoted spreads more than double following halts and volatility increases to more than nine times normal levels for intraday halts that reopen after only a five-minute

37 quotation period. However, for halts that reopen the following morning with a 90-minute quotation period, inside spreads are no wider than on non-halt days and volatility increases are significantly dampened. The result supports the hypothesis that increased information transmission during the halt results in reduced post-halt uncertainty. In fact, the result is consistent with the arguments in Greenwald and Stein (1991) that trading halts can be beneficial if they are used to transmit information during times of unusually high transaction-price uncertainty. Like the NYSE studies by Lee, Ready, and Seguin

(1994) and Corwin and Lipson (2000), the paper also finds that even with information transmission during the halt, post-halt volume and volatility are unusually high following

NASDAQ halts. This consistency in volume and volatility patterns suggests that the response of investors to trading halts is independent of market structure and halt mechanisms. However, the finding may suggest either trading halts result in increased uncertainty or trading halts are called in response to high-expected volatility. It is extremely difficult to identify the actual reasons because we do not know what would happen without the imposition of the trading halts.

Bhattacharya and Spiegel (1998) analyze all trading suspensions that occurred in the NYSE during the period 1974-1988 and reveal that suspensions occur for two main reasons: the firm announces impending news (49.1%) or the market maker observes a severe order imbalance (48.5%). They find that the desire to maintain price continuity is an important motivation to suspend trade. Inventory-imbalance fears are pronounced for large firms. News-related suspensions are related to adverse-selection concerns.

Besides, they also conclude that the ability of the NYSE to absorb unusually large shocks has improved in this time period.

38 While the existing literature on trading halts is based on the U.S. market, Wu

(1998) adds new evidence to the literature by investigating trading suspensions on the

Stock Exchange of Hong Kong (SEHK). Wu evaluates the effectiveness of trading suspensions in the price discovery process, changes in return volatility and changes in trading volume. The SEHK suspensions differ from NYSE trading halts in two ways.

First, the SEHK employs an electronic automated trading system so no market makers are involved during suspensions. Second, in SEHK, a trading suspension is either ordered by a regulatory body or requested by a listed company. In fact, the issuer-initiated suspensions consist of more than 90% of all trading suspensions in Hong Kong. It is found that higher variance and trading volume in the post-suspension period are observed than those in the pre-suspension period on the SEHK, which suggests that suspensions do not immediately ease unusual volatility and trading activity.

In sum, the empirical evidence on trading halts is consistent across different markets and different halting mechanisms. Trading activity and volatility around trading halts are higher than the normal level. That is, trading halts do not calm the halted securities. However, the information transmission during trading halts promotes the price discovery process, which is the intended goal of trading halts. Future studies on trading halts should focus on the informed investors’ trading behavior and provide more international evidence.

I.4. Market-wide circuit breakers

I.4.A. NYSE circuit breakers

The well-known example of market-wide circuit breakers is the NYSE circuit breakers. Following the recommendation by the Brady report, the NYSE imposed the market-wide circuit breakers in October 1988. Table IV shows the detailed rule of the circuit breakers as well as the changes over time. The trigger point is based on the movement of the Dow Jones Industrial Average (DJIA). Initially, it was set at 250 points. On May 1, 1998, the NYSE increased circuit breaker thresholds according to a quarterly adjusted percentage-based system. The threshold point is adjusted based on the

DJIA average closing values of the previous month, rounded to the nearest 50 points. A

10% drop in DJIA will lead to a trading halt for one hour if it occurs before 2:00 p.m., for

30 minutes if occurs between 2:00 and 2:30 p.m., and no trading halt if occurs after 2:30 p.m.

One problem with circuit breaker criticisms is that they are based on models that ignore market imperfections that circuit breakers are designed to address. An important imperfection in asset markets identified by the Brady report is the market’s limited capacity to absorb massive one-sided shocks. Another important issue associated with circuit breakers is the effect of these impediments to trade on the ex ante trading decisions of market participants and consequently on market liquidity and price variability.

I.4.B. Theoretical model and prediction

40 Subrahmanyam (1994) presents a model to show that circuit breakers may yield results that are exactly the opposite of what regulation intended it to accomplish.

Specifically, if the price is very close to the breaker limit and if agents place a high value on their desire to trade, the circuit breaker may actually increase price variability and the probability of the price crossing the circuit breaker bounds because strategic traders may advance their trades to assure their ability to trade. This is usually called magnet or gravitation effect. In a two-market situation, a highly liquid dominant market and a less liquid satellite market, agents with intertemporal needs to trade may migrate out of dominant markets and trade on the satellite market. This phenomenon has the effect of transferring price variability from the dominant market to the satellite market. The model assumes that the circuit breaker is triggered based on extreme price movements in the dominant market, and if triggered, results in trading halts on both markets.

I.4.C. Empirical evidence

Because circuit breakers are rarely triggered, it is difficult to obtain sufficient evidence to evaluate their effectiveness. On October 13, 1989, the Dow Jones Industrial

Average fell 6.9% and trading was halted in several high-capitalization NYSE stocks.

The S&P 500 futures contract fell to its price limit on that day. Kuhn, Kuserk, and Locke

(1991) analyze the cash and stock index futures markets on October 13 and 16, 1989 to examine whether circuit breakers can moderate volatility. There is no evidence that circuit breakers moderate volatility in cash or stock index futures markets. The sharp move on October 13 is in response to fundamental information in the markets. The

41 circuit breakers triggered in the futures market merely postponed the price discovery process.

Gerety and Mulherin (1992) investigate how the daily opening and closing of financial markets affect trading volume by studying NYSE data from 1933-1988. They find that closing volume is positively related to expected overnight volatility, while opening volume is positively related to expected and unexpected volatility from the previous night. Given the results, they conclude that the desire of investors to trade prior to market closings indicates a cost of mandating market-wide circuit breakers. Since circuit breakers on the NYSE has been activated only once in 1997, this study provides a good and simple prediction of the performance of circuit breakers. However, the difference between the regular closing and circuit-breaker closure makes it difficult to apply this finding. For example, circuit-breaker closure is associated with large market movements and high volatility while regular closing is not.

Lauterbach and Ben-Zion (1993) examine the official order imbalance data from the Tel-Aviv Stock Exchange in October 1987 to investigate the net effect of circuit breakers. Surprisingly, they find that sell pressures were concentrated in higher-beta, larger-company, and lower-leverage stocks. One possible explanation is that traders may have concluded that less-solid stocks had a good chance of selling at an inferior price. As to the performance of circuit breakers, they find that circuit breakers reduce the next-day opening order imbalance and the initial price loss, but they have no effect on the long-run response.

42 Goldstein and Kavajecz (2000) examine liquidity provision by limit order traders and floor members during extreme market movements. On October 27, 1997, circuit breakers caused the NYSE to halt trading for the first time in history as the Dow Jones

Industrial Average (DJIA) lost 554 points. However, on the next day, the DJIA increased by 337 points, the largest single-day point gain up to that date. By investigating data on the limit order books and specialists’ quotes, they find that a substantial liquidity drain occurs not on the day of the market-wide trading halts but rather on the following day as limit order traders are reluctant to replace expired day limit orders. The liquidity drain is characterized by wider limit order book spreads as well as diminished depth throughout the limit order book. On the other hand, specialists perform their function as liquidity providers of last resort by maintaining narrow spreads and normal depth, despite the significantly diminished liquidity provision by the limit order traders. They also find that, surprisingly, the high volume stocks showed the most dramatic liquidity drain over this period.

Since circuit breakers are rarely triggered, Ackert, Church, and Jayaraman

(2001a) conduct an experimental study to examine the effects of mandated market closures and temporary halts on market behavior. Even though circuit breakers can take many forms, they focus on the market-wide mandated trading halts triggered by extreme market movements. Using nine experimental asset markets from three different regulatory regimes, namely, market closure, temporary halt, and no interruption, they find that the deviations from the expected price are not affected by the presence of circuit breakers. The main driver of price deviations from fundamental value is information in the market. Their analysis of trading volume indicates that circuit breakers accelerate

43 trading activity when an interruption in trading is imminent. From setting up the experiments, they are able to specify the level of uncertainty, the distribution of information across traders, the fundamental determinants of asset value, and the alternative market structures, which are the advantage of experimental studies over other empirical researches.

Following their previous study, Ackert, Church, and Jayaraman (2001b) conduct experimental asset markets to examine the effects of circuit breakers on market behavior when agents are uncertain about the presence of private information. In particular, they investigate whether circuit breakers dampen or prevent price movements away from fundamental value in periods without private information. They find that circuit breakers perform no useful function in their experimental asset markets. With a temporary halt, price moves away from fundamental value in periods without private information. They propose that the result may occur because a temporary halt provides traders time for introspection, allowing them to dwell on irrelevant or unimportant information. Thus, agents may be more likely to mistakenly infer that others possess private information.

However, they do not find any evidence that circuit breakers affect trading volume in the absence of private information. Similar results are not observed from the case of market closure. They argue that the reason is that traders have fewer opportunities to reverse decisions or act on information.

In sum, there is only limited number of studies examining the performance of market-wide circuit breakers. Of course, the main reason is that data are not available.

The NYSE circuit breakers only triggered once since the implementation in 1988. Even if we can obtain sufficient data to examine the performance of circuit breakers, we do not

44 know what would have happened without circuit breakers. Two experimental studies provide insight to this issue. Experimental designs are beneficial because they can control the level of uncertainty and the distribution of information across traders. Future studies on the market-wide circuit breakers may want to focus on the magnet effect as well as the satellite market model designed by Subrahmanyam (1994).

I.5. Comparison between different circuit breakers

Given that a variety of circuit breakers exist on exchanges around the world, what is the type of halting procedure that is likely to help the exchange to achieve its policy goals? A trading halt can either be rule-based or at the discretion of exchange officials or security regulators. Telser (1981) informally suggests that the rule-based price limits in futures markets are superior to the discretionary trading halts on the NYSE because the former are more predictable.

Subrahmanyam (1995) makes the first attempt to build perspective on the relative desirability of various types of halting procedures. He compares discretionary and rule- based procedures to halt trade and argues that the former can be more effective than the latter by bringing more information into the closure decision. In relation to

Subrahmanyam (1994), he shows that discretionary trading halts may be less susceptible than rule-based halts to the gravitational or magnet effect because the former introduce some degree of randomness into the halting rule and thus are less predictable. He also points out that more theoretical and, more importantly, empirical work is required to

45 draw definitive policy conclusions on the relative desirability of various halting procedures.

Coursey and Dyl (1990) compare the market’s adjustment to significant new information when price limits or trading suspensions are present in laboratory markets.

This is the first experimental study on price limits and trading halts. There are three institutional settings: (1) a completely unregulated market where no controls were imposed; (2) a market where the arrival of new information caused a rule limiting the allowable price change to take effect; and (3) a market where the arrival of new information triggered a trading suspension. By conducting a total of 18 laboratory sessions, six replications of the same experiment in each of the three settings, they find that both price limits and trading suspensions result in a loss of efficiency as compared to the unregulated markets. Besides, the loss of allocative and informational efficiency resulting from price limits is not as great as when trading is suspended. The major drawback of this study is that all traders have the same information in those laboratory markets, which is unlikely in the real world. Thus, the result cannot be generalized to the real market.

Kim, Yagüe, and Yang (2003) compare the relative performance between trading halts and price limits using Spanish market data. According to their empirical evidence, trading activity increases after trading halts and limit hits. Volatility stays at the same level after trading halts, but increases after limit hits. They also demonstrate that the degree of information asymmetry is reduced after trading halts, but is even higher after lower limit hits. For price discovery, it is found that both upper price limits and good-

46 news trading halts cause delayed price discovery and dampen market efficiency. Overall, trading halts seem to perform better than price limits in achieving their intended goals.

In sum, due to the data availability, only one theoretical, one experimental, and one empirical study compare the relative performance between different types of circuit breakers. It is very unfortunate because exchanges, security regulators, and investors deserve more understanding on this issue to help them make their best decision. Future studies on the comparison between circuit breakers are strongly encouraged.

I.6. Conclusion

Are circuit breakers effective in achieving their intended goals? Do the benefits of the impediment to trade outweigh the apparent liquidity cost? This paper reviews existing circuit breaker studies and finds that the answer to these two questions is “we do not know”.

Theoretical studies present models to either support or against circuit breakers.

Several attempts have also been made to determine the optimal level of price limits.

However, existing empirical work does not seem to provide consistent results to support the theoretical predictions. Conflicting results are mostly associated with price limits.

The conflicting results suggest that price limits deserve further examination and a serious policy debate.

The main conclusion of studies on trading halts is that trading halts do not reduce volatility and liquidity in stock markets. Both volume and volatility increase right after the trading halt. Regulators argue that the advantage of trading halts is that they lessen

47 the breakdown in the transmission of information among market participants. The existence of trading halts seems to suggest that the advantage is at least large enough to cover the costs. Further research is necessary to justify this net advantage of trading halts.

Table V provides the classification of existing circuit-breaker studies. Although many studies have been done on price limits, the lack of consistency suggests that more studies are necessary in the future to help us understand price limits better. Future studies on circuit breakers should focus on the effectiveness of price limits, the optimal design of price limits, the comparison between different types of circuit breakers, and the research methodology related to overcoming the constrained return generating process. Since circuit breakers exist in many markets in the world, international evidence is also encouraged. If circuit breakers are necessary from the policy consideration, which type of circuit breakers currently imposed in the world is the most effective? The relative performance between different types of circuit breakers appears to be the next most needed research topic in the circuit breaker literature.

48 References I

Ackert, Lucy F., Bryan Church, and Narayanan Jayaraman, 2001a, An experimental study of circuit breakers: the effects of mandated market closures and temporary halts on market behavior, Journal of Financial Markets 4, 185-208.

Ackert, Lucy F., Bryan K. Church, and Narayanan Jayaraman, 2001b, Circuit breakers with uncertainty about the presence of informed agents: I know what you know…I think, Working paper.

Ackert, Lucy F. and William C. Hunter, 1994, Rational price limits in futures markets: tests of a simple optimizing model, Review of Financial Economics 4, 93-108.

Anshuman, V. Ravi, and Avanidhar Subrahmanyam, 1999, Price limits, information acquisition, and bid-ask spreads: theory and evidence, Economic Notes 28, 91-118.

Arak, Marcelle, and Richard E. Cook, 1997, Do daily price limits act as magnets? The case of Treasury bond futures, Journal of Financial Services Research 12, 5-20.

Berkman, Henk, and Onno W. Steenbeek, 1998, The influence of daily price limits on trading in Nikkei futures, Journal of Futures Markets 18, 265-279.

Bhattacharya, Utpal, and Matthew Spiegel, 1998, Anatomy of a market failure: NYSE trading suspensions (1974-1988), Journal of Business & Economic Statistics 16, 216- 226.

Brady, Nicholas F. et al., 1988, Report of the Presidential Task Force on Market Mechanisms (Washington, D.C.: Government Printing Office).

Brennan, Michael J., 1986, A theory of price limits in futures markets, Journal of Financial Economics 16, 213-233.

Broussard, John P., 2001, Extreme-value and margin setting with and without price limits, Quarterly Review of Economics and Finance 41, 365-385.

Chan, Soon Huat, Kenneth A. Kim, and S. Ghon Rhee, 2001, Price limit performance: evidence from transactions data and the limit order book, Working paper.

Chen, Haiwei, 1998, Price limits, overreaction, and price resolution in futures markets, Journal of Futures Markets 18, 243-263.

Chen, Haiwei, 2002, Price limits and margin requirements in futures markets, Financial Review 37, 105-121.

Chen, Yea-Mow, 1993, Price limits and stock market volatility in Taiwan, Pacific-Basin Finance Journal 1, 139-153.

Chiang, Raymond, K.C. John Wei, and Soushan Wu, 1990, Price limits in Taiwan and risk-return estimation, In: S.G. Rhee and R.P. Chang, eds., Pacific-Basin Capital Markets Research (Elsevier Science Publishers B.V., North-Holland) 173-180.

Cho, David D., Jeffrey Russell, George C. Tiao, and Ruey Tsay, 2003, The magnet effect of price limits: evidence from high-frequency data on Taiwan Stock Exchange, Journal of Empirical Finance 10, 133-168.

Choi, Woo Suk, and Sang Bin Lee, 2001, A transitory and asymmetric properties in price limits: Korean evidence with intra-day data, Working paper.

Chou, Pin-Huang, 1997, A Gibbs sampling approach to the estimation of linear regression models under daily price limits, Pacific-Basin Finance Journal 5, 39-62.

Chou, Pin-Huang, Mei-Chen Lin, and Min-Teh Yu, 2000, Price limits, margin requirements, and default risk, Journal of Futures Markets 20, 573-602.

Christie, William G., Shane A. Corwin, and Jeffrey H. Harris, 2002, Nasdaq trading halts: the impact of market mechanisms on prices, trading activity, and execution costs, Journal of Finance 57, 1443-1478.

Chung, Jeff, and Li Gan, 2000, Empirical estimates of effect of price limits on limit- hitting days, Unpublished working paper, University of Texas, Austin.

Chung, Jong-Rock, 1991, Price limit system and volatility of Korean stock market, In: S.G. Rhee and R.P. Chang, eds., Pacific-Basin Capital Markets Research, Vol. 2 (Elsevier Science Publishers B.V., North-Holland) 283-294.

Corwin, Shane A., and Marc L. Lipson, 2000, Order flow and liquidity around NYSE trading halts, Journal of Finance 55, 1771-1801.

Coursey, Don L., and Edward A. Dyl, 1990, Price limits, trading suspensions, and the adjustment of prices to new information, The Review of Futures Markets 9, 343-360.

Fama, Eugene F., 1989, Perspectives on October 1987, or, what did we learn from the crash?, In: R.W. Kamphuis, Jr., R.C. Kormendi, and J.W.H. Watson, eds., Black Monday and the Future of Financial Markets (Mid America Institute for Public Policy Research, Chicago, IL) 71-82.

Ferris, Stephen P., Raman Kumar, and Glenn A. Wolfe, 1992, The effect of SEC-ordered suspensions on returns, volatility, and trading volume, The Financial Review 27, 1- 34.

50 George, Thomas J., and Chuan-Yang Hwang, 1995, Transitory price changes and price- limit rules: evidence from the Tokyo Stock Exchange, Journal of Financial and Quantitative Analysis 30, 313-327.

Gerety, Mason S., and J. Harold Mulherin, 1992, Trading halts and market activity: an analysis of volume at the open and the close, Journal of Finance 47, 1765-1784.

Goldstein, Michael A., and Kenneth A. Kavajecz, 2000, Liquidity provision during circuit breakers and extreme market movements, Unpublished working paper.

Greenwald, Bruce C., and Jeremy C. Stein, 1988, The Task Force Report: the reasoning behind the recommendations, Journal of Economic Perspectives 2, 3-23.

Greenwald, Bruce C., and Jeremy C. Stein, 1991, Transactional risk, market crashes, and the role of circuit breakers, Journal of Business 64, 443-462.

Hall, Anthony D., and Paul Kofman, 2001, Limits to linear price behavior: futures prices regulated by limits, Journal of Futures Markets 21, 463-488.

Harris, Lawrence E., 1998, Circuit breakers and program trading limits: what have we learned? In: Litan, R.E., and A.M. Santomero (eds.), Brookings-Wharton Papers on Financial Services, Brookings Institution Press, Washington, DC, pp. 17-64.

Howe, John S., and Gary G. Schlarbaum, 1986, SEC trading suspensions: empirical evidence, Journal of Financial and Quantitative Analysis 21, 323-333.

Kao, G. Wenchi, and Christopher K. Ma, 1992, Memories, heteroscedasticity, and price limit in currency futures markets, Journal of Futures Markets 12, 679-692.

Kim, Kenneth A., 2001, Price limits and stock market volatility, Economics Letters 71, 131-136.

Kim, Kenneth A., and Piman Limpaphayom, 2000, Characteristics of stocks that frequently hit price limits: empirical evidence from Taiwan and Thailand, Journal of Financial Markets 3, 315-332.

Kim, Kenneth A., and S. Ghon Rhee, 1997, Price limit performance: evidence from the Tokyo Stock Exchange, Journal of Finance 52, 885-901.

Kim, Kenneth A., and Richard J. Sweeney, 2001, Effects of price limits on information revelation: theory and evidence, Working paper.

Kim, Yong H., José Yagüe, and J. Jimmy Yang, 2003, The relative performance between trading halts and price limits: Spanish evidence, Working paper.

51 Kodres, Laura E., and Daniel P. O’Brien, 1994, The existence of Pareto-superior price limits, American Economic Review 84, 919-932.

Kuhn, Betsey A., Gregory J. Kuserk, and Peter Locke, 1991, Do circuit breakers moderate volatility? Evidence from October 1989, The Review of Futures Markets 10, 136-175.

Kyle, Albert S., 1988, Trading halts and price limits, The Review of Futures Markets 7, 426-434.

Lauterbach, Beni, and Uri Ben-Zion, 1993, Stock market crashes and the performance of Circuit Breakers: empirical evidence, Journal of Finance 48, 1909-1925.

Lee, Charles M.C., Mark J. Ready, and Paul J. Seguin, 1994, Volume, volatility, and New York Stock Exchange trading halts, Journal of Finance 49, 183-214.

Lee, Sang Bin, and Dae Joong Kim, 1997, Price limits and beta, Review of Quantitative Finance and Accounting 9, 35-52.

Lee, Sang-Bin, and Kwang-Jung Kim, 1995, The effect of price limits on stock price volatility: empirical evidence in Korea, Journal of Business Finance & Accounting 22, 257-267.

Lehmann, Bruce N., 1989, Commentary: Volatility, price resolution, and the effectiveness of price limits, Journal of Financial Services Research 3, 205-209.

Ma, Christopher K., Ramesh P. Rao, and R. Stephen Sears, 1989, Volatility, price resolution, and the effectiveness of price limits, Journal of Financial Services Research 3, 165-199.

Miles, David K., 1990, An appraisal of stock market circuit breakers, Managerial Finance 16, 25-32.

Miller, Merton H., 1989, Commentary: Volatility, price resolution, and the effectiveness of price limits, Journal of Financial Services Research 3, 201-203.

Moser, James T., 1990, Circuit breakers, Economic Perspectives 14, 2-13.

Park, Chul Woo, 2000, Examining futures price changes and volatility on the trading day after a limit-lock day, Journal of Futures Markets 20, 445-466.

Pettway, Richard H., and Takashi Kaneko, 1996, The effects of removing price limits and introducing auctions upon short-term IPO returns: the case of Japanese IPOs, Pacific- Basin Finance Journal 4, 241-258.

52 Phylaktis, Kate, Manolis Kavussanos, and Gikas Manalis, 1999, Price limits and stock market volatility in the Athens Stock Exchange, European Financial Management 5, 69-84.

Shen, Chung-Hua, and Lee-Rong Wang, 1998, Daily serial correlation, trading volume and price limits: evidence from the Taiwan stock market, Pacific-Basin Finance Journal 6, 251-273.

Subrahmanyam, Avanidhar, 1994, Circuit breakers and market volatility: a theoretical perspective, Journal of Finance 49, 237-254.

Subrahmanyam, Avanidhar, 1995, On rules versus discretion in procedures to halt trade, Journal of Economics and Business 47, 1-16.

Subrahmanyam, Avanidhar, 1997, The ex ante effects of trade halting rules on informed trading strategies and market liquidity, Review of Financial Economics 6, 1-14.

Telser, Lester G., 1981, Margins and futures contracts, Journal of Futures Markets 1, 225-253.

Wei, K.C. John, and Raymond Chiang, 2001, A GMM approach for estimation of volatility and regression models when daily prices are subject to price limits, Working paper.

Wei, Steven X., 2002, A censored-GARCH model of asset returns with price limits, Journal of Empirical Finance 9, 197-223.

Wu, Lifan, 1998, Market reactions to the Hong Kong trading suspensions: mandatory versus voluntary, Journal of Business Finance & Accounting 25, 419-437.

Table I: Types of Circuit Breakers

Circuit Breakers

Price limits Firm-specific trading halts Market-wide trading halts

Futures Markets Stock Markets SEC-initiated Exchange-initiated NYSE Circuit Breakers

Contract-specific Percentage-based Price level-based Order imbalance News

54 Table II Price limits in futures markets

Agricultural Contracts:

20 cents/bu ($1,000/contract) above or below the previous day's Corn Futures settlement price 20 cents/bu ($1,000/contract) above or below the previous day's Oats Futures settlement price. 50 cents/cwt ($1,000/contract) above or below the previous day's Rough Rice Futures settlement price. $20/ton ($2,000/contract) above or below the previous day's Soybean Meal Futures settlement price. 2 cents per pound ($1,200/contract) above or below the previous Soybean Oil Futures day's settlement price. 50 cents/bu ($2,500/contract) above or below the previous day's Soybeans Futures settlement price. 30 cents/bu ($1,500/contract) above or below the previous day's Wheat Futures settlement price.

Index Contracts: Dow Jones Industrial Average Successive 10%, 20%, and 30% price limits based on the average Futures daily close of the cash index in the last month of the preceding quarter. Daytime price limits are coordinated with NYSE circuit-breakers.

Financial Contracts no price limits

Source: Chicago Board of Trade website

55 Table III Price limits in stock markets

Country Price limits Notes

Austria 5% Belgium 10% China 10% Finland 15% France 10-20% different limits apply to different groups of stocks Greece 4% and 8% 8% on highly active stocks, since August 1992 Italy 10-20% Japan 10-60% price-level based price limits Korea 15% increased from 6% Luxembourg 5% Malaysia 30% in each of the two trading sessions per day Mexico 10% Peru 15% Romania 15% South Africa 2-6% price-level based price limits Spain 15% triggers for trading halts since 2001 Switzerland 10-15% trading halts for 15 minutes Taiwan 7% since 1989, but adjusted down to 3.5% several times Thailand 30% increased from 10% in 1997 Turkey 10% in each of the two trading sessions per day, since 1994

Source: from stock exchange websites and Fact Books

56 Table IV Circuit Breaker System on the NYSE

October 19, 1988: first implementation of circuit breakers 250 points drop in DJIA—trading halt for one hour 400 points drop in DJIA—trading halt for two hours if there are additional 150 points drop after trading resumes

February 3, 1997: widen the circuit breakers 350 points drop in DJIA—trading halt for half an hour 550 points drop in DJIA—trading halt for one hour if there are additional 200 points drop after trading resumes

May 1, 1998: increase circuit breaker thresholds according to a quarterly adjusted percentage-based system. The threshold point is adjusted based on the DJIA average closing values of the previous month, rounded to the nearest 50 points. 10% drop in DJIA—trading halt for one hour if occurred before 2:00 p.m., trading halt for 30 minutes if occurred between 2:00 and 2:30 p.m., no trading halt if occurred after 2:30 p.m. 20% drop in DJIA—trading halt for 2 hours if occurred before 1:00 p.m., trading halt for one hour if occurred between 1:00 and 2:00 p.m., market closure if occurred after 2:00 p.m. 30% drop in DJIA—market closure, regardless of when hit.

Source: NYSE website

57 Table V Classification of Circuit-Breaker Studies

Types Price limits Firm-specific Market-wide circuit Comparison trading halts breakers Theoretical studies Moser(1990) Moser(1990) Subrahmanyam Telser(1981) Brennan(1986) Kyle(1988) (1994) Subrahmanyam Kodres and Greenwald and (1995) O’Brien(1994) Stein(1988) Fama(1989) Fama(1989) Lehmann(1989) Greenwald and Telser(1981) Stein(1991) Chou et al.(2000) Subrahmanyam Ackert and (1997) Hunter(1994) Anshuman and Subrahmanyam (1999) Kim and Sweeney(2001) Empirical studies: Ma et al.(1989) Kuhn et al.(1991) futures markets Kao and Ma(1992) Arak and Cook(1997) Berkman and Steenbeek(1998) Chen(1998) Park(2000) Hall and Kofman(2001) Chen(2002) Empirical studies: Chung(1991) Howe and Gerety and Kim et al.(2003) stock markets Chen(1993) Schlarbaum (1986) Mulherin(1992) George and Ferris et al.(1992) Lauterbach and Hwang(1995) Lee et al.(1994) Ben-Zion(1993) Lee and Corwin and Goldstein and Kim(1995) Lipson(2000) Kavajecz(2000) Pettway and Christie et Kaneko(1996) al.(2002) Kim and Bhattacharya and Rhee(1997) Spiegel Shen and (1998) Wang(1998) Wu(1998) Phylaktis et al.(1999) Kim and Limpaphayom (2000) Chan et al.(2001) Choi and Lee(2001) Kim(2001) Kim and Sweeney(2001)

58 Cho et al.(2003) Experimental studies Ackert et Coursey and al.(2001a) Dyl(1990) Ackert et al.(2001b) Methodology Chiang et al.(1990) Lee and Kim(1997) Chou(1997) Chung and Gan(2000) Wei(2002) Broussard(2001) Wei and Chiang(2001)

59 Chapter II: The Performance of Price Limits under Asymmetric Information: Evidence from Taiwanese IPOs

II.1. Introduction

After the October 1987 market crash, the market stabilization issue attracted a great deal of attention from both academic researchers and securities regulators. The

Brady Report (1988) suggested the imposition of “circuit breakers” to prevent excessive fluctuations in the stock market. Basically, there are two different types of circuit breakers: trading halts and price limits. The existing literature on trading halts consistently finds that both volatility and trading volume increased significantly after trading halts (see, e.g., Christie, Corwin, and Harris, 2002; Corwin and Lipson, 2000;

Lee, Ready, and Seguin, 1994). Results from price limit studies, on the other hand, are conflicting (see, e.g., Kim and Rhee, 1997; Lee and Kim, 1995; Phylaktis, Kavussanos, and Manalis, 1999). Hence, the main objective of this paper is to shed some light on the price limit literature by empirically investigating the performance of price limits using

Taiwanese IPOs.

By definition, price limits are boundaries set by market regulators to restrict daily security price changes in an effort to curb excess volatility. The U.S. futures markets and many stock exchanges around the world currently adopt price limits. For example,

China, Japan, Korea, Malaysia, Taiwan and Thailand in Asia and Austria, Belgium,

France, Italy, Netherlands, Spain, and Switzerland in Europe impose price limits on their stock markets. However, are price limits effective in achieving the intended goal? Kim and Limpaphayom (2000) point out that very little is known regarding price limit effects despite their significant presence. Apparently, the academic research has yet to provide a

60 satisfactory answer to this question. Harris (1998) also states that the ignorance of price limit issue is unfortunate because we need to make informed decisions as to how our markets should be protected.

Theoretically, no consensus has been achieved on whether the imposition of price limits has the desired result of reducing stock price volatility. Price limit supporters argue that it can provide a cooling off period, allowing investors to re-evaluate market information so that more rational investment strategies can be formulated. Thus, price limits are believed to be able to reduce overreaction and decrease stock price volatility.

Proponents of price limits also emphasize loss reduction and risk sharing among investors. Brennan (1986) shows that price limits may act as a partial substitute for margin requirements without resorting to costly litigation. During periods of volatile price movements, price limits lower conditionally expected losses so investors are more likely to pay margin calls on time rather than default. Kodres and O’Brien (1994) examine the welfare effects of price limits. They show that price limits may promote better risk sharing than unconstrained trading when price fluctuations are driven by news about fundamentals. However, the opponents of price limits argue that they serve no purpose other than to delay price discovery (e.g., Fama, 1989). Even though price limits can stop the price of a share from falling or rising beyond the limit on a given trading day, they argue, the price will continue to move in the direction towards equilibrium as new trading limits are established in subsequent trading day(s). Given the lack of consensus, empirical evidence is required to provide insights to the debate over the effectiveness of price limits.

61 Unfortunately, there is no perfect way to empirically test the effectiveness of price limits because we cannot observe what would have happened in the absence of price limits. Rather than directly testing the effectiveness of price limits, we focus on examining the performance of price limits under asymmetric information. The IPO literature has documented that significant information asymmetry exists in IPOs. For example, Rock (1986) uses the information asymmetry between informed and uninformed investors to explain the underpricing of IPOs. The information cascades or herding in IPOs proposed by Welch (1992) further demonstrate that information asymmetry is indeed prominent in IPOs. Thus, IPOs can represent stocks experiencing significant asymmetric information. Our result can be used to predict price limit effects during periods with information shocks or high information asymmetries (e.g., terrorists attack or other unexpected news announcements). During these periods, stock prices are likely to generate significant movement due to fundamental changes of asset value and/or possible overreaction. Since price limits are designed to reduce investor overreaction, our analysis of IPOs provides insights to their capability of achieving this goal during these periods.

We choose Taiwanese data for four reasons. First, unlike other stock markets, the

Taiwan Stock Exchange (TSE) implements relatively tight price limits and thus limit hits are not rare events. Accordingly, a reasonable sample size can be obtained from the TSE.

Second, the TSE price limit system is based on a fixed rate so all traded stocks are subject to the same percentage movement. Unlike the TSE, the Tokyo Stock Exchange has established daily price limits that are based on the previous day’s closing prices and vary according to price levels. In terms of percentage, the price limits range from 5% to 50%

62 or even higher. Studies of price limits on this market may be contaminated by the varying price-limit rates because the observed difference in return volatility may be due to the variation in price-limit rates. Using TSE data, we can avoid this potential contamination. In addition, among the six non-Japan Asian equity markets (Hong Kong,

Korea, Malaysia, Singapore, Taiwan, and Thailand), Taiwan has the largest market capitalization as well as trading volume (Rhee and Chang, 1993). Thus, the problem of thin trading or lack of liquidity is not a concern on the TSE. Furthermore, in Taiwan,

IPOs are subject to little underwriters’ price support so the price discovery process is less distorted as compared to countries with significant underwriters’ support. Based on these four reasons, we believe that the Taiwanese IPO sample is an ideal candidate for studying the performance of price limits under asymmetric information.

Price limits can impact volatility, trading activity, and price discovery. In this paper, we empirically test the performance of price limits by examining questions raised by the opponents of price limits, namely, volatility spillover, trading interference, and delayed price discovery. However, instead of following the volatility spillover argument, we form an information-based hypothesis to test the impact of price limits on volatility.

Our tests are based on the comparison between IPOs and their industry-and-size matched seasoned equities (MSEs). The information asymmetry hypothesis, which states that price limits cause higher volatility levels on subsequent days for IPOs than MSEs, is supported. The trading interference hypothesis, which implies that price limits interfere with trading and cause higher trading activity on subsequent days for IPOs than MSEs, is supported only by lower limit hits. The delayed price discovery hypothesis is not supported by our data. Overall, our results represent the performance of price limits for

63 IPOs and can be used to predict the performance of price limits during periods with high information asymmetry.

This paper is structured as follows. Section 2 offers institutional background of the TSE. Section 3 provides a review of the previous literature. Section 4 presents all hypotheses. Section 5 discusses the data selection and research methodology. In Section

6, detail results of the performance of price limits as well as tests of each hypothesis are presented. Section 7 concludes the paper.

II.2. Institutional background

II.2.A. Taiwan Stock Exchange

According to the Monthly Bulletin of Statistics of the Republic of China (January,

2000), there are 462 stocks listed in TSE. The total market value is New Taiwanese

Dollar (NT$) 11,787,330 million (or about US$ 375.5 billion at the exchange rate of NT$

31.39/US$). The average daily trading value, defined as the sum of the products of trading volume and transaction prices, is NT$110,120 million (US$ 3.5 billion).

The TSE is an order-driven market with no market makers or specialists.

According to the Taiwan Stock Exchange Corporation (TSEC), in August 1985, the open outcry system was gradually replaced with a computer-aided trading system (CATS), which was eventually upgraded to a fully automated securities trading (FAST) system in

1993. The trading session of the centralized market is 9:00 a.m. to 12:00 p.m., Monday through Friday. On Saturday, trading takes place from 9:00 a.m. to 11:00 a.m. Thirty

64 minutes before the market opens, orders can be submitted via security firms on a first- come-first-serve basis. The opening price is the one that maximizes trading volume.

The TSE has been imposing daily price limits since its inception in 1962. The purpose of the price limits is to prevent excessive volatility and to protect investors by limiting potential daily losses. The TSE sets its daily price limits at a predetermined rate, both upward and downward, based on the previous day’s closing prices. The price-limit rate has been adjusted up or down several times in accordance with market conditions.

Appendix A provides the dates of each adjustment of price-limit rates as well as the reason announced by the TSEC. For stocks listed on the TSE, tick sizes (the minimum allowable unit that stock prices may deviate) vary with market prices. Appendix B reports the tick size for each price range. Stocks that hit their price limits are still allowed to trade as long as the transaction prices are within the limits. Thus, the TSE price limits are simply boundaries, not triggers for trading halts.

II.2.B. IPOs in Taiwan

In Taiwan, most IPOs are underwritten through the best efforts underwriting agreement. As reported by Ritter (1987), IPOs associated with best efforts offers generate a higher degree of underpricing than those with firm commitment offers, which might partially explain the significant underpricing on the TSE. Another reason is that the regulatory authority requires issuing firms to set an offering price based on a standard formula that usually results in a relatively low offering price, almost guaranteeing the successful sale of IPOs. Each investor is allowed to subscribe for a fixed number of

65 shares, typically one or two thousand shares. When an oversubscription occurs, shares are allocated on a random basis through a lottery drawing system. Starting in November

1997, the public offering of all TSEC-listed securities can be distributed through the

TSEC’s computer drawing system. Because of best efforts offers and low mandatory offering prices, the reputation effect of underwriters suggested by Carter and Manaster

(1990) may be less important for the success of promoting an IPO in Taiwan. Besides, since the reputation effect is not a major issue in IPO business, the stabilization activities by underwriters documented by Aggarwal (2000) may be less profound on the TSE.

Furthermore, short sales are not allowed for the first six months after going public.

It should be noted that price limit regulation varies from country to country. For

IPOs, the price limit is applied to each IPO’s offering price in Taiwan; therefore, the first opening price is also subject to the limit and thus can only be within the 7% range, either above or below the offering price. In Japan and Korea, however, the price limit is applied to the first opening price. In other words, there is no limit on the price movement from the offering price to the first opening price. Given the international evidence of IPO underpricing provided by Loughran, Ritter, and Rydqvist (1994), it would definitely take more days for IPOs in Taiwan to fully reflect the degree of underpricing than IPOs in

Japan and Korea due to regulatory differences. In fact, Huang (1999) examines the stock price behavior of 311 IPOs on the TSE for the period 1971-1995 and finds that they are significantly underpriced. Based on his study, the average risk-adjusted initial return is

42.6% from the listed day to the first non-limit trading day. Given the 7% price limit, it would take an average of five to six trading days to reflect the high degree of underpricing on the TSE.

II.3. Literature review

Even though the price-limit issue has attracted a lot of attention from both

academic researchers and securities regulators since the 1987 market crash, most of the

studies examine the futures markets rather than the stock markets.1 In contrast, Kim and

Rhee (1997) examine the daily stock price data of the Tokyo Stock Exchange from 1989 to 1992 to investigate the performance of price limits. Based on the belief that stocks that reach their price limit are prevented from correcting their order imbalance, while stocks that almost hit their limit are not, they divide stocks into three different groups. The first group includes stocks that reach their daily price limit. The second group includes stocks that experience a price change of at least 90% of the limit, but do not reach the price limit. The third group includes stocks that experience a price change between 80% and

90% of the limit. With the comparison of those three groups, they find evidence that supports the arguments of price limits opponents, i.e. volatility spillover, trading interference, and delayed price discovery. They then conclude that the price limit system of the Tokyo Stock Exchange may be ineffective. However, we argue that the formation of the comparison groups in their study is arbitrary given the fact that the price limits range from 5% to 50% or higher in terms of percentage. More importantly, using the daily high or low price, instead of the closing price, as the criterion for selecting limit-hit stocks cannot really capture stocks with actual order imbalance problem. If the closing

1 For studies of price limit on the futures markets, please see Arak and Cook (1997), Berkman and Steenbeek (1998), Kao and Ma (1992), and Ma, Rao, and Sears (1989) for example. However, those studies found no consensus evidence on the effectiveness of price limits.

67 price is less than the daily high or higher than the daily low, the order imbalance problem has already been resolved by the market itself. The inclusion of those stocks in the first group would exaggerate the degree of order imbalance.

Lee and Kim (1995) investigate the effect of price limits on stock price volatility using the daily stock price data of the Korea Stock Exchange from 1980 to 1989 and find, contrary to Kim and Rhee (1997), that price limits serve to reduce stock price volatility.

However, the issue of multiple limit hits, which occur when prices hit the limit for consecutive days, is ignored in their study. Our study shows that the results are sensitive to the inclusion or exclusion of multiple limit hits. In fact, Miller (1989) points out that the selection bias associated with the exclusion of multiple limit hits is harmful because the next day’s move could have been an equally large or larger move in the same direction. However, if multiple limit hits are included, whether to treat each limit hit independently or not is troublesome. Furthermore, it is difficult to conclude that the difference in volatility is due to differences in price limits because volatility can vary across time for a variety of reasons (see, e.g., Schwert, 1989).

Phylaktis, Kavussanos, and Manalis (1999) assess the impact of price limits on stock volatility in the Athens Stock Exchange where price limits were imposed in August

68 of several years need to be interpreted cautiously. It is difficult to associate a change in long-run volatility to a specific change in policy.

As mentioned earlier, price limits are considered one form of circuit breaker, but they differ in several ways. For instance, price limits are applied to individual stocks while circuit breakers are triggered based on the movement of an overall market index.

Furthermore, price limits are simply boundaries because trading is allowed as long as the price is at or within the limits. Circuit breakers, on the other hand, halt trading if they are triggered. Even though there are differences between price limits and circuit breakers, the objective of both mechanisms is the same—to stabilize the market. Subrahmanyam

(1994) provides a theoretical analysis on the "circuit breaker" system and Lauterbach and

Ben-Zion (1993) examine it empirically for the Israeli stock market. Subrahmanyam argues that the existence of circuit breakers may distort the optimal trading decisions of large institutions and thus may have perverse effects on stock volatility and market liquidity. However, according to Lauterbach and Ben-Zion, circuit breakers actually reduce the next-day opening order imbalance and the initial price loss, even though no long-run response is observed. Therefore, for circuit breakers, the empirical result seems to contradict the theoretical prediction.

Overall, the existing studies provide mixed evidence on the performance of price limits and circuit breakers in terms of controlling stock price volatility. Furthermore, the results are debatable due to the exclusion of multiple limit hits, the arbitrarily selected comparison groups, and the varying price limit rates over different price levels or over time. Our study is based on the examination of IPOs, which tend to have greater asymmetric information than seasoned equities, so the results should provide a better

69 ground for empirically evaluating the performance of price limits under asymmetric

information. Since the price limit rate is fixed at 7% for the period of study2, the data

provides a clean sample that cannot be obtained from Greece, Japan, or Korea. We

examine all IPOs issued during the period from 1989 to 2000; therefore, no arbitrarily

selected groups are formed.

II.4. Hypotheses

Opponents of price limits raise three potential problems: volatility spillover,

trading interference, and delayed price discovery. In order to investigate the performance

of price limits, we form three hypotheses based on these potential drawbacks attributed to

the imposition of price limits. Detailed descriptions of each hypothesis as well as

supporting and rejecting criteria are as follows.

H1: Information asymmetry hypothesis – Price limits cause higher volatility levels

for IPOs than for the matching seasoned equities (MSEs) on days following a limit hit.

With price limits, investors are restricted from trading beyond the limit prices. If the

equilibrium price is outside the allowable price range, there will be no trading. Given

that information has to be transmitted through trading, price limits maintain some degree

of asymmetric information between traders following a limit-hit day. Since by nature,

the degree of information asymmetry is high for IPOs, we expect to see greater impact of

2Even though the price limit has been set at 7% since Oct. 11, 1989, during special periods such as the 9/21/1999 earthquake or the presidential elections, the lower limit was shrunk to 3.5%. (See Appendix A for detail.) No IPO was issued during these periods, so all IPOs in our sample period are subject to 7% price limit.

70 price limits on IPOs than on MSEs3. The higher degree of information asymmetry would then lead to higher volatility. If we observe higher volatility on subsequent day(s) for

IPOs than for MSEs, this hypothesis is supported; otherwise, it is rejected.

H2: Trading interference hypothesis – Price limits interfere with trading due to

limitations imposed by price limits. When price limits are hit, investors have to wait until

the next day or even the subsequent days to fill unexecuted orders. This implies that

trading activity will be intensified on the day(s) following the limit-hit day. However,

since IPOs are subject to higher degree of information asymmetry, investors might be less

willing to trade IPOs on the limit-hit day as compare to MSEs. Accordingly, trading

activity will be intensified on subsequent day(s) at a higher degree for IPOs than for

MSEs. This hypothesis suggests that price limits reduce market liquidity and deprive

traders the opportunity to change their portfolio allocations on the limit-hit day.

However, on the other hand, if price limits provide a cooling-off period to reduce

overreacting activities, trading activity might not be intensified following a limit-hit day.

If we observe higher trading activity on subsequent day(s) for IPOs than for MSEs, this

trading interference hypothesis is supported; otherwise it is rejected.

H3: Delayed price discovery hypothesis – Price limits prevent prices from

efficiently reaching their equilibrium levels. That is, when price limits are hit, the

equilibrium price will not be reached until the next trading day or even the subsequent

days. This hypothesis suggests that price limits are ineffective because they reduce

market efficiency. Because of price limits, it takes longer for information to be reflected

3 Because IPOs have come to symbolize the insider abuses in recent years, Peristiani (2003) examines IPOs for the 1980-2000 period to investigate their post-issue riskiness. He finds that the riskiness of IPOs is indeed higher relative to that of seasoned equities.

71 into stock prices. There are two criteria for testing this hypothesis. First, based on the belief that information arrives randomly, we assume that price continuations and reversals are equally likely to happen. Thus, the delayed price discovery hypothesis holds if there are more price continuations than price reversals after the limit-hit day.

Based on the information asymmetry argument, we expect to see more price continuations from IPOs than from MSEs because trading is interfered at a higher degree for IPOs than MSEs. Second, we adopt the event methodology to measure the speed of price discovery given price limits. On the limit-hit day, the abnormal return cannot be fully reflected if price limits delay price discovery. Therefore, we except to see abnormal return on days following the limit-hit day. Again, due to the information asymmetry, the abnormal return of IPOs is expected to be higher for upper limit hits and lower for lower limit hits than that of MSEs.

II.5. Data and methodology

II.5.A. Data description

We obtained a list of all IPOs issued on the TSE during the period from October

11, 1989 to July 31, 2000 and their offering prices from the Taiwan Stock Exchange

Corporation (TSEC). The main reason to start with October 11, 1989 is that the price limit was extended from 5% to 7% on that day. Since previous studies suggest that regulatory changes have significant impact on IPO markets (see, e.g., Hebner and Hiraki,

1993), we exclude the period with varying price-limit rates and focus on the period with

72 only 7% price limits. Even though the price limit has been set at 7% since October 11,

1989, during special periods such as earthquakes or presidential elections, the lower limit was reduced to 3.5%. However, no IPOs were issued during those periods, so all IPOs in our sample period are subject to the 7% price limits. As to the end date, July 31, 2000 is the most current data available at the time of this study. Based on the list, there are 296

IPOs issued during this period.

Trading data, including opening price, closing price, daily high price, daily low price, trading volume, trading value, and daily number of transactions for IPOs are obtained from the Taiwan Economic Data Center (TEDC). Trading data are available for

294 of the 296 IPOs in the TSEC list from TEDC. Therefore, there are 294 IPOs in the sample.

It was noted earlier that the price limit has been extended to 7% from 5% since

October 11, 1989. That is, the price of each stock can only move up or down within the positive and negative 7% range based on the closing price of the previous trading day.

For IPOs, since the previous trading day’s closing prices are not available, the 7% price limit is applied to the offering prices. Due to tick size, set by the TSE, only a few IPOs can hit the exact 7% price limit. For example, with an offering price of NT$80, the price on the first trading day is allowed to go up to NT$85.6 or go down to NT$74.4.

However, the tick size for the price range between NT$50 and NT$150 is NT$0.5.

Therefore, a limit hit occurs when the price goes up to NT$85.5 or goes down to

NT$74.5 in this case, even though the return is only 6.875% or negative 6.875%, respectively.

73 Table I reports the yearly breakdown of the number of listings and shows the price-limit-hit occurrences for both upper and lower price movements on the first trading day. Limit-hit IPOs are those that hit the price limits on their first trading day. Due to the severe IPO underpricing in the TSE, there are 245 (83.33%) upper limit-hit IPOs and only 14 (4.76%) lower limit-hit IPOs. Non-hit IPOs are those that did not hit the price limits on their first trading day. Upward (downward) refers to those whose first closing price is higher (lower) than the offering price, while unchanged refers to those whose first closing price is the same as the offering price. There are 16 upward, 16 downward, and three unchanged non-hit IPOs. In sum, there are 259 limit-hit IPOs and 35 non-hit IPOs.

Table I also reports the average market returns based on the TSEC Capitalization

Weighted Price Index (TAIEX) on the first trading day of each IPO. For limit-hit IPOs, the overall market behavior seems to have little impact on the IPO performance because the average market return is higher for lower limit-hit IPOs than for upper limit-hit IPOs.

However, for non-hit IPOs, the overall market plays an important role for the IPO return because the average market return is higher for upward non-hit IPOs than for downward ones. In fact, the average market return was negative when IPOs had downward movements.

Due to the high degree of underpricing, many IPOs consecutively hit the price limit after going public. When a limit is hit, an order imbalance occurs and the equilibrium price is not observable. For example, if an IPO hits the price limit for five consecutive days after going public, the price limit causes order imbalances on the first five days. In this case, the equilibrium price is not achieved until the sixth day when the order imbalance is digested by the market. If the price limit is never hit after going

74 public, there is no order imbalance and the equilibrium price can be reached on the first trading day. Thus, we define the number of delayed days as the number of days required for IPOs to reach their equilibrium prices after going public.

Table II reports the number of delayed days by the year of listing. Without price limits, the equilibrium price can be reached on the first trading day. However, with the

7% price limit, it takes an average of 6.24 days for IPOs to reach their equilibrium prices.

Since the median of delayed days is only four, the distribution is highly skewed. Fig. 2 provides the distribution of delayed days and shows that it is skewed to the right. The number of delayed days ranges from one to 31. With price limits, the order imbalance prevents prices from moving toward their equilibrium prices. For some hot IPOs, the numbers of delayed days are more than 25. That is, it took more than 25 trading days for the stock prices to reach their equilibrium prices. On those days, there are nearly no sell orders and accordingly the trading volume is extremely low. Most of the time, we observe only single digit trading volume during these days. Apparently, because of price limits and the IPO underpricing, the trading is interfered and the price discovery is delayed during these days.

In order to test the performance of price limits under asymmetric information, we examine each IPO’s first limit hit occurring after the equilibrium price is reached but within one month after going public. We believe that once the equilibrium price is reached, the price behavior of IPOs is less distorted. For IPOs taking more than one month to reach their equilibrium prices, the observation period is extended to two months. If no limit hit is observed during the two-month period, the IPO is dropped out of the sample since the degree of information asymmetry decreases over time. We find

75 174 IPOs that meet our selection criteria. 97 of them hit the upper limit and 77 of them hit the lower limit.

Multiple hits occur when two or more consecutive limit hits are observed. Since

Miller (1989) points out that there appears to be no solution as to whether to keep multiple hits or throw them out of the sample, both samples with and without multiple hits are investigated. There are 121 IPOs in the sample if multiple hits are excluded. Of which, 71 hit the upper limit and 50 hit the lower limit.

Since trading behavior could be different for firms with different sizes and in different industries, we construct a control sample consisting of industry-and-size matched seasoned equities (MSEs). Trading data for those MSEs are obtained from the

Pacific-Basin Capital Markets (PACAP) Databases. For each IPO, we obtain seasoned equities that also hit the price limits on the same day the IPO hit the limit. We keep only those that have been traded on the market for at least one year to prevent from getting firms that are also subject to information asymmetry problems. From the remaining firms, we choose the one that is in the same industry as the IPO and whose market capitalization is closest to the IPO to be our MSE. However, we cannot always find limit-hit seasoned equities in the same industry as the IPO. In these cases, MSEs are selected only based on their market capitalization. In order to make pair-wise comparisons, we drop IPOs that do not have MSEs. For some missing and questionable data on the PACAP database, we obtain and correct them from the TEDC.

Table III reports the sample sizes as well as the means and medians of market capitalizations for IPOs and MSEs. Although initially there are 174 IPOs in our sample, we are able to find only 146 MSEs, with 87 hitting the upper limit and 59 hitting the

76 lower limit. If multiple hits are excluded, there are 106 observations with 65 upper limit hits and 41 lower limit hits. Even though we try to match our samples by industry and size, only about 50% of the MSEs are matched by industry. 34 out of the 87 upper limit hits and 40 out of the 59 lower limit hits are industry matched. This seems to suggest that the industry effect is more profound when a price hits the lower limit.

The market capitalization in Table III is defined as the last closing price of the limit-hit month multiplied by the number of shares outstanding at month end. The unit of market capitalization is millions of Taiwanese currency. For those industry matched observations, the market capitalization might be so different between IPOs and MSEs if only one seasoned equity in the same industry is available. However, Table III shows that the difference of market capitalization between IPOs and MSEs is statistically insignificant in terms of means and medians. That is, the difference on trading behavior between IPOs and MSEs caused by the potential size effect can be ignored.

II.5.B. Methodology

We define the event day t as the limit-hit day for both IPOs and MSEs. When multiple hits occur, the last limit-hit day is the event day t. Line (a) of Fig. 1 depicts the timeline from the first trading day (the IPO date) to the equilibrium price (EP)-reached day. Line (b) of Fig. 1 represents the Post-EP timeline. Our focus is to investigate the volatility, trading activity, and price discovery of IPOs and MSEs on days surrounding the event day.

77 II.5.B.1. Information asymmetry hypothesis test

In order to test the information asymmetry hypothesis, an appropriate measurement of volatility is necessary and is crucial to the test result. Ma, Rao, and

Sears (1989) adopt the extreme value method proposed by Parkinson (1980) to measure daily volatility and point out that this measurement provides a more current approximation of volatility than historical measures. As suggested by Parkinson (1980) and Garman and Klass (1980), the daily volatility of security price measured by daily high and low prices is superior to the traditional method4. Grossman (1988) also points out that the close-to-close return is a market direction measure, not a volatility measure.

Like Parkinson, he uses daily high and low prices to calculate daily volatility.

Economists usually identify two types of volatility: fundamental volatility and transitory volatility. Fundamental volatility reflects the uncertainty about underlying security values, but transitory volatility is often caused by the trading process. Price limit proponents are more interested in the ability to reduce transitory volatility rather than the fundamental volatility. Harris (1998) defines transitory volatility as the tendency of prices to bounce around their fundamental values. We believe that the use of daily high and low prices can capture the transitory volatility better than using the daily closing prices. For daily comparison between IPOs and MSEs, therefore, the Grossman and the

4 Kim and Rhee (1997) use the traditional simple method to measure daily volatility from the close–to- close returns. Using this measurement, the daily volatility is always the highest on the limit-hit day. Lehmann (1989) argues that daily volatility measured by extreme values is subject to measurement error of the high and low prices. One way to overcome Lehmann’s argument is to use Grossman’s (1988) logarithmic volatility measure, which is designed to flatten out extreme values.

78 Parkinson measures are adopted in our investigation5. The volatility measures based on

the methods used by Grossman can be calculated as follows:

 HPT  VolatilityT (VGT) = ln  (1)  LPT 

Alternatively, the volatility measure based on the extreme value method suggested by

Parkinson can be expressed as follows:

2 ()HPT− LPT VolatilityT (VPT) = (2) 42ln where HPT is the high price and LPT is the low price on day T, and ln is the natural log

transformation. We calculate the means and medians of daily volatilities from t-1, one

day prior to the event day t, to t+10, 10 days after t, across IPOs and MSEs in order to

investigate the daily volatility changes over the 12-day window. Previous studies usually

report the daily volatility 10 days before and 10 days after the event day, a 21-day

window, but for IPOs, the volatility 10 days before the event day does not exist because

the event day is also the second trading day for some IPOs.

II.5.B.2. Trading interference hypothesis test

The second hypothesis in evaluating the performance of price limits focuses on

the possibility of trading interference. If price limits prevent investors from trading, this

may cause intensified trading activity on the following day(s). As suggested by Chordia,

Roll, and Subrahmanyam (2001), we use daily trading volume, daily trading value, and

5 Wei and Chiang (2000) suggest a generalized method of moments (GMM) estimator for volatility in markets with daily price limits. Even though the estimator captures reliable volatility over a given period of time when both limit hit and non-hit exist, it is not applicable for determining the daily volatility.

79 daily number of transactions to measure trading activity. Daily trading volume is

measured by the number of units traded on a given day. The unit of trading volume is

1000 shares in the TSE. Daily trading value, the dollar amount of trading volume, is the

sum of the products of each trading volume and its corresponding trading price. Daily

number of transactions is the total number of transactions that occur on each day. For

each IPO, the daily trading volume, trading value, and number of transactions are scaled

by the 12-day (from t-1 to t+10) average trading volume, trading value, and number of

transactions, respectively. The means and medians of daily trading volume, trading

value, and number of transactions from t-1 to t+10 across IPOs are calculated in order to

investigate the daily trading activity change over the 12-day window.

II.5.B.3. Delayed price discovery hypothesis test

In order to examine the delayed price discovery, we adopt a method suggested by

d Choi and Lee (2001). Let rt be the daytime (open-to-close) return on event day, t, and

n d rt +1 be the overnight (close-to-open) return from t to t+1, one day after the event day. rt

n and rt +1 are calculated as follows.

 P c  r d = n t  (3) t l o  Pt 

 P o  r n = n t+1  (4) t +1 l c  Pt 

c o where Pt is the closing price on event day t, Pt is the opening price on event day t, and

o Pt+1 is the opening price on t+1. Stock returns can be positive (+), negative (-) or zero

80 d n (0). For upper limit hits, we classify the set of {[rt , rt +1 ] | [+, +], [0, +]} as price

d n d continuations, the set of {[ rt , rt +1 ] | [+, -], [0, -]} as price reversals, and the set of {[rt ,

n rt +1 ] | [+, 0], [0, 0]} as no price changes around the closing. As to lower limit hits, we

d n d n classify the set of {[rt , rt +1 ] | [-, -], [0, -]} as price continuations, the set of {[ rt , rt +1 ] | [-

d n , +], [0, +]} as price reversals, and the set of {[ rt , rt +1 ] | [-, 0], [0, 0]} as no price changes

around the closing. Because price limits prevent stock prices from reaching their

equilibrium level, prices continue to move toward that level on the following day to

reflect the remaining information. Based on the information asymmetry argument, we

expect to see more price continuations from IPOs than from MSEs because trading is

interfered at a higher degree for IPOs than MSEs.

We also examine how effective price limits are in conveying new information by

investigating stock return behavior around the event day using event study methodology.

In this analysis we use the daily individual stock returns and daily returns of the TSEC

Capitalization Weighted Price Index (TAIEX). Returns are measured by logarithmic

price differences adjusted for cash dividends and stock splits. Excess returns are

calculated based on the market adjusted returns model. We define the abnormal return of

firm i on day t (ARi,t), as:

ARit = Rit – Rmt (5)

where Rit is the observed return for security i on day t, and Rmt is the return of the TAIEX on day t. In the literature, alternative methods have been used to detect abnormal returns, like the market model or the mean-adjusted returns model. We choose the market- adjusted returns model based on the following reasons. First, for our IPO sample, it is difficult to apply the other two models because the estimation period needed to generate

81 expected or normal returns is not available. Furthermore, Brown and Warner (1985)

show that the abilities of the three models to correctly detect the presence of abnormal

performance are similar when analyzing non-clustered daily returns data.

The daily average abnormal return (AARt) for a given day t across n stocks is defined as:

1 n AARt = ∑ ARi,t (6) n i=1

Based on the event window [t-1, t+10], we compute the CAAR from a set of windows embedded in this event window. The Cumulative Average Abnormal Return in the window (T1, T2) ( CAAR ) is: (T1,T2 )

T2 CAAR = AAR (7) (T1,T2 ) ∑ t t=T1

We perform t-tests to determine the statistical significance of the abnormal returns. If

price limits delay the price discovery process, we expect to find significant positive

abnormal returns after upper limit hits and negative abnormal returns after lower limit

hits. Since IPOs are associated with high degree of information asymmetry, we also

expect to see high abnormal returns for IPOs than for MSEs on day t+1.

II.6. Results

II.6.A. Information asymmetry hypothesis

82 Table IV reports the medians of three different volatility measures6 from t-1 to

t+10 for IPOs and MSEs. Since the Shapiro-Wilk normality test of the distribution of

volatilities is rejected, the Wilcoxon signed rank test is used to determine the level of

significance. Daily volatilities are calculated using both the measurements suggested by

Grossman (1988), VGT, and the extreme value method introduced by Parkinson (1980),

VPT to capture the transitory volatility. In addition, we also calculate VCT to measure the day-to-day movement. VCT is defined as follows.

2    CPT  VCT =  n  (8) l CP   T −1 

where CPT is the closing price on day T and CPT-1 is the closing price on day T-1, and ln is the natural log transformation.

Panel A reports the results for upper limit hits. When multiple hits are included, both VPT and VGT measures show that the volatility on t+1 is significantly higher for

IPOs than for MSEs. However, the VCT measure presents the opposite result. That is,

even though the price movement from t to t+1 is higher for MSEs, the intraday volatility

on t+1 is higher for IPOs. No significant differences on day t between IPOs and MSEs

are observed. This supports the information asymmetry hypothesis that price limits cause

higher volatility levels for IPOs than for MSEs on subsequent day(s). The volatility

differences disappear after t+3 from VPT measure and after t+2 from VGT measure. This

indicates that the information asymmetry inducement problem caused by price limits is

resolved within two days after the limit-hit day. Overall, the volatility reaches the highest

6 In order to examine whether the Asian financial crisis had any impact on the results, we also group daily volatility by periods prior to and after 1997, the year when the crisis began. However, no significant differences were observed.

83 level on t and gradually decreases to lower levels on subsequent days for both IPOs and

MSEs. That is, the limit-hit day is the most volatile day. There is also evidence showing

that the volatility on t-1 is higher for IPOs than for MSEs. Both the VPT and VGT

measures show that IPOs are more volatile than MSEs on t-1, but the difference is not

significant for the VGT measure.

When multiple hits are excluded, the result still holds. The information asymmetry hypothesis is supported because higher volatility is observed for IPOs than for

MSEs on t+1. However, the evidence is not as strong because the volatility difference from the VGT measure is not significant. Besides, the higher volatility of IPOs on t-1 no

longer exists when multiple hits are excluded. The result indicates that when multiple

hits occur, the volatility on t-1 for IPOs is higher than that for MSEs.

Panel B of Table IV reports the results for lower limit hits. Basically, the result is

similar to the upper limit hit case. The information asymmetry hypothesis is supported

because higher volatility is observed for IPOs than for MSEs on t+1. However, when

multiple hits are included, we observe higher volatility for MSEs than IPOs on t. One

possibility is that trading is interfered at a higher degree for IPOs on t when lower limit

hits occur than when upper hits occur. Because trading is interfered, the volatility is

suppressed and will be spilled over to the following day(s). Our analysis of trading

interference hypothesis in the next section supports this argument.

II.6.B. Trading interference hypothesis

84 Table V reports the medians of daily trading volume, trading value, and number of transactions from t-1 to t+10 for IPOs and MSEs. Like volatility, the Shapiro-Wilk normality tests of the distribution of trading volume, trading value, and number of transactions are rejected, so the Wilcoxon signed rank test is used to determine the level of significance. Panel A reports the results for upper limit hits. All three measures show that trading activity is higher for IPOs than for MSEs on both t-1 and t. No significant differences are observed on t+1, except for the measure from the number of transactions.

Since IPOs are subject to higher degree of information asymmetry, the trading interference hypothesis suggests that trading activity will be intensified on subsequent day(s) at a higher degree for IPOs than for MSEs. The lack of strong evidence of higher trading activity on t+1 for IPOs than for MSEs fails to support this hypothesis. When multiple hits are excluded, the results are similar. Again, the trading interference hypothesis is not supported. Overall, for both IPOs and MSEs, the trading activity reaches the highest level on t+1 and gradually decreases to lower levels on subsequent days.

Panel B of Table V reports results for lower limit hits. Unlike the case of upper limit hits, no significant differences of trading activity on t are observed. IPOs have higher trading activity than MSEs on t+1. That is, the trading interference hypothesis is supported when lower limit hits occur. When multiple hits are excluded, the results are similar. On the limit-hit day t, price limits reduce market liquidity and deprive traders the opportunity to change their portfolio allocations. Since trading is interfered on t, higher trading activity on t+1 when new price limits are established is expected. All three measures of trading activity support this trading interference hypothesis. Trading activity

85 reaches the highest level on t+1 and gradually decreases to lower levels on subsequent

days. However, when multiple hits are excluded, the highest trading activity takes place

on t-1 for IPOs. Our explanation is that when multiple hits occur, trading is interfered on

t-1 (also a limit-hit day) and thus lower trading activity on t-1 is expected. That is why

trading activity is lower when multiple hits are included than when they are excluded.

II.6.C. Delayed price discovery hypothesis

As to the delayed price discovery hypothesis, Table VI reports the proportions of

price continuations, reversals, and no change for both IPOs and MSEs. The significance

level is based on the Z-statistic from a standard binomial test. Although the proportions of price continuations of IPOs are higher than those of MSEs, the differences are not significant. The only exception is the case of upper limit hits when multiple hits are excluded. In this case, the proportion of price continuation of IPO firms is significantly higher than that of MSEs. The delayed price discovery hypothesis is supported because trading is interfered at a higher degree for IPOs than MSEs. However, the evidence is weak because only one of the sub-samples supports this hypothesis.

It is important to notice that for either IPOs or MSEs, the proportions of continuations are significantly higher than those of reversals and no changes. The delayed price discovery hypothesis is weakly supported in our previous discussion because we focus on the comparison between IPOs and MSEs. However, if we look at

IPOs and MSEs separately, the delayed price discovery hypothesis is strongly supported.

That is consistent with the finding from the Japanese study by Kim and Rhee (1997).

86 Table VII reports the daily average abnormal returns (AART) and the cumulative

average abnormal returns ( CAAR ) for IPOs and MSEs. If the delayed price (T1,T2 ) discovery hypothesis holds, we except to see abnormal return on days following the limit- hit day. Due to the information asymmetry, the AART of IPOs is expected to be higher

for upper limit hits and lower for lower limit hits than that of MSEs. However, the

results show the opposite. For upper limit hits, the AART of MSEs is significantly higher than that of IPOs on t+1 and t+2. For lower limit hits, the AART of IPOs is significantly

higher than that of MSEs on t+1 when multiple hits are excluded. That is, the evidence

does not support the delayed price discovery hypothesis. Even though IPOs are subject

to higher degree of information asymmetry than are MSEs, the price discovery process

seems to be delayed at a higher degree for MSEs. When prices hit the upper limits, they

tend to generate positive abnormal returns on the following day for MSEs. On the other

hand, when prices hit the lower limits, they tend to generate negative abnormal returns on

the following day. Apparently, the price discovery process is delayed for MSEs.

However, for IPOs, negative abnormal returns are observed after hitting the upper limits.

This IPO evidence is consistent with the cooling-off effect of price limits.

Table VII also shows that, for upper limit hits, both the CAAR (t-1, t+2) and

CAAR (t-1, t+10) are significantly higher for MSEs than for IPOs. When multiple hits

are excluded, all three windows of CAAR are significantly higher for MSEs than for

IPOs. As to lower limit hits, higher CAAR is observed for IPOs only when multiple hits are excluded. Since we do not examine the reasons for the big moves on the event day for IPOs and MSEs, there is no way we can figure out why MSEs have higher CAAR for upper limit hits and lower CAAR for lower limit hits than IPOs.

II.6.D. Summary results

Table VIII summarizes the testing results for our three hypotheses. Strong support means that the hypothesis is supported and the results are robust across different measurements and methodology. Weak support means that the hypothesis is supported but the result is sensitive to the measurements and methodology. No support means that the hypothesis is not supported.

The information asymmetry hypothesis suggests that price limits cause higher volatility levels for IPOs than for MSEs on days following the limit-hit day. We find strong support for the information asymmetry hypothesis when multiple hits are included and only weak support when multiple hits are excluded. Basically, the volatility of IPOs is higher than that of MSEs on t+1. This suggests that price limits induce information asymmetry and lead to higher volatility after limit hits.

The trading interference hypothesis suggests that price limits cause higher trading activity for IPOs than for MSEs on days following the limit-hit day. We find strong support for the trading interference hypothesis for lower limit hits but no support for upper limit hits. When prices reach the lower limits, trading is interfered and intensified trading activity is observed on the next trading day. However, when prices reach the upper limits, no evidence of trading interference is observed. This asymmetric property of price limits between upper and lower limit hits is also observed by Choi and Lee

(2001) and Cho et al. (2003).

88 The delayed price discovery hypothesis indicates that price limits prevent prices from efficiently reaching their equilibrium levels and thus suggests that more price continuations will be observed from IPOs than from MSEs following limit hits. Besides, we also adopt the event methodology to measure the speed of price discovery given price limits. On the limit-hit day, the abnormal return cannot be fully reflected if price limits delay price discovery. Therefore, we except to see abnormal return on days following the limit-hit day. Due to the information asymmetry, the abnormal return of IPOs is expected to be higher for upper limit hits and lower for lower limit hits than that of MSEs. We find basically no support for the delayed price discovery hypothesis except for the case of upper limit hits when multiple hits are excluded. In this case, we observe more price continuations from IPOs. However, the abnormal return analysis does not support this hypothesis under any circumstances.

II.7. Conclusion

Are price limits effective? We do not know because only limited studies have tried to answer this question and their results are conflicting. Besides, empirical studies are subject to the same criticism that we do not know what would have happened without price limits.

This paper examines IPOs in the Taiwan Stock Exchange in an effort to investigate the performance of price limits under information asymmetry. Since most countries in the Asia and Europe impose price limits on their stock markets, it is important to understand whether price limits can achieve their intended goals or not,

89 especially during periods with information shocks or high information asymmetry. Our results from IPOs can be used to predict the performance during these periods because we understand that, by nature, IPOs are subject to high information asymmetry.

In essence, the investigation covers issues such as volatility, trading activity, and price discovery. The evidence from IPOs in the TSE shows that, on average, it takes 6.24 days for IPOs to reach their equilibrium prices in the presence of the 7% price limit. The delayed price discovery is apparent for IPOs before the equilibrium price is reached. The information asymmetry hypothesis, which states that price limits cause higher volatility levels on subsequent days for IPOs, is supported. The trading interference hypothesis, which implies that price limits interfere with trading and cause higher trading activity on subsequent days for IPOs, is supported by lower limit hits. The delayed price discovery hypothesis is not supported by our data. Overall, our results represent the performance of price limits for IPOs and can be used to predict the performance of price limits during periods with high information asymmetry.

Our results also generate important policy implications for the Taiwan Stock

Exchange. With price limits, the price discovery process is delayed during the period from the first trading day to the day when equilibrium price is reached due to the high degree of IPO underpricing. One way to circumvent the ineffectiveness is to impose price limits on the first opening price instead of the offer price of IPOs. If price limits are applied to the first opening price, which is currently the case in Japan and Korea, the equilibrium price is more likely to be achieved on the first trading day. As we have shown, it takes 6.24 days for IPOs in TSE to reach their equilibrium prices given the current price limit policy. The cost associated with the low liquidity observed during

90 those days could be reduced if price limits were applied to the first opening price rather

than the offer price. Therefore, based on this study, we recommend that regulators of the

TSE as well as other countries with similar regulations, such as France7, should

reevaluate the current policy and adjust it accordingly to avoid the unnecessary costs

associated with the low market liquidity of IPOs.

Further studies on the performance of price limits are encouraged in order to

complete a thorough picture of the price limit issue. Recent increases of the price-limit

rate in both Korea (1998) and Thailand (1997) cast doubt on the effectiveness of price

limits from the regulators’ perspectives. Starting from 2001, Spain also modifies its price

limit system to one that combines both price limits and trading halts. That is, when price

limits are hit, a certain period of trading halt is also triggered. If price limits are indeed

effective, the increase of the price-limit rate will reduce its effectiveness because the

chance to hit the limit is lower. On the other hand, if price limits are ineffective, the

increase of price-limit rate does not help because the existence of price limits is

unnecessary. In addition, whether a price-based limit system or a fixed-rate limit system

is better is worth examining.

7 In French stock market, a 10% price limit is applied to the offering prices for IPOs. If the potential clearing price is higher or less than the offering price by more than 10%, no transaction occurs on the first trading day. In this case, trading is postponed to the following day(s) till a clearing price within the price limit is found. We thank Francois Derrien for providing the regulatory detail.

91 Appendix A Adjustments of Price-limit Rates Over Time at the Taiwan Stock Exchange

Date Price-limit Rates Change Reasons

2/9/1962 5% upward and downward Centralized market began operation, to protect investors 4/9/1973 3% upward and downward Stock markets too volatile 8/7/1973 5% upward and downward Stock markets back to normal 2/19/1974 5% upward and 1% downward Oil crisis, prevent stocks from falling 3/9/1974 5% upward and one tick size downward Prevent stocks from falling 4/15/1974 1% upward and downward Prevent stocks from falling 5/21/1974 3% upward and downward Falling trend alleviated 6/17/1974 5% upward and downward Stock markets back to normal 12/19/1978 2.5% upward and downward Diplomatic relation between U.S. and China, prevent stocks from falling 1/5/1979 5% upward and downward Stock markets back to normal 10/27/1987 3% upward and downward Global market crash, prevent stocks from falling 11/14/1988 5% upward and downward Requested by investors and the press 10/11/1989 7% upward and downward Requested by investors and the press 9/27/1999 7% upward and 3.5% downward Sep. 21, 1999 earthquake, prevent stocks from falling 10/8/1999 7% upward and downward Stock markets back to normal 3/20/2000 7% upward and 3.5% downward Presidential election, prevent stocks from falling 3/27/2000 7% upward and downward Stock markets back to normal

Source: Taiwan Stock Exchange Corporation

92 Appendix B Tick Size for each Price Range

Price Range Tick Size

P < NT$ 5.00 NT$ 0.01

NT$ 5.00 ≤ P < NT$ 15.00 NT$ 0.05

NT$ 15.00 ≤ P < NT$ 50.00 NT$ 0.10

NT$ 50.00 ≤ P < NT$ 150.00 NT$ 0.50

NT$ 150.00 ≤ P < NT$ 1,000.00 NT$ 1.00

NT$ 1,000.00 ≤ P NT$ 5.00

Source: Taiwan Stock Exchange Corporation

93 References II

Aggarwal, R., 2000, Stabilization activities by underwriters after initial public offerings, Journal of Finance 55, 1075-1103.

Arak, M. and R.E. Cook, 1997, Do daily price limits act as magnets? The case of Treasury bond futures, Journal of Financial Services Research 12, 5-20.

Berkman, H. and O.W. Steenbeek, 1998, The influence of daily price limits on trading in Nikkei futures, Journal of Futures Markets 18, 265-279.

Brady, N. et al., 1988, Report of the Presidential Task Force on Market Mechanisms (Washington, D.C.: Government Printing Office).

Brennan, M. J., 1986, A theory of price limits in futures markets, Journal of Financial Economics 16, 213-233.

Brown, S.J. and J.B. Warner, 1985, Using daily stocks returns: The case of Event Studies, Journal of Financial Economics 14, 3-31.

Carter, R. and S. Manaster, 1990, Initial public offerings and underwriter reputation, Journal of Finance 45, 1045-1067.

Cho, D.D., J. Russell, G.C. Tiao, and R. Tsay, 2003, The magnet effect of price limits: evidence from high-frequency data on Taiwan Stock Exchange, Journal of Empirical Finance 10, 133-168.

Choi, W.S. and S.B. Lee, 2001, A transitory and asymmetric properties in price limits: Korean evidence with intra-day data, Unpublished working paper, Han Yang University, Seoul.

Chordia, T., R. Roll, and A. Subrahmanyam, 2001, Market liquidity and trading activity, Journal of Finance 56, 501-530.

Christie, W.G., S.A. Corwin, and J.H. Harris, 2002, Nasdaq trading halts: the impact of market mechanisms on prices, trading activity, and execution costs, Journal of Finance 57, 1443-1478.

Corwin, S.A. and M.L. Lipson, 2000, Order flow and liquidity around NYSE trading halts, Journal of Finance 55, 1771-1801.

Fama, E.F., 1989, Perspectives on October 1987, or, What did we learn from the crash?, in: R.W. Kamphuis, R.C.Kormendi, and J.W.H. Watson, eds., Black Monday and the future of the financial markets (Irwin, Homewood, IL.) 71-82.

94 Garman, M.B. and M.J. Klass, 1980, On the estimation of security price volatilities from historical data, Journal of Business 53, 67-78.

Grossman, S.J., 1988, Program trading and market volatility: A report on interday relationships, Financial Analysts Journal July-August, 18-28.

Harris, L.E., 1998, Circuit breakers and program trading limits: what have we learned?, in: R.E. Litan and A.M. Santomero, eds., Brookings-Wharton papers on financial services (Brookings Institutions Press, Washington D. C.).

Hebner, K.J. and T. Hiraki, 1993, Japanese initial public offerings, in: I. Walter and T. Hiraki, eds., Restructuring Japan’s financial markets (Business One/Irwin, Homewood, IL.).

Huang, Y.S., 1999, The price behaviour of initial public offerings on the Taiwan Stock Exchange, Applied Financial Economics 9, 201-208.

Kao, G.W. and C.K. Ma, 1992, Memories, heteroscedasticity, and price limit in currency futures markets, Journal of Futures Markets 12, 679-692.

Kim, K.A. and P. Limpaphayom, 2000, Characteristics of stocks that frequently hit price limits: Empirical evidence from Taiwan and Thailand, Journal of Financial Markets 3, 315-332.

Kim, K.A. and S.G. Rhee, 1997, Price limit performance: Evidence from the Tokyo Stock Exchange, Journal of Finance 52, 885-901.

Kodres, L.E., and D.P. O’Brien, 1994, The existence of Pareto superior price limits, American Economic Review 84, 919-932.

Lauterbach, B. and U. Ben-Zion, 1993, Stock market crashes and the performance of circuit breakers: empirical evidence, Journal of Finance 48, 1909-1925.

Lee, C.M.C., M.J. Ready, and P.J. Seguin, 1994, Volume, volatility, and New York Stock Exchange trading halts, Journal of Finance 49, 183-214.

Lee, S.B. and K.J. Kim, 1995, The effect of price limits on stock price volatility: Empirical evidence from Korea, Journal of Business Finance & Accounting 22, 257- 267.

Lehmann, B.N., 1989, Commentary: volatility, price resolution, and the effectiveness of price limits, Journal of Financial Services Research 3, 205-209.

Loughran, T., J.R. Ritter, and K. Rydqvist, 1994, Initial public offerings: international insights, Pacific-Basin Finance Journal 2, 165-199.

95 Ma, C.K., R.P. Rao, and R.S. Sears, 1989, Volatility, price resolution, and the effectiveness of price limits, Journal of Financial Services Research 3, 165-199.

Miller, M.H., 1989, Commentary: volatility, price resolution, and the effectiveness of price limits, Journal of Financial Services Research 3, 201-203.

Parkinson, M., 1980, The extreme value method for estimating the variance of the rate of return, Journal of Business 53, 61-65.

Peristiani, S., 2003, Evaluating the riskiness of initial public offerings: 1980-2000, Unpublished working paper, Federal Reserve Bank of New York, New York.

Phylaktis, K., M. Kavussanos, and G. Manalis, 1999, Price limits and stock market volatility in the Athens Stock Exchange, European Financial Management 5, 69-84.

Rhee, S.G. and R.P. Chang, 1993, The microstructure of Asian equity markets, Journal of Financial Services Research 6, 437-454.

Ritter, J.R., 1987, The costs of going public, Journal of Financial Economics 19, 269- 281.

Rock, K., 1986, Why new issues are underpriced?, Journal of Financial Economics 15, 187-212.

Schwert, G.W., 1989, Why does stock market volatility change over time?, Journal of Finance 44, 1115-1154.

Subrahmanyam, A., 1994, Circuit breakers and market volatility: a theoretical perspective, Journal of Finance 49, 237-254.

Wei, K.C.J. and R. Chiang, 2000, A GMM approach for estimation of volatility and regression models when daily prices are subject to price limits, Unpublished working paper, HKUST, Hong Kong.

Welch, I., 1992, Sequential sales, learning, and cascades, Journal of Finance 47, 695- 732.

96 Figure 1: Timeline

IPO Date is the first trading date for each IPO. EP-reached Day stands for Equilibrium Price-reached Day and is defined as one day after the last limit-hit day. Number of Delayed Days is the total number of days from IPO Date to EP-reached Day. Line (a) represents the timeline from the IPO Date to the EP-reached Day. Line (b) represents the timeline for our event study after the EP was reached. If the first limit-hit day in Line (b) was followed by another limit-hit day(s), the last limit-hit day of those multiple hits is selected to be the event day, t.

Number of Delayed Days (6.24 Days)*

(a) IPO Date Last Limit-hit Day EP-reached Day

(b) - - - - - EP-reached Day First Limit-hit Day One Day after t Ten Days after t (Event Day t) (t+1) (t+10)

* This is the average number. The median of the number of Delayed Days is 4. Details are presented in Table II.

Figure 2: Distribution of Delayed Days for a sample of 294 IPOs from October 11, 1989 to July 31, 2000

The horizontal axis represents the number of Delayed Days, which ranges from 1 to 31. Number of Delayed Days is the total number of days from the first trading day to the Equilibrium Price-reached Day, which is defined as one day after the last limit-hit day. The vertical axis represents the number of observations, which ranges from 1 to 41.

30 ons

20 er of Observati b

Num 15

0 0 5 10 15 20 25 30 35 Delayed Days

98 Table I Summary Statistics of IPOs

There are 294 IPOs from October 11, 1989 to July 31, 2000 on TSE. The total number of IPO listings is reported by their issuing year. Limit-hit IPOs are those that hit the price limits on their first trading day. The number of upper limit-hit IPOs and that of lower limit-hit IPOs are reported by their issuing year. Non-hit IPOs are those that did not hit the price limits on their first trading day. Upward (downward) refers to those whose first closing price is higher (lower) than the offering price, while unchanged refers to those whose first closing price is the same as the offering price. % = (total number of observations / 294) * 100. Average market return is based on the TSEC Capitalization Weighted Price Index (TAIEX) on the first trading day of each IPO.

Year IPO listings Limit-hit IPOs Non-hit IPOs Upper Lower Upward Downward Unchanged 1989 13 13 0 0 0 0 1990 19 18 0 1 0 0 1991 24 23 0 1 0 0 1992 35 34 0 1 0 0 1993 27 24 0 2 1 0 1994 30 28 1 1 0 0 1995 39 28 5 4 2 0 1996 36 26 1 4 4 1 1997 21 17 2 1 1 0 1998 23 17 2 0 3 1 1999 19 12 1 1 4 1 2000 8 5 2 0 1 0 Total 294 245 14 16 16 3 % 100 83.33 4.76 5.44 5.44 1.02

Average market return 0.009% 0.258% 0.127% -0.159% 0.192%

99 Table II Number of Delayed Days

Since many IPOs consecutively hit the price limit after going public, the event day, t, for each IPO is defined as the last limit-hit day. EP stands for equilibrium price. Pre-EP refers to the period before EP was reached. The equilibrium price of each IPO was not reached till t+1, one day after the event day. The number of delayed days for each IPO is obtained from the number of t+1. If the price limit was not hit on the first trading day, the number of delayed days is one. The table reports the mean, median, and S.D. of the number of delayed days by year of listing. S. D. stands for standard deviation.

Year(s) Sample Size Number of Delayed Days Mean Median S. D. 1989 13 23.15 23.0 4.10 1990 19 11.68 10.0 7.48 1991 24 8.00 8.0 3.44 1992 35 6.80 6.0 3.30 1993 27 6.00 6.0 3.16 1994 30 5.03 4.5 3.31 1995 39 3.46 3.0 2.59 1996 36 4.06 3.0 3.57 1997 21 3.71 2.0 2.55 1998 23 3.96 3.0 2.72 1999 19 3.53 2.0 3.34 2000 8 6.50 5.5 4.54 ALL 294 6.24 4.0 5.56

100 Table III Market Capitalization

This table reports the market capitalization of limit-hit IPOs and MSEs. The market capitalization is defined as the last closing price of the limit-hit month multiplied by the number of shares outstanding at month end. The unit of market capitalization is millions of Taiwanese dollar. MSEs are industry-and-size matching seasoned equities. Since not all MSEs are industry matched, IND refers to the number of industry matched MSEs. Multiple hits occurred when two or more consecutive limit hits are observed. N is the sample size. The P-value of the test that the figure is equal to 0 is provided in the parenthesis. We use t-test for means and Wilcoxon signed rank test for medians.

Mean Median Limit Hit Multiple Hits N IND IPOs MSEs IPOs-MSEs IPOs MSEs IPOs-MSEs Upper Included 87 34 11,706 12,868 -1,162 5,537 6,273 -736 (0.7063) (0.9649) Lower Included 59 40 7,603 7,990 -387 5,165 4,639 526 (0.8211) (0.5843) Total Included 146 74 10,048 10,897 -849 5,217 5,426 -209 (0.6643) (0.7721)

Upper Excluded 65 26 11,805 11,822 -17 5,010 5,750 -740 (0.9958) (0.5937) Lower Excluded 41 27 8,058 6,657 1,401 4,770 4,117 653 (0.1241) (0.4426) Total Excluded 106 53 10,356 9,824 531 4,897 5,238 -341 (0.7947) (0.9797)

101 Table IV Volatility Analysis This table reports medians of daily volatility measures for IPOs and their industry-and-size matched seasoned equities (MSEs) from t-1, one 2 trading day prior to the event day t, to t+10, 10 trading days after t. The event day, t, is defined as the limit-hit day. VPT = (HPT – LPT) / 4ln2, 2 VGT = ln(HPT/LPT) and VCT = [ln(CPT/CPT-1)] where HPT is the high price, LPT is the low price, and CPT is the closing price on day T, while CPT-1 is the closing price on day T-1, one trading day prior to T. Multiple hits occur when two or more consecutive limit hits are observed. ***, **, and * indicate that the figure is higher than that of the comparison group on the same day at the 0.01, 0.05, and 0.10 levels of significance, respectively. Wilcoxon signed rank test is used to determine the level of significance.

Panel A: Upper limit hits Multiple hits included Multiple hits excluded

VPT VGT VCT VPT VGT VCT IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs t-1 **2.2542 1.4427 0.0410 0.0359 0.0008 0.0007 1.4427 1.1686 0.0355 0.0373 0.0005 0.0007 t 4.4183 3.2461 0.0592 0.0580 0.0046 0.0046 4.4183 2.2542 0.0629 0.0580 0.0046 0.0046 t+1 ***4.4183 2.2542 *0.0494 0.0437 0.0005 ***0.0012 **3.2461 1.4427 0.0460 0.0445 0.0003 ***0.0015 t+2 *2.2542 1.5906 0.0386 0.0438 0.0007 0.0009 2.0775 1.4427 0.0364 0.0438 0.0004 **0.0011 t+3 2.2542 1.5906 0.0432 0.0421 0.0004 0.0004 1.7457 1.4427 0.0405 *0.0421 0.0003 *0.0004 t+4 1.4427 1.5906 0.0360 *0.0435 0.0005 0.0005 1.4427 1.4427 0.0343 **0.0447 0.0006 0.0005 t+5 2.2542 1.4427 0.0319 *0.0403 0.0003 0.0004 1.4427 1.4427 0.0284 ***0.0388 0.0003 0.0004 t+6 1.4427 1.4427 0.0337 *0.0415 0.0004 0.0005 1.4427 1.4427 0.0293 **0.0418 0.0003 0.0005 t+7 1.4427 1.4427 0.0307 *0.0390 0.0003 0.0003 1.4427 0.8115 0.0302 **0.0359 0.0003 0.0003 t+8 2.2542 1.4427 0.0296 0.0364 0.0002 **0.0006 0.8115 1.0423 0.0283 0.0379 0.0002 0.0004 t+9 2.2542 1.3020 0.0333 0.0343 0.0002 *0.0005 1.4427 0.8115 0.0294 *0.0332 0.0001 ***0.0005 t+10 1.4427 1.4427 0.0290 ***0.0348 0.0002 0.0003 1.4427 1.4427 0.0282 ***0.0361 0.0002 ***0.0004

102

Panel B: Lower limit hits Multiple hits included Multiple hits excluded

VPT VGT VCT VPT VGT VCT IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs t-1 **3.2461 1.4427 0.0572 0.0434 0.0016 0.0009 ***4.4183 1.4427 *0.0545 0.0412 0.0008 0.0004 t 2.2542 2.8277 0.0513 *0.0594 0.0046 0.0046 3.2461 3.4661 0.0562 0.0594 0.0046 0.0046 t+1 *3.4661 1.9080 *0.0597 0.0588 0.0006 ***0.0025 **3.2461 1.9080 0.0559 0.0548 0.0006 ***0.0012 t+2 1.4427 1.4427 0.0426 0.0496 0.0008 0.0011 *1.4427 0.9233 0.0366 0.0442 0.0004 0.0008 t+3 1.3020 1.4427 0.0442 0.0480 0.0012 0.0018 1.4427 1.4427 0.0431 0.0426 0.0008 0.0010 t+4 1.9080 1.4427 0.0433 0.0473 0.0006 0.0012 1.4427 1.4427 0.0385 0.0449 0.0003 0.0012 t+5 0.9233 1.4427 0.0317 *0.0438 0.0004 ***0.0015 0.8115 1.4427 0.0301 **0.0425 0.0002 *0.0014 t+6 1.4427 1.4427 0.0386 0.0460 0.0003 0.0011 1.4427 0.8115 0.0348 0.0405 0.0003 0.0011 t+7 1.0423 1.1686 0.0377 0.0465 0.0004 ***0.0021 0.8115 0.8115 0.0367 *0.0476 0.0004 **0.0013 t+8 1.0423 0.8115 0.0351 *0.0388 0.0005 0.0005 0.8115 0.7069 0.0290 **0.0354 0.0003 0.0002 t+9 1.4427 0.9233 0.0462 0.0457 0.0009 0.0009 1.4427 0.8115 0.0363 0.0382 0.0008 0.0007 t+10 1.4427 1.0423 0.0364 0.0403 0.0005 ***0.0014 0.8115 0.8115 0.0288 *0.0409 0.0003 ***0.0014

103 Table V Trading Activity Analysis

This table reports the medians of daily trading volume, trading value, and number of transactions for IPOs and their industry-and-size matched seasoned equities (MSEs) from t-1, one trading day prior to the event day t, to t+10, 10 trading days after t. The event day, t, is defined as the limit-hit day. The unit of trading volume is 1000 shares in TSE. For each observation, the daily trading volume, trading value, and number of transactions are scaled by the 12-day (from t-1 to t+10) average trading volume, trading value, and number of transactions, respectively. Multiple hits occur when two or more consecutive limit hits are observed. ***, **, and * indicate that the figure is higher than that of the comparison group on the same day at the 0.01, 0.05, and 0.10 levels of significance, respectively. Wilcoxon signed rank test is used to determine the level of significance.

Panel A: Upper limit hits Multiple hits included Multiple hits excluded Volume Value Transactions Volume Value Transactions IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs t-1 ***1.0109 0.7196 ***0.9510 0.6899 ***1.0641 0.7124 **0.9783 0.7151 **0.9743 0.6341 ***1.1244 0.7124 t **1.5221 1.3784 **1.5118 1.3204 ***1.4439 1.2865 ***1.5579 1.3525 ***1.5750 1.2979 ***1.5028 1.2754 t+1 1.5524 1.5342 1.5738 1.5118 *1.5830 1.4370 1.5526 1.5289 1.5980 1.5028 1.6051 1.4370 t+2 0.9925 1.1218 1.0243 1.1679 1.0533 1.1669 0.9860 1.1218 1.0119 1.1146 1.0372 1.1264 t+3 0.8174 0.9875 0.8277 0.9756 0.8789 *1.0090 0.8139 0.9956 0.8277 0.9756 0.8587 **1.0224 t+4 0.8451 0.8169 0.8285 0.8289 0.8825 0.9101 0.8295 0.8291 0.8213 0.8348 0.8592 0.8782 t+5 0.8245 0.8325 0.8124 0.8218 0.8518 0.8126 0.8163 0.8233 0.7867 0.7823 0.8226 0.8073 t+6 0.7236 0.7458 0.7127 0.7407 0.7793 0.8082 0.7236 0.7971 0.7127 0.7840 0.7817 0.8366 t+7 0.6303 0.6874 0.6242 0.6648 0.6745 **0.7839 0.6383 0.6976 0.6193 0.6862 0.6673 **0.8043 t+8 0.6062 0.6642 0.5821 0.6338 0.6561 0.7391 0.6062 0.6400 0.5948 0.6270 0.6561 0.7211 t+9 0.6761 0.7298 0.6473 0.7002 0.6651 0.7908 0.6200 0.6898 0.6140 0.6559 0.6167 0.7730 t+10 0.6290 0.6903 0.6028 0.7256 0.6485 0.7637 0.6290 0.6847 0.6028 0.7033 0.6489 0.7763

104

Panel B: Lower limit hits Multiple hits included Multiple hits excluded Volume Value Transactions Volume Value Transactions IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs t-1 ***1.2300 0.8995 **1.3061 0.9650 ***1.3245 0.9315 ***1.4369 0.9340 **1.5402 0.9800 ***1.5244 0.9744 t 0.9921 0.9938 0.9528 1.0064 1.0666 1.0351 1.0149 0.9987 1.0052 1.0064 1.1179 0.9980 t+1 ***1.3328 1.0114 ***1.3434 1.0423 ***1.4358 1.0502 **1.3258 0.9772 **1.2844 1.0423 ***1.3505 1.0502 t+2 0.9853 0.9383 0.9820 0.9280 0.9309 1.0568 0.9435 0.9787 0.9212 1.0398 0.9245 1.0568 t+3 0.8585 **1.0488 0.8175 **1.0376 0.9041 *1.0270 0.8186 ***0.9444 0.7813 ***1.0253 0.8645 **1.0557 t+4 0.8680 0.9020 0.8145 0.8852 0.9269 0.9491 0.8244 0.9015 0.8110 0.8397 0.8756 0.9351 t+5 0.7293 0.8108 0.7202 0.7968 0.7658 0.8665 0.7434 0.8575 0.7268 0.8264 0.7653 0.8980 t+6 0.7127 0.8802 0.6897 0.8486 0.7153 0.8843 0.6745 0.8802 0.6466 0.8337 0.6762 0.8843 t+7 0.7791 0.8478 0.7781 0.8053 0.7951 **0.8460 0.7689 0.7937 0.7505 0.7494 0.7947 0.8326 t+8 0.7367 0.7784 0.6921 0.7473 0.6698 **0.8483 0.6743 0.7784 0.6633 0.7473 0.6036 **0.8615 t+9 0.6582 0.7349 0.6130 0.7261 0.6751 **0.8232 0.6051 0.7136 0.5646 0.6925 0.6362 *0.7983 t+10 0.7199 0.7372 0.7347 0.7397 0.7922 0.8519 0.7164 0.7372 0.6600 0.7397 0.7922 *0.8519

105 Table VI Price Discovery Analysis

This table reports the proportions of continuations, reversals, and no change for IPOs and their industry-and-size matched seasoned equities (MSEs). Multiple d c o n o c c hits occur when two or more consecutive limit hits are observed. The event day, t, is the limit-hit day. Let rt = ln ( Pt / Pt ) and rt+1 = ln ( Pt+1 / Pt ), where Pt o o is the closing price on event day t, Pt is the opening price on event day t, and Pt+1 is the opening price on t+1. Stock returns can be positive (+), negative (-) or d n d n zero (0). For upper limit hits, we classify the set of {[rt , rt+1 ] | [+, +], [0, +]} as price continuations, the set of {[rt , rt+1 ] | [+, -], [0, -]} as price reversals, and the d n d n set of {[rt , rt+1 ] | [+, 0], [0, 0]} as no changes in prices around the closing. As to lower limit hits, we classify the set of {[rt , rt+1 ] | [-, -], [0, -]} as price d n d n continuations, the set of {[rt , rt+1 ] | [-, +], [0, +]} as price reversals, and the set of {[rt , rt+1 ] | [-, 0], [0, 0]} as no changes in prices around the closing. The proportions may not add to 1 due to rounding error. ***, **, and * indicate that the figure is higher than that of the comparison group under the same category (Continuation, Reversal or No Change) at the 0.01, 0.05, and 0.10 levels of significance, respectively. The significance levels are based on the Z-statistic from a standard binomial test.

IPOs MSEs Limit Hit Multiple Hits Sample Size Continuation Reversal No Change Continuation Reversal No Change Upper Included 87 0.70 0.13 0.17 0.64 0.18 0.17 Lower Included 59 0.63 0.17 0.20 0.64 0.14 0.22 Total Included 146 0.67 0.14 0.18 0.64 0.16 0.19 Upper Excluded 65 *0.74 0.12 0.14 0.60 0.23 0.17 Lower Excluded 41 0.68 0.15 0.17 0.66 0.10 0.24 Total Excluded 106 0.72 0.13 0.15 0.62 0.18 0.20

106 Table VII Abnormal Return Analysis

This table reports the daily average abnormal returns (AART) and the cumulative average abnormal returns

( CAAR ) in the window (T1, T2) for IPOs and their industry-and-size matched seasoned equities (T1,T2 ) (MSEs) from t-1, one trading day prior to the event day t, to t+10, 10 trading days after t. The event day, t, is defined as the limit-hit day. Multiple hits occur when two or more consecutive limit hits are observed. ARiT = RiT – RmT , where RiT is the observed return for security i on day T, and RmT is the return of the market T 1 n 2 index (TAIEX) on day T. AAR = AR and CAAR = AAR ***, **, and * indicate that t ∑ i,t (T1,T2 ) ∑ t n i=1 t=T1 the figure is higher than that of the comparison group on the same day at the 0.01, 0.05, and 0.10 levels of significance, respectively. T-test is used to determine the significance level.

Abnormal returns Multiple hits included Multiple hits excluded Upper limit hits Lower limit hits Upper limit hits Lower limit hits IPOs MSEs IPOs MSEs IPOs MSEs IPOs MSEs t-1 0.0116 0.0110 -0.0166 -0.0133 -0.0041 ***0.0093 -0.0066 -0.0130 t 0.0611 0.0615 -0.0399 -0.0400 0.0635 0.0639 -0.0405 -0.0407 t+1 -0.0003 *0.0084 -0.0073 -0.0148 -0.0038 *0.0086 *-0.0059 -0.0178 t+2 -0.0097 **0.0006 *0.0014 -0.0082 -0.0078 *0.0041 *0.0046 -0.0078 t+3 -0.0013 -0.0055 0.0011 -0.0055 -0.0007 -0.0061 0.0020 -0.0054 t+4 -0.0030 -0.0025 0.0005 -0.0084 -0.0058 -0.0027 **0.0007 -0.0139 t+5 0.0006 0.0024 -0.0022 -0.0081 -0.0014 -0.0005 -0.0021 -0.0069 t+6 0.0003 -0.0004 -0.0023 -0.0082 -0.0008 0.0001 -0.0044 -0.0098 t+7 -0.0017 0.0030 0.0004 0.0014 -0.0048 *0.0020 0.0011 0.0038 t+8 0.0050 0.0076 -0.0021 -0.0027 0.0055 0.0086 -0.0010 -0.0028 t+9 0.0014 0.0027 -0.0044 -0.0025 0.0008 0.0041 -0.0006 0.0035 t+10 -0.0010 0.0026 -0.0023 -0.0006 -0.0019 0.0015 -0.0030 0.0000

(t-1, t+1) 0.0724 0.0810 -0.0638 -0.0681 0.0556 ***0.0818 *-0.0529 -0.0715 (t-1, t+2) 0.0626 **0.0816 -0.0624 -0.0763 0.0478 ***0.0859 **-0.0483 -0.0794 (t-1, t+10) 0.0630 *0.0915 -0.0737 -0.1111 0.0386 ***0.0928 *-0.0557 -0.1109

107 Table VIII Summary Table

This table summarizes the testing results for our three hypotheses. Strong support means that the hypothesis is supported and the results are robust across different measurements and methodology. Weak support means that the hypothesis is supported but the result is sensitive to our measurements and methodology. No support means that the hypothesis is not supported.

Limit Hit Multiple Hits Information asymmetry Trading Interference Delayed Price hypothesis Hypothesis Discovery Hypothesis

Upper Included Strong support No support No support Lower Included Strong support Strong support No support

Upper Excluded Weak support No support Weak support Lower Excluded Weak support Strong support No support

108 Chapter III: Price Limits and Overreaction

III.1. Introduction

Behavioral finance has cast doubt on the market efficiency hypothesis and provided empirical evidence as well as theoretical models of market overreaction. Since the 1987 market crash, the Brady Report (1988) and several academic researchers have suggested the imposition of “circuit breakers” to prevent the market from fluctuating excessively due to investor overreaction. Even though the New York Stock Exchange (NYSE) implemented the circuit breaker mechanism on October 19, 1988, because the circuit breakers are rarely activated8, we cannot be certain whether the circuit breakers actually

reduce market overreaction. However, many countries in Asia have been imposing price

limits, one form of circuit breakers, for decades. In general, price limits regulate the

magnitude of the change in price that can occur for a given asset during a single trading

session. The purpose of this paper is to examine whether price limits can reduce market

overreaction by investigating a natural experiment from the Taiwan Stock Exchange

(TSE), one of the major stock markets in Asia.

The most popular rationale for imposing price limits is to reduce market overreaction.

From the proponents’ point of view, price limits can provide a cooling-off period that

allows investors to re-evaluate market information and make more rational decisions

during periods of extreme price changes. Hence, price limits can reduce traders’

overreaction and diminish price volatility. However, opponents of price limits argue that

8 Circuit breakers were triggered for the first and only time on Oct. 27, 1997, when the DJIA fell 350 points at 2:35 p.m. and 550 points at 3:30 p.m. That reflected an approximate 7% overall decline and the market was closed for the remainder of the day. See Goldstein and Kavajecz (2000) for more detail.

109 they serve no purpose other than to slow or delay the price discovery process (see, e.g.,

Fama, 1989). Even though price limits can stop the price of a share from falling or rising

at the limit on a given trading day, they argue that the price will continue to move in the

direction towards equilibrium as new trading limits are established in subsequent trading

day(s). Furthermore, rather than generating a stabilizing effect that calms market

movements, price limits may have a magnet effect that acts to pull prices toward the

limit. When prices move toward the limits, traders may rush to trade for fear that orders

might not be executed if the limit is hit. That is, price limits induce investors’

overreaction when prices are approaching the limits. The resolution of these two contrary

arguments relies on empirical evidence.

The empirical literature does not definitively answer whether price limits reduce or

induce overreaction. Studies that test directly for the effect of price limits use relatively

small data sets and reach different conclusions. Besides, most studies examine the

futures market rather than the stock market because there are price limits on futures

markets but not on the stock market in the U.S.9. Studies investigating stock markets use

data from Asian markets and focus only on daily stock prices. However, to determine

whether price limits reduce or induce overreaction, transaction data rather than daily data

should be used because the transitory volatility caused by overreaction is more clearly

reflected in transaction data. For example, if the significant change in volatility occurs

only during the first hour of trading before and/or after a limit hit, the use of daily data

would fail to detect the real impact of price limits. Therefore, this paper examines

9 We cannot generalize the results from futures markets to the stock market because of the different characteristics between them. For example, the difference in the degree of margin requirement, the mark- to-market feature in futures markets, and the market makers’ obligation to provide liquidity on NYSE may affect investors’ trading behavior between the futures markets and the stock market.

110 transaction data in an effort to provide insight into the ongoing debate over the relation between price limits and overreaction.

We test two hypotheses to investigate whether price limits reduce or induce overreaction. The first hypothesis is called the cooling-off hypothesis, which suggests that price limits can reduce overreaction because they provide a cooling-off period for investors to re-evaluate market information and make more rational trading decisions. To test this hypothesis, we identify three different limit hits, namely, closing limit hits, single limit hits, and consecutive limit hits. A closing limit hit occurs when a price hits the limit and no other trades occur the remainder of the day. A single limit hit occurs when a limit hit is followed by non-limit-hit transactions. Consecutive limit hits occur when a limit hit is followed by further trades at the limit price. The cooling-off hypothesis is only supported by consecutive limit hits. The second hypothesis is called the magnet hypothesis, which suggests that price limits induce overreaction because investors may rush to submit orders when prices are approaching the limits, even if those orders do not meet their optimal trading strategy. We test this hypothesis by examining return autocorrelations, trading volume, and relative spread.10 We find support for the magnet hypothesis in our measures of trading volume and relative spread, but no support from the return autocorrelations. Overall we conclude that price limits induce overreaction when prices are approaching the limits, but they also reduce overreaction when the limit price is traded at consecutively.

This paper makes several contributions. This is the first paper to examine the relation between price limits and overreaction using transaction data. We believe that the use of

10 Description and formula are presented in Section IV.

111 transaction data better captures the transitory volatility caused by price limits than does daily data. Second, the empirical results support opponents’ magnet effect argument and demonstrate that the proponents’ cooling-off effect argument is valid only for consecutive limit hits. Third, our method of distinguishing among different types of limit hits provides important insights into the effect of price limits on trading behavior. Lastly, our findings have important regulatory implications. Since price limits can reduce and induce overreaction, policy makers need to evaluate the net effect from price limits and set a rule to optimally reduce overreaction. For example, since the cooling-off hypothesis is supported by the consecutive limit hits, would a hybrid or combination of price limits and trading halts reduce overreaction more efficiently than only price limits or only trading halts?

The remainder of this paper is organized as follows. The next section provides the institutional background of the TSE. Section II discusses the theory and relevant empirical literature. Section III proposes two testing hypotheses. Section IV describes the data and sets out the research design. Section V presents the empirical findings and section VI concludes.

III.2. Institutional Background

According to the Monthly Bulletin of Statistics of the Republic of China (March,

2001), 531 stocks are listed on the TSE at the end of 2000. The total market value is

New Taiwanese Dollar (NT$) 8,191,170 million (or about US$ 248.5 billion at the exchange rate of NT$ 32.96/US$). The average daily trading value is NT$112,640

112 million (US$ 3.4 billion), with daily trading value being the sum of the product of trading volume and the trading price for every transaction on any given trading day.

The TSE is an order-driven market with no market makers or specialists. Investors can submit either market orders or limit orders. Orders are accumulated and matched against each other via the automated central limit order book. Since there are no official market makers, the bid and ask quotations are the best prices in the limit order book provided by various traders. According to the Taiwan Stock Exchange Corporation

(TSEC), the open outcry system has been gradually replaced since August 1985 by a computer-aided trading system (CATS), and was eventually upgraded to a fully automated securities trading (FAST) system in 1993. The trading session of the centralized market is 9:00 a.m. to 12:00 p.m., Monday through Friday. On the first, third, and fifth (if there is one) Saturdays of each month, trading also takes place from 9:00 a.m. to 12:00 p.m. Thirty minutes before the market opens, orders can be submitted via security firms and are ranked based on price-time priority. The opening price is the one that maximizes trading volume. Following the opening, orders are matched on a periodic basis till the closing with each round of clearing taking around one minute. The actual time interval of each round of clearing may vary slightly according to trading intensity.

The TSE has been imposing daily price limits since its inception in 1962. The purpose of the price limits is to avoid excessive volatility and to protect investors by limiting potential daily losses. The TSE sets its daily price limits at a predetermined rate, both upward and downward, based on the previous day’s closing price. The price-limit rate has been adjusted up or down several times in accordance with the market conditions. Panel A of Table I provides the price-limit rates during different periods in

113 year 2000. For stocks listed on the TSE, tick sizes (the minimum allowable unit that a stock price may change) vary with market prices. Panel B of Table I reports the tick size for each price range. In addition to the daily price limits, the clearing price in each round of matching cannot exceed two tick sizes from the clearing price in the preceding round.

Stocks that hit their price limits are still allowed to trade as long as the transaction prices are within the limits. Thus, the TSE price limits are simply boundaries, not triggers for trading halts.

III.3. Literature Review

There are two important theoretical studies on price limits and circuit breakers.

Brennan (1986) provides a theory of price limits in futures markets. He shows that price limits may act as a partial substitute for margin requirements in ensuring contract performance without resorting to costly litigation. However, the implementation of a daily price limit imposes clear costs on market participants by prohibiting trades at prices outside the limits. Subrahmanyam (1994) provides a model showing that a circuit breaker may actually increase price variability, thus increasing the probability that the price will reach the circuit breaker bounds if it is already very close to the breaker limit.

If traders fear that a halt will occur before they can submit their orders, to increase the probability of execution they may submit them earlier than they would otherwise. This is called the gravitation effect or magnet effect.

Because there are no price limits imposed on the U.S. equity markets, most studies examine the futures markets to investigate the effects of price limits. Arak and Cook

114 (1997) examine the U.S. Treasury bond futures market to test whether the daily price limits act as magnets to pull prices toward the limit. They find that the proximity to the limit tends to cause a small price reversal. In other words, the daily price limits act as a stabilizer in the futures markets. Chen (1998) investigates 19 futures contracts to test the overreaction hypothesis. The author finds little evidence to support the hypothesis. By arguing that futures prices are extremely noisy in the opening and closing minute, Chen uses the difference between the closing price on the event day and the average of the opening, closing, daily high and low prices on the next day to measure overreaction.

However, Chen points out that transaction data appears to be superior to the average daily price in measuring overreaction.

Several studies examine the effects of price limits on stock prices. Kim and Rhee

(1997) examine daily stock price data from the Tokyo Stock Exchange from 1989 to

1992 to test the effectiveness of price limits and find evidence that supports the arguments advanced by the opponents of price limits, i.e. delayed price discovery, volatility spillover, and trading interference. They then conclude that the price limit system of the Tokyo Stock Exchange may be ineffective. Choi and Lee (2001) examine both the inter-day and intra-day data from the Korean market to investigate the transitory and asymmetric properties of price limits. Using variance ratio tests and the modified

Kim and Rhee (1997) method, they provide evidence of delayed price discovery due to price limits. They further show that the delayed price discovery and trading interference are transitory because they are resolved once the constraint of price limits is removed at the open on the next day following the limit days. More importantly, they identify the asymmetric feature of price limits by showing that price limits act differently on the

115 upper and lower limit activities. They find that criticisms of price limits are partially supported by upper limit moves while price limits are found effective in the case of lower limit moves. Because of this asymmetric effect, they suggest a price-limit system with an upper limit wider than the lower limit to enhance market efficiency and reduce market volatility. However, the appropriate magnitude of limit rates is left for future studies.

While this paper uses intra-day data, its major focus is on the first five transactions after the opening.

Recently, Cho et al. (2003) use the intraday data from Taiwan Stock Exchange to test the magnet effect of price limits. It is the first attempt to test the magnet effect using intraday data. They find a statistically and economically significant tendency for stock prices to accelerate toward the upper bound and weak evidence of acceleration toward the lower bound as the price approaches the limits. That is, the magnet effect is supported.

Based on Subrahmanyam (1994), if the price is very close to the limit, the price limit may actually increase price variability and the probability of the price crossing the limits because strategic traders may advance their trades to assure their ability to trade. Thus, the magnet effect is supported only if significant results are observed when the price is very close to the limit. Besides, the discarding of limit-hit observations in their study may underestimate the means, standard deviations and correlation coefficients, as pointed out by Chiang, Wei, and Wu (1990).

116 Several studies examine the effect of trading halts, another form of circuit breakers, using U.S. data. Corwin and Lipson (2000) study the order flow and liquidity around

NYSE trading halts. They find that limit order book depth near the quotes is unusually low before, during, and after trading halts, which reflects investors’ reduced willingness to supply liquidity at these times. They also find that the quoted spreads are unusually high at the reopen. In addition, they find evidence of a dramatic increase in quoted spreads prior to order imbalance halts. Gerety and Mulherin (1992) investigate NYSE data from 1933-1988 and find that closing volume is positively related to expected overnight volatility, while volume at the open is positively related to both the expected and unexpected volatility from the previous night. They then conclude that the desire of investors to trade prior to market closings indicates a cost of mandating circuit breakers.

Lee, Ready, and Seguin (1994) examine the effect of firm-specific NYSE trading halts on volume and price volatility. They find that the period after a trading halt is characterized by higher levels of both volume and volatility. That is, the trading halt is ineffective in reducing price volatility. Like the NYSE studies by Lee, Ready, and Seguin (1994) and

Corwin and Lipson (2000), the NASDAQ-based research by Christie, Corwin, and Harris

(2002) also finds that even with information transmission during the halt, post-halt volume and volatility are unusually high following NASDAQ halts. This consistency in volume and volatility patterns suggests that the response of investors to trading halts is independent of the market structure and halt mechanisms. In sum, these studies suggest that trading halts are ineffective in reducing volatility and transactions increase after halts.

117 III.4. Hypotheses

Two hypotheses emerge from the relationship between price limits and overreaction.

The first hypothesis is based on the belief that price limits can provide a cooling-off period for investors to gain more information and re-evaluate market conditions and thus can reduce overreaction. Therefore, it is called the cooling-off hypothesis. The second hypothesis is based on the argument that when prices are approaching the limits, market participants, fearing the inability to trade, will alter their trading strategies to make sure their desired positions are taken before price limits are hit. This is called the magnet hypothesis. Instead of reducing overreaction, this hypothesis suggests that price limits actually induce overreaction.

III.4.A. Cooling-off hypothesis (reduce overreaction)

For this cooling-off hypothesis, we are more interested in knowing what happens after price limits are hit. If price limits can provide traders a cooling-off period to obtain information, reassess the market price, and avoid overreaction, the degree of return volatility after a limit hit should be less than that before a limit hit. On the other hand, if price limits delay the price discovery process and interfere with trading, the degree of return volatility after a limit hit may be higher than that before a limit hit. Therefore, the hypothesis is supported if we observe lower return volatility after limit hits than before limit hits. Otherwise, it is rejected.

118 According to Easley and O’Hara (1987) and Stoll (1989), when liquidity providers perceive an increase in the degree of information asymmetry, they tend to widen the bid- ask spread to compensate for expected losses to informed traders. Even though these studies are based on quote-driven markets, Biais, Hillion, and Spatt (1995), Hamao and

Hasbrouck (1995), and Ahn, Bae and Chan (2001) investigate major order-driven markets in the world and find similar results. In an order-driven market, such as the

Tokyo Stock Exchange and the Paris Bourse, all liquidity is provided by traders who submit limit orders. The difference between the price of the lowest sell limit order and that of the highest buy limit order determines the effective bid-ask spread. The bid-ask spread represents expected compensation for the costs of supplying immediacy. Because limit order prices are fixed, investors face adverse selection risk due to the arrival of informed traders. Glosten (1994) shows that the existence of adverse selection costs generate positive bid-ask spreads in an order-driven trading environment. Therefore, bid- ask spreads can be used as the proxy for the degree of information asymmetry.

If investors overreact and price limits reduce overreaction by providing a cooling-off period for them to obtain more information, the degree of information asymmetry is expected to be lower after limit hits. Consequently, liquidity providers will face less adverse selection risks from informed traders after limit hits and, accordingly, narrow the bid-ask spreads. If we observe lower bid-ask spreads after limit hits than before limit hits, the cooling-off hypothesis is supported. Otherwise, the hypothesis is rejected.

III.4.B. Magnet hypothesis (induce overreaction)

119 Theoretical support of this hypothesis is modeled by Subrahmanyam (1994). The model suggests that price limits may increase price variability and the probability that the price will reach the limit if it is already very close to the limit. If the magnet effect holds, we should observe three phenomena. First, the rush of market participants to trade actually exacerbates the problem by pushing the price closer to the price limits. That is, when a price is approaching the limit, the possibility of hitting the limit will increase due to the magnet effect. In other words, instead of reducing overreaction, price limits actually induce overreaction. Second, under this hypothesis, market participants should have an increasing demand for liquidity as prices approach the limits. Accordingly, trading volume should increase. Third, because of the increasing demand of liquidity, the cost of liquidity is expected to rise. Brockman and Chung (1999) examine an order- driven market, the Stock Exchange of Hong Kong, and find that the liquidity costs are realized through both spreads and depths. They demonstrate that the bid-ask spread is positively related to the liquidity costs while depth is negatively related to the liquidity costs. Since TSE is also an order-driven market, we expect to see increasing spreads and decreasing depths as prices approach the limits if the magnet hypothesis holds.

III.5. Data and Methodology

III.5.A. Data description

Transaction data of all TSE-listed stocks in year 2000 are obtained from the Taiwan

Economic Journal Data Bank. The data contain time-stamped records of all transactions

120 on the Taiwan Stock Exchange. 541 stocks traded on TSE during year 2000, but only

439 of them traded through the entire year, either due to delisting or IPOs. Trading volume, trading price, and transaction time for each transaction as well as the bid and ask prices are recorded in the transaction data. After August 27, 2000, the bid and ask sizes when the bid price or the ask price is also the limit-hit price are also recorded in the data.

Year 2000 is an ideal period for examining the relation between price limits and overreaction. First of all, the stock market is relatively volatile in 2000, with the index rising from 8756.55 at the beginning of the year to 10202.20 on the 17th of February and then dropping down to 4743.94 at the end of 2000. Given this high volatility, the chance for stocks to hit their limits is also high; so more observations of limit hits can be obtained. Thus, the concern of small sample sizes raised in previous studies is alleviated.

Second, some important events, such as the presidential election and the resignation of the prime minister, occurred in Taiwan, so the lower limit rate was adjusted downward from 7% to 3.5% four times during the year while keeping the upper limit rate unchanged at 7% (see Panel A of Table I for detail). Therefore, we are able to form comparison groups based on those different price-limit rates during different periods to test our overreaction hypotheses.

III.5.B. Methodology

For limit hits, we identify three different cases. The first case is called a closing limit hit, which occurs when a price hits the limit and no other trades occur the remainder of the day. In this case, information may not be fully reflected during the day and the

121 remaining information needs to be reflected on the following day. The second case is called a single limit hit, which occurs when a limit hit is followed by non-limit-hit transactions. In this case, the limit hit is transitory and information may be fully reflected during the day. The last case is called a consecutive limit hit, which occurs when a limit hit is followed by additional trades at the limit price. In this case, information may be fully reflected and the price could be the equilibrium price because market participants are willing to trade at the limit price.

III.5.B.1. Cooling-off hypothesis

For the case of closing limit hits, we compare the percentage of price reversals and price continuations between the limit-hit sample and the comparison sample. The limit- hit sample includes those days when closing prices hit the price limits, while the comparison sample includes all other days with non-zero daily returns. We exclude those days with zero daily returns from the comparison sample in order to make more legitimate comparisons. The event day, t , is the day when the closing price hits the price limit for the limit-hit sample while it is the day when the closing price does not hit the

price limit and when the daily return is not zero for comparison sample. Let rt be the

d n daily return on t , rt be the daytime (open-to-close) return on t , and rt +1 be the overnight (close-to-open) return from t to t + 1, one trading day after t .

 P c  r = n t  (1) t l c  Pt−1 

122  P c  r d = n t  (2) t l o  Pt 

 P o  r n = n t+1  (3) t +1 l c  Pt 

c o where ln is the natural logarithm, Pt is the closing price on t , Pt is the opening price

c o on t , Pt −1 is the closing price on t − 1, the trading day prior to t , and Pt +1 is the opening

price on t + 1, the trading day following t . Stock returns can be positive (+), negative (-) or zero (0). For comparison sample, upward (downward) movements include those days

d when rt is positive (negative). For all upward movements, we classify the set of {[ rt ,

n d n rt +1 ] | [+, +], [0, +]} as price continuations, the set of {[rt , rt +1 ] | [+, -], [0, -]} as price

d n reversals, and the set of {[rt , rt +1 ] | [+, 0], [0, 0]} as no changes in prices around the

d n closing. As to all downward movements, we classify the set of {[rt , rt +1 ] | [-, -], [0, -]}

d n as price continuations, the set of {[rt , rt +1 ] | [-, +], [0, +]} as price reversals, and the set

d n of {[ rt , rt +1 ] | [-, 0], [0, 0]} as no changes in prices around the closing. For the

d n comparison sample, we also add the set of {[rt , rt +1 ] | [-, +], [-, 0], [-, -]} to price

d n reversals for upward movements and the set of {[rt , rt +1 ] | [+, +], [+, 0], [+, -]} to price

reversals for downward movements. If more price reversals are observed from the limit-

hit sample than from the comparison sample, the cooling-off hypothesis is supported. In

this case, the closing limit hit is caused by traders overreacting to news, so a price

reversal would support the cooling-off effect. In contrast, if more price continuations are

observed from the limit-hit sample than from the comparison sample, which is more

likely to happen given the belief that information is not fully reflected on the limit-hit day

123 and the remaining information needs to be reflected on the following day, the cooling-off

hypothesis is not supported. In this case, price limits delay the price discovery process.

However, for the case of single limit hit and consecutive limit hit, by definition, price

reversals are guaranteed. Hence, the cooling-off effect should be tested using a different

methodology. Based on previous discussion, if price limits can provide traders with a

cooling-off period to obtain information, reassess the market price, and avoid

unnecessary overreaction, the degree of return volatility after a limit hit should be less

than that before a limit hit. For return volatility, we calculate the average 3-minute and 5-

minute return volatility 30 minutes before and after limit hits to test the cooling-off

3 5 hypothesis. We obtain the 3-minute returns ( Rt ), 5-minute returns ( Rt ), and the mean

return ( R ) based on the following calculation.

 P  3  t  Rnt = l (4)  Pt−3 

 P  5  t  Rnt = l (5)  Pt −5 

1 R = R (6) n ∑ t

where ln is the natural logarithm, Pt is the transaction price at t , Pt −3 is the transaction

price three minutes prior to t , Pt −5 is the transaction price five minutes prior to t , and

3 5 Rt is either Rt or Rt . In order to gain robust results, we use two different volatility

measures. V1 is the standard deviation of returns and V2 is the mean of absolute returns.

Those measures are calculated as follows:

1 2 V = R− R (7) 1 n − 1∑(t )

124 1 V = R (8) 2 n ∑ t

3 5 where n is the number of observations, Rt is equal to Rt ( Rt ) for 3-minute (5-minute)

return volatility, and R is the mean of Rt .

If we observe lower return volatility after limit hits than before limit hits, the cooling- off hypothesis is supported. Otherwise, it is rejected. Since the bid-ask bounce might affect the volatility measure, we use the bid-ask midpoint to replace the transaction price

for the calculation of returns and volatilities. The bid-ask midpoint ( M t ) is defined as

follows:

()PPbt + at M = (9) t 2

where Pbt is the bid price at time t and Pat is the ask price at time t .

As to the degree of information asymmetry, we use the bid-ask spread as its proxy.

For the purpose of comparison, we calculate the relative spread ( RSt ) and the mean of

relative spreads ( RS ) with the following formula.

()PPat − bt RSt = (10) M t

1 RS = RS (11) n ∑ t

where Pat is the ask price at time t , Pbt is the bid price at time t , M t is the bid-ask

midpoint at time t , and n is the number of observations. If we observe lower RS after limit hits than before limit hits, the cooling-off hypothesis is supported. Otherwise, the hypothesis is rejected.

125 To perform a robustness check, we also construct a control sample called “pseudo” limit hits. As Panel A of Table I shows, there were periods with 7% upward and 7% downward price limits and periods with 7% upward and 3.5% downward price limits.

Pseudo limit hits occur when transaction prices hit the pseudo 3.5% downward price limits during periods with 7% downward price limits. That is, even though the actual downward price limits are 7%, we treat them as 3.5% to identify our pseudo limit hits.

We also split pseudo limit hits to single pseudo limit hits and consecutive pseudo limit hits. The methodology used for single limit hits and consecutive limit hits is applied to pseudo limit hits. However, we do not expect to see a significant difference in return volatility and relative spread between the pre- and post- pseudo limit hits.

In sum, the cooling-off hypothesis leads to the following testable implications.

Proposition 1: For the case of closing limit hits, the percentage of price reversals

(continuations) is expected to be higher (lower) for the limit-hit sample than for the comparison sample.

Proposition 2: For the cases of single limit hits and consecutive limit hits, the return volatility after limit hits is expected to be lower than that before limit hits.

Proposition 3: For the cases of single limit hits and consecutive limit hits, the relative spread after limit hits is expected to be lower than before limit hits.

126 Proposition 4: For the case of pseudo limit hits, the return volatility and the relative spread after limit hits are expected to be insignificantly different from those before limit hits.

We perform the normality test of the volatility measures and the relative spread to determine if nonparametric tests are required. The Wilcoxon signed rank test will be used if it is necessary.

III.5.B.2. Magnet hypothesis

In order to test the magnet hypothesis, it is crucial to find a price that is so close to the price limits that the magnet effect is likely to occur. On TSE, the clearing price in each round of periodic matching cannot exceed two tick sizes from the clearing price in the preceding round. Therefore, a price that is two tick sizes below the upper limit price or a price that is two tick sizes above the lower limit price is the most desirable choice. All qualifying prices are called “magnet prices”. We then calculate the first-order

3 5 autocorrelations of 3-minute returns ( Rt ) and 5-minute returns ( Rt ) for the 30 minutes after the magnet prices are traded. If the magnet hypothesis holds, we expect to see positive autocorrelations. In order to have a benchmark, a control sample needs to be constructed for the purpose of comparison. The control sample includes the same stocks in the sample, matched by time of day, day of week, and duration, during which no

“magnet prices” are observed. Again, the Wilcoxon signed rank test is used to determine the significance level. If we observe higher return autocorrelations from the magnet

127 sample than those from the control sample, the magnet hypothesis is supported.

Otherwise, it is rejected. However, according to Roll (1984), the bid-ask bounce may

result in negative serial correlation, which will reject the hypothesis. Therefore, we also

calculate the first-order autocorrelations of 3-minute and 5-minute returns for the 30

minutes after the magnet prices are traded based on the bid-ask midpoint ( M t ) to mitigate the distortion from the bid-ask bounce. The calculation of returns is the same

except the transaction prices are replaced by M t .

If there is magnet effect, trading volume is expected to increase after the magnet

prices are reached because investors may rush to place market orders. Total trading

volume during the 30 minutes prior to a magnet price is compared to the total trading

volume during the 30 minutes after the magnet price. To make valid comparison, we

divide the trading volume by its corresponding total daily trading volume. The magnet

hypothesis is supported if we observe significantly higher trading volume after the

magnet prices than that before the magnet prices for the magnet sample and no significant

difference between the trading volume before and after for the control sample.

As mentioned earlier, due to increased demand for liquidity, we expect to see

increasing spreads and decreasing depths as prices approach the limits if the magnet

hypothesis holds. Since depth data are not available, we focus on only the bid-ask

spread. The mean of the relative spreads ( RS ) during the 30 minutes prior to the magnet

price is compared to that during the 30 minutes after the magnet price. If the RS prior to

the magnet price is lower than that after the magnet price for the magnet sample and no

significant difference between the RS before and after for the control sample, the magnet hypothesis is supported; otherwise, it is rejected.

128

III.6. Results

III.6.A. Summary Statistics

Panel A of Table II reports the number of observations and ratios for limit hits, magnet hits, and pseudo limit hits during period 1 and period 2 for all 439 stocks traded through out the whole year 2000 on TSE. Period 1 represents all periods with 7% upward and 7% downward price limits while period 2 includes all periods with 7% upward and 3.5% downward price limits. 17,188,194 transactions occurred during period

1 and 3,539,160 transactions during period 2 for all 439 stocks. Upper (Lower) limit hits occur when transaction prices hit the upward (downward) price limits. Upper (lower) magnet hits occur when transaction prices hit the upper (lower) magnet prices. Upper

(lower) magnet prices are defined as the prices that are two ticks below (above) the upward (downward) price limits. Pseudo limit hits occur when transaction prices hit the pseudo 3.5% downward price limits during period 1 when the actual downward price limits are 7%. Ratios, displayed in Panel A of Table II, are defined as the number of observations divided by the number of transactions in each period.

During Period 1, there are more upper limit hits than lower limit hits. The overall limit-hit ratio is about 4%. That is, for every one hundred transactions, 4 hit the price limits. The fact that the sum of the numbers of observations for single, consecutive, and closing limit hits is much less than the total number of limit-hit observations shows that many limit hits are categorized into consecutive limit hits. On average, each consecutive

129 limit hit observation consists of 12 limit-hit transactions. As for magnet hits, there are more lower hits than upper hits and the overall ratio is about 1.55%. The ratio of pseudo limit hits is 1.73%.

Unlike Period 1, there are more lower limit hits than upper limit hits during Period 2.

The main reason is the imposition of 3.5% downward price limits. Further, Period 2 is more volatile than Period 1 given the occurrence of several important events such as the

Presidential Election and the resignation of Prime Minister. The overall limit-hit ratio is about 10%, with 7% being lower limit hits. As for magnet hits, there are more lower hits than upper hits and the overall ratio is 5.37%.

Panel B of Table II reports the summary statistics of daily market returns during both

Periods 1 and 2 based on the TSE Capitalization Weighted Price Index (TAIEX), the most frequently quoted index among the many stock indices published by TSE. Since the standard deviation (S.D.) of daily market returns in Period 2 is higher than that in Period

1, Period 2 is indeed more volatile than Period 1. We also perform the Wilcoxon Rank

Sum test to see if the median in Period 1 is significantly different from that in Period 2, but we find no significant difference between them given a P-value of 0.4104. Since the stock market in Period 2 is more volatile than in Period 1, we would expect to see higher limit-hit ratios in Period 2 than those in Period 1. Besides, given the asymmetric price- limit rates during Period 2, we would expect to see more lower limit hits than upper limit hits. The results reported in Panel A of Table II are consistent with our expectations. All ratios in Period 2 are significantly higher than those in Period 1. In Period 2, there are more lower limit hits than upper limit hits and more lower magnet hits than upper magnet hits.

130

III.6.B. Cooling-off hypothesis

We report the results from the tests of the cooling-off hypothesis based on the three different limit hits: closing limit hits, single limit hits, and consecutive limit hits. To find a benchmark, we form a comparison group for the closing limit hits. For single limit hits and consecutive limit hits, we compare the results with those from the pseudo limit hits.

III.6.B.1. Closing limit hits

Table III reports the proportions of price continuations, reversals, and no changes for both upward movements and downward movements for the closing limit-hit group and the comparison group. Panel A reports the results for periods with 7% upward and 7% downward price limits and Panel B reports those for periods with 7% upward and 3.5% downward price limits. The limit-hit sample includes those days when closing prices hit the price limits, while the comparison sample includes all other days when daily returns are not zero. We exclude those zero-return observations from our comparison sample to gain a more legitimate comparison since they do not exist in our limit-hit sample.

As mentioned earlier, if more price reversals are observed from the limit-hit sample than from the comparison sample, the cooling-off hypothesis is supported. In other words, if the closing limit hit is caused by traders overreacting to news, the cooling-off effect from price limits implies a price reversal. Results from both Panels A and B of

Table III show that the ratio of continuations for the limit-hit sample is significantly

131 higher than that for the comparison sample. The ratio difference, defined as the ratio of the limit-hit sample minus the ratio of the comparison sample, for continuations is positive and significant based on the standard binomial test. On the other hand, the ratio difference of reversals is negative and significant. That is, the ratio of reversals for the limit-hit sample is significantly less than that for the comparison sample. Therefore, the results from the closing limit hits do not support the cooling-off hypothesis. When the closing price hits the price limits, the price on the next trading day tends to move in the same direction toward its equilibrium level. In fact, this is consistent with the delayed price discovery hypothesis, which states that price limits delay the price discovery process due to the limitations imposed on the price movements.

III.6.B.2. Single limit hits

Table IV reports the results from various tests of the cooling-off hypothesis for single limit hits. A single limit hit occurs when a limit hit is followed by non-limit-hit transactions. Upper (Lower) limit hits occur when prices hit the upward (downward) price limits. Pseudo limit hits occur when prices hit the pseudo 3.5% downward price limits during Period 1 when actual price limits are 7%. Pre-Hits refer to the period 30 minutes prior to the single limit hit, while Post-Hits refer to the period 30 minutes after

the single limit hit. R is the mean of the returns, V1 is the standard deviation of the

returns, V2 is the mean of the absolute returns, and N is the sample size. All figures reported are multiplied by 1000. >> (<<) indicates that the left-side figure is higher

(lower) than the right-side figure at the 1% level of significance. > (<) indicates that the

132 left-side figure is higher (lower) than the right-side figure at the 5% level of significance.

The Wilcoxon signed rank test is used to determine the level of significance. The sample

sizes for upper limit hits, lower limit hits, and pseudo limit hits are much smaller than

those reported in Table II because we eliminate many observations in the process of

selecting the qualified sample. Basically, we delete the observations that occur during

the first and the last 30 minutes of each trading day in order to obtain trading data both 30

minutes prior to and after each selected observation. Further, for each single limit hit, if

another limit hit occurs during the period 30 minutes prior to or after the trading took

place, the observation is deleted from our sample. The purpose for these strict selection

criteria is to obtain a clean sample for testing the cooling-off hypothesis.

Panel A reports the average return, R , and average volatilities, V1 and V2 , from both

the 3-minute and the 5-minute return analyses based on transaction prices. The average

3-minute return is the average of ten 3-minute returns during the 30-minute period and

the average 5-minute return is the average of six 5-minute returns during the 30-minute

period. If no trading occurs during the 3-minute and 5-minute interval, the previous

trading price is used to determine the return, which is zero in this case. Basically, the

results from the 3-minute and the 5-minute return analyses are similar. For upper limit

hits, R is significantly higher during Pre-Hits than during Post-Hits, but is significantly

lower during Pre-Hits than during Post-Hits for lower limit hits. In fact, R is positive during Pre-Hits and negative during Post-Hits for upper limit hits. These results are intuitive because prices need to keep going up to hit the upper price limits and then going down to be away from the limits. For lower limit hits, same intuitive argument applies to explain the negative R during Pre-Hits and positive R during Post-Hits. However,

133 without the actual price limits constraint, no significant difference of R between Pre-Hits and Post-Hits is observed for pseudo limit hits.

As to volatility measures, V1 and V2 provide similar results. No significant

differences for V1 and V2 between Pre-Hits and Post-Hits are observed except for the

upper limit hits during periods with 7% upward and 3.5% downward price limits. Both

volatility measures are significantly higher during Pre-Hits than during Post-Hits for

upper limit hits during Period 2. That is, the cooling-off hypothesis is supported only

from upper single limit hits during periods with 7% upward and 3.5% downward price

limits using transactions prices to determine returns and volatilities. For pseudo limit

hits, no significant difference is observed between Pre-Hits and Post-Hits.

Panel B reports the return and volatility from both the 3-minute and 5-minute return

analysis based on the bid-ask midpoint to avoid excess return volatility caused by bid-ask

bounce. Sample sizes are smaller than those in Panel A because transactions with either

zero bid or zero ask prices are excluded from the sample. The results from Panel B are

similar to those in Panel A for both returns and volatilities. However, the support of the

cooling-off hypothesis from upper single limit hits during Period 2 is not as strong as that

in Panel A because the low volatility during Post-Hits is only observed from the 3-minute

return analysis, not the 5-minute return analysis.

Panel C reports the mean of the relative spreads for each type of limit hits. The

relative spread is defined as the bid-ask spread divided by the bid-ask midpoint. Since

the relative spreads are proxies for the degree of information asymmetry, we expect to

see lower relative spreads after limit hits to support the cooling-off hypothesis. Results

from Panel C show that relative spreads increased from Pre-Hits to Post-Hits for upper

134 limit hits during both Periods 1 and 2. For lower and pseudo limit hits, no significant differences between Pre-Hits and Post-Hits are observed. Therefore, results from the spread analysis do not support the cooling-off hypothesis.

III.6.B.3. Consecutive limit hits

Table V reports the results from various tests of the cooling-off hypothesis for consecutive limit hits. A consecutive limit hit occurs when a limit hit is followed by additional trades at the limit price. Pre-Hits refer to the period 30 minutes prior to the first limit hit of a consecutive limit hit, while Post-Hits refer to the period 30 minutes after the last limit hit of the consecutive limit hit. All notations used in Table V are the same as those in Table IV. However, unlike the results from Table IV, Panel A of Table

V shows that return volatilities decrease from Pre-Hits to Post-Hits for both upper and lower limit hits during Period 1 and for lower limit hits during Period 2. No significant difference is found for pseudo limit hits. Therefore, the cooling-off hypothesis is supported by consecutive limit hits except for the upper limit hits during Period 2. Panel

B reports the return and volatility measures from both the 3-minute and 5-minute return analysis based on the bid-ask midpoint to avoid excess return volatility caused by bid-ask bounce. Again, the cooling-off hypothesis is supported because both volatility measures are significantly lower during Post-Hits than during Pre-Hits for all limit hits during all periods except for the pseudo limit hits. Thus, the results from Panel B are stronger than those in Panel A.

135 As to the spread analysis, similar to single limit hits, we find that the average relative spread is higher during Post-Hits than during Pre-Hits for upper limit hits. However, for lower limit hits during Period 1, we observe lower average relative spreads during Post-

Hits than during Pre-Hits. Therefore, the cooling-off hypothesis is supported by lower consecutive limit hits from the spread analysis.

III.6.B.4. Regression analysis

Since the cooling-off hypothesis is supported by consecutive limit hits, but not by single limit hits, we further examine our sample to find out what features associated with consecutive limit hits may contribute to this result. The most apparent feature is the number of transactions at the limit price. Another feature is the time duration between the first limit hit and the last limit hit for each consecutive limit hit. Panel A of Table VI provides descriptive statistics of duration and the number of limit hit transactions. The mean (median) duration is 293 (131) seconds with the maximum and minimum duration being 2504 and 0 seconds, respectively. There are 27 consecutive limit hits whose durations are 0 because two separate limit hit transactions occur at the same time. The mean (median) number of transactions is 5 (3), with the maximum and minimum being

75 and 2, respectively. The correlation between duration and number of transactions is

0.4922.

To determine whether the duration or the number of transactions can explain the magnitude of the volatility change, we run the following regressions:

V(Pre)-V(Post) = α + β1 Duration + ε (12)

136 V(Pre)-V(Post) = α + β1 Transaction + ε (13)

V(Pre)-V(Post) = α + β1 Duration + β2 UP + β3 SEVEN + ε (14)

V(Pre)-V(Post) = α + β1 Transaction + β2 UP + β3 SEVEN + ε (15) where V(Pre) is the pre-hit volatility multiplied by 1,000, while V(Post) is the post-hit volatility multiplied by 1,000. Duration is the total time (in seconds) from the first to the last limit hit, while transaction refers to the number of limit hit transactions for each consecutive limit hit. UP takes the value 1 for upper limit hits and takes the value 0 for lower limit hits. SEVEN takes the value 1 for periods with 7% upward and downward price limits and takes the value 0 for periods with 7% upward and 3.5% downward price limits.

Panel B of Table VI shows that the coefficients of Duration are negative and significant, which indicates that there is negative relationship between Duration and the magnitude of volatility change. That is, the longer the duration of consecutive limit hits, the less the volatility will be reduced. The result may appear to be counter intuitive.

That is, if price limits provide a cooling-off period for market participants to obtain and evaluate information and make rational decisions, we would expect to see a positive relationship between Duration and the magnitude of volatility reduction. However, our regression result is consistent with the time pressure argument widely recognized in psychology literature. Ben Zur and Breznitz (1981) argue that individuals may adapt and accelerate information processing and focus on important information given time pressure. Easterbrook (1959) also points out that with moderate time pressure, decision makers’ performance improves as they focus on relevant cues and exclude the peripheral.

When price limit is hit, market participants are eager to obtain and evaluate information.

137 If the duration is short, given the time pressure, they make decisions based on important and relevant information. On the other hand, if the duration is long, they may obtain some irrelevant information or slow down the speed of information processing due to the decreased time pressure. Thus, the magnitude of volatility reduction can be negatively related to the duration of limit hits.

Besides volatility, we also perform similar regression analysis for the relative spread.

However, unlike the previous analysis, neither duration nor the number of transactions can explain the change in relative spread. The only factor that affects the change in relative spread is the dummy variable UP.

III.6.C. Magnet hypothesis

We report the results from the tests of the magnet hypothesis regarding three different aspects: return autocorrelations, trading volume, and relative spreads. To find a benchmark for comparison, we form a control sample for our magnet sample. The magnet sample includes transactions whose prices have hit the magnet prices. A magnet price is two ticks below the upper limit price or two ticks above the lower limit price.

The control sample includes the same stocks in the magnet sample, matched on time of day, day of week, and duration, during which no “magnet prices” are observed.

III.6.C.1. Return autocorrelations

138 Table VII reports the average return autocorrelations from both the 3-minute return analysis and the 5-minute return analysis during Period 1 and Period 2. Results in Panel

A are based on the transaction prices, while those in Panel B are based on the bid-ask midpoints. Before refers to the period 30 minutes prior to the trading time of each observation and After refers to the period 30 minutes after the trading time of that observation. After-Before is equal to After minus Before. N is the sample size. The P- value of After-Before is based on the Wilcoxon signed rank test.

Panel A of Table VII reports the sample sizes and return autocorrelations of the magnet sample and the control sample during Period 1 and 2 from the 3-minute and 5- minute return analysis. It should be noted that the sample sizes for the 5-minute return analysis are smaller than those for the 3-minute return analysis. The major reason is because some observations have zero 5-minute returns for the 30-minute periods and thus the autocorrelations are not available. In addition, the sample sizes of the magnet sample are much smaller than those reported in Table II due to our strict sample selection criteria. Basically, we delete the observations that occur during the first and the last 30 minutes of each trading day in order to obtain trading data for 30 minutes prior to and after each selected observation. Besides, when a magnet price is hit, it is possible to have another trade(s) at the magnet price. In order to test the magnet hypothesis, we only select the first magnet price to be in the magnet sample. Therefore, the size of the magnet samples decrease significantly from those reported in Table II. For the control sample, we try to find as many matched observations as possible without losing comparability. The control sample needs to match with the magnet sample in terms of stocks, day of the week, time of the day, and duration, during which no “magnet prices”

139 are observed. To obtain a reasonable number of matched observations, we follow four matching rounds with the first being one week prior to the day of each magnet observation. For those unmatched magnet observations after the first round, we perform the second, third, and fourth matching rounds by examining one week after, two weeks prior to, and two weeks after the day of each magnet observation, respectively. To avoid capturing too much noise, we do not go beyond the fourth matching round. Fortunately, the TSE is a very active market, so we are able to find the majority of the magnet- matched observations in the control sample.

We expect to see positive return autocorrelations for the magnet sample and the magnitude should be higher than that from the control sample if the magnet hypothesis holds. However, as shown in Panel A, all autocorrelations are negative. For the magnet sample, the average autocorrelations during After are significantly lower than those during Before except for the 5-minute return analysis during Period 2. Some similar results are found for the control sample, but mainly caused by downward movements.

Therefore, based on the results, the magnet hypothesis is not supported from the return autocorrelations using transaction prices. If these negative autocorrelations are due to the bid-ask bounce suggested by Roll (1984), the use of bid-ask midpoint for determining return autocorrelations should give us more promising results.

Panel B of Table VII reports the sample sizes and return autocorrelations of the magnet sample and the control sample during Period 1 and 2 from the 3-minute and 5- minute return analysis based on bid-ask midpoint. Again, the sample sizes are smaller than those reported in Panel A because observations with zero bid or ask prices are deleted from the sample. The results show that autocorrelations are still negative for both

140 the magnet sample and the control sample. Apparently, the magnet hypothesis is not supported even with those return autocorrelations calculated using the bid-ask midpoints.

III.6.C.2. Trading volume

Table VIII reports the average trading volume of the magnet sample and control sample during Period 1 and Period 2. Trading volume during each 30-minute period, both Before and After, is scaled by its corresponding daily trading volume. N is the sample size. >> (<<) indicates that the left-side figure is higher (lower) than the right- side figure at the 1% level of significance. > (<) indicates that the left-side figure is higher (lower) than the right-side figure at the 5% level of significance. The Wilcoxon signed rank test is used to determine the level of significance.

If the magnet hypothesis holds, we expect to see higher average trading volume during After than during Before because investors rush to trade when the magnet price is reached. Results in Table VIII show that average trading volumes are actually higher during Before than during After for both the magnet sample and the control sample except during the downward movement of the magnet sample. It seems that the magnet hypothesis is rejected. However, since the same results are found for both the magnet sample and the control sample, we cannot attribute this result to the magnet effect.

Instead, we argue that these results might support the magnet hypothesis. The average trading volume reported is actually the proportion of the daily trading volume that occurs during the 30-minute period. If we consider the total proportion of the daily trading volume occurring during the one-hour period (30 minutes before and 30 minutes after

141 each observation) for the magnet sample and the control sample, we find that the proportion (Before+After) is significantly higher for the magnet sample than for the control sample. That is, even though we do not observe higher trading volume during

After than during Before, during the one-hour window, the trading volume of the magnet sample is actually higher than that of the magnet sample for both upward and downward movements and during both Periods 1 and 2. Therefore, we feel safe in concluding that the magnet hypothesis cannot be rejected by this trading volume analysis. Furthermore, the results indicate that the intensive trading activity may occur prior to the time when the magnet price is reached.

III.6.C.3. Relative spreads

Table IX reports the mean of the relative spreads for the magnet sample and the control sample during Period 1 and Period 2. Relative spread is defined as the bid-ask spread divided by the bid-ask midpoint. All relative spreads reported are multiplied by

1000. If the magnet hypothesis holds, we expect to see higher average relative spreads during After than during Before. Results from Table IX show that the average relative spreads are significantly higher during After than during Before for magnet sample during Period 2 and for upward magnet sample during Period 1. Results for the control sample are simply the opposite, with higher average relative spreads during Before than during After during Period 2 and no difference during Period 1. Therefore, the magnet hypothesis is supported by this relative spread analysis. That is, when investors’ demand

142 for liquidity increases, the cost of liquidity is expected to rise and that is reflected in the increasing bid-ask spreads.

III.7. Conclusion

Given the recent development of behavioral finance and its empirical evidence on market overreaction, this paper intends to answer the question “can price limits, one form of circuit breakers, reduce overreaction”? Even though the most popular rationale for imposing price limits is to reduce market overreaction, opponents of price limits argue that it serves no purpose other than to slow down or delay the price discovery process.

Furthermore, rather than generating a stabilizing effect that calms market movements, price limits have a magnet effect that acts to pull prices toward the limit. That is, price limits induce investors’ overreaction when prices are approaching the limits. We use transaction data from the Taiwan Stock Exchange to examine the relation between price limits and overreaction. More specifically, we try to investigate whether price limits reduce or induce overreaction.

143 the magnet hypothesis, which suggests that price limits induce overreaction because investors may submit sub-optimal orders when prices are approaching the limits. We test this hypothesis from three different aspects, namely, return autocorrelations, trading volume, and relative spread. The magnet hypothesis is supported by trading volume and relative spread, but it is not supported by the return autocorrelations. Therefore, our overall result is that price limits induce overreaction when prices are approaching the limits, but they also reduce overreaction when the price hit the limit consecutively.

This paper makes no attempt to investigate whether investors overreact or not. In fact, we assume that investor overreaction exists and examine the relation between price limits and overreaction. Our findings have important regulatory implications. Since price limits can reduce overreaction after consecutive limit hits and induce overreaction when prices are approaching the limits, policy makers need to evaluate the net effect from price limits and set a rule to optimally reduce overreaction. For example, since the cooling-off hypothesis is supported by consecutive limit hits, would a hybrid or combination of price limits and trading halts perform better than a pure price limits or trading halts in reducing overreaction? That might be a good future research topic.

144

References III

Ahn, Hee-Joon, Kee-Hong Bae, and Kalok Chan, 2001, Limit orders, depth, and volatility : evidence from the Stock Exchange of Hong Kong, Journal of Finance 56, 767-788.

Arak, Marcelle, and Richard E. Cook, 1997, Do daily price limits act as magnets ? The case of Treasury bond futures, Journal of Financial Services Research 12, 5-20.

Ben Zur, H., and S.J. Breznitz, 1981, The effects of time pressure on risky choice behavior, Acta Psychologica 47, 89-104.

Biais, Bruno, Pierre Hillion, and Chester Spatt, 1995, An empirical analysis of the limit order book and the order flow in the Paris Bourse, Journal of Finance 50, 1655- 1689.

Brennan, Michael J., 1986, A theory of price limits in futures markets, Journal of Financial Economics 16, 213-233.

Brady, Nicholas F. et al. 1988, Report of the Presidential Task Force on Market Mechanisms. Washington, D.C.: Government Printing Office.

Brockman, Paul, and Dennis Y. Chung, 1999, An analysis of depth behavior in an electronic, order-driven environment, Journal of Banking & Finance 23, 1861-1886.

Chen, Haiwei, 1998, Price limits, overreaction, and price resolution in futures markets, Journal of Futures Markets 18, 243-263.

Choi, Woo Suk, and Sang Bin Lee, 2001, A transitory and asymmetric properties in price limits: Korean evidence with intra-day data, Unpublished working paper. Han Yang University, Seoul.

Corwin, Shane A., and Marc L. Lipson, 2000, Order flow and liquidity around NYSE trading halts, Journal of Finance 55, 1771-1801.

145

Easley, David, and Maureen O’Hara, 1987, Price, trade size, and information in securities markets, Journal of Financial Economics 19, 69-90.

Easterbrook, J.A., 1959, The effect of emotion cue utilization and the organization of behavior, Psychological Review 66, 183-201.

Fama, Eugene F., 1989, Perspectives on October 1987, or, What did we learn from the crash?, In: Kamphuis, R.W., Jr., Kormendi, R.C., Watson, J.W. H. (Eds.), Black Monday and the future of the financial markets, Irwin, Homewood, IL., pp. 71-82.

Gerety, Mason S., and J. Harold Mulherin, 1992, Trading halts and market activity: an analysis of volume at the open and the close, Journal of Finance 47, 1765-1784.

Glosten, Lawrence R., 1994, Is the electronic open limit order book inevitable?, Journal of Finance 49, 1127-1161.

Goldstein, Michael A., and Kenneth A. Kavajecz, 2000, Liquidity provision during circuit breakers and extreme market movements, Unpublished working paper.

Hamao, Yasushi, and Joel Hasbrouck, 1995, Securities trading in the absence of dealers: trades and quotes on the Tokyo Stock Exchange, Review of Financial Studies 8, 849- 878.

Kim, Kenneth A., and Piman Limpaphayom, 2000, Characteristics of stocks that frequently hit price limits: empirical evidence from Taiwan and Thailand, Journal of Financial Markets 3, 315-332.

Kim, Kenneth A., and S. Ghon Rhee, 1997, Price limit performance: evidence from the Tokyo Stock Exchange, Journal of Finance 52, 885-901.

Lee, Charles M.C., Mark J. Ready, and Paul J. Seguin, 1994, Volume, volatility, and New York Stock Exchange trading halts, Journal of Finance 49, 183-214.

Roll, Richard, 1984, A simple implicit measure of the effective bid-ask spread in an efficient market, Journal of Finance 39, 1127-1139.

Stoll, Hans R., 1989, Inferring the components of the bid-ask spread: theory and empirical tests, Journal of Finance 44, 115-134.

Subrahmanyam, Avanidhar, 1994, Circuit breakers and market volatility: a theoretical perspective, Journal of Finance 49, 237-254.

146 Table I Price-Limit Rates and Tick Size

Panel A reports the price-limit rates during different periods in year 2000. Panel B reports the tick sizes for different price ranges on TSE. NT$ is the Taiwanese currency. Information from Both Panels is obtained from the Taiwan Stock Exchange Corporation.

Panel A: Price limits in year 2000 Periods Price-limit Rates

01/01/2000 to 03/19/2000 7% upward and 7% downward 03/20/2000 to 03/26/2000 7% upward and 3.5% downward 03/27/2000 to 10/03/2000 7% upward and 7% downward 10/04/2000 to 10/11/2000 7% upward and 3.5% downward 10/12/2000 to 10/19/2000 7% upward and 7% downward 10/20/2000 to 11/07/2000 7% upward and 3.5% downward 11/08/2000 to 11/20/2000 7% upward and 7% downward 11/21/2000 to 12/31/2000 7% upward and 3.5% downward

Panel B: Tick size Price Range Tick Size

P < NT$ 5.00 NT$ 0.01 NT$ 5.00 ≤ P < NT$ 15.00 NT$ 0.05 NT$ 15.00 ≤ P < NT$ 50.00 NT$ 0.10 NT$ 50.00 ≤ P < NT$ 150.00 NT$ 0.50 NT$ 150.00 ≤ P < NT$ 1,000.00 NT$ 1.00 NT$ 1,000.00 ≤ P NT$ 5.00

147 Table II Summary Statistics

There are 439 stocks traded through out the whole year 2000 on TSE. Panel A reports the number of observations and ratios for both upper and lower limit hits during period 1 and period 2. Upper (Lower) limit hits occur when transaction prices hit the upward (downward) price limits. Period 1 represents all periods with 7% upward and 7% downward price limits while period 2 represents all periods with 7% upward and 3.5% downward price limits in year 2000. There are 17,188,194 transactions occurred during period 1 and 3,539,160 transactions during period 2 for all 439 stocks. A single limit hit occurs when a limit hit is followed by non-limit-hit transactions. Consecutive limit hits occur when a limit hit is followed by further trades at the limit price. Closing limit hit occurs when a price hits the limit and no other trades occur the remainder of the day. Upper (lower) magnet hits occur when transaction prices reach two ticks below (above) the upper (lower) limit prices. Pseudo limit hits occur when transaction prices hit the pseudo 3.5% downward price limits during Period 1 when actual price limits are 7%. Ratios are defined as the number of observations divided by the number of transactions in each period. Ratio2-Ratio1 is the ratio of period 2 minus the ratio of period 1. Z-value of Ratio2-Ratio1 is based on the standard binomial test. Panel B reports the summary statistics of daily market return during both Periods 1 and 2 based on the TSE Capitalization Weighted Price Index (TAIEX).

Panel A:

Period 1 Period 2 Ratio2-Ratio1 Z-value # of observations Ratios # of observations Ratios Upper limit hits 399890 0.0233 104124 0.0294 0.0062 68.46 Single 12455 5528 Consecutive 26826 9156 Closing 5343 2114

Lower limit hits 305466 0.0178 251757 0.0711 0.0534 565.21 Single 15246 14973 Consecutive 28978 25285 Closing 4471 4359

Total limit hits 705356 0.0410 355881 0.1006 0.0595 462.62

Upper magnet hits 119522 0.0070 37549 0.0106 0.0036 72.22 Lower magnet hits 146749 0.0085 152490 0.0431 0.0346 496.19 Total magnet hits 266271 0.0155 190039 0.0537 0.0382 446.05

Pseudo limit hits 298162 0.0173

Panel B: Daily market return Period N Mean S.D. Min Max Median 1 213 -0.00375 0.01990 -0.06774 0.04483 -0.00292 2 57 0.00325 0.02518 -0.02757 0.06172 -0.00125

148 Table III Cooling-off hypothesis: Closing Limit Hits

There are 439 stocks traded through the whole year 2000 on TSE. This table reports the proportions of price continuations, reversals, and no changes for both upward movements and downward movements for the limit-hit group and the comparison group. Panel A reports the results for periods with 7% upward and 7% downward price limits and Panel B reports those for periods with 7% upward and 3.5% downward price limits. Limit-hit sample includes those days when price hits the limit and no other trades occur the remainder of the day, while the comparison sample includes all other days when daily returns are not zero. = ( c / c ) d = ( c / o ) n = ( o / c ) c Let rt ln Pt Pt −1 , rt ln Pt Pt and rt +1 ln Pt +1 Pt , where Pt is the closing price on o c day t , Pt is the opening price on day t , Pt −1 is the closing price on t − 1 , the trading day prior to t , o and Pt +1 is the opening price on t + 1 , the trading day following t . Stock returns can be positive (+), negative (-) or zero (0). For comparison sample, upward (downward) movements include those days when d n rt is positive (negative). For upward movements, we classify the set of {[ rt , rt +1 ] | [+, +], [0, +]} as price d n d n continuations, the set of {[ rt , rt +1 ] | [+, -], [0, -]} as price reversals, and the set of {[ rt , rt +1 ] | [+, 0], [0, d 0]} as no changes in prices around the closing. As to downward movements, we classify the set of {[ rt , n d n rt +1 ] | [-, -], [0, -]} as price continuations, the set of {[ rt , rt +1 ] | [-, +], [0, +]} as price reversals, and the d n set of {[ rt , rt +1 ] | [-, 0], [0, 0]} as no changes in prices around the closing. Besides, for comparison d n sample, we also add the set of {[ rt , rt +1 ] | [-, +], [-, 0], [-, -]} to price reversals for upward movements and d n the set of {[ rt , rt +1 ] | [+, +], [+, 0], [+, -]} to price reversals for downward movements. The proportions may not add to 1 due to rounding error. N is the sample size. Ratio difference is the ratio of the limit-hit sample minus the ratio of the comparison sample. Z-value of the ratio difference is based on the standard binomial test.

Price Behavior Limit-hit Sample Comparison Sample Ratio Difference Z-value Panel A: Periods with 7% upward and 7% downward price limits Upward Movements N = 5343 N = 30594 Continuation 0.770 0.410 0.359 48.60 Reversal 0.121 0.382 -0.261 -37.11 No change 0.110 0.208 -0.098 -16.76 Downward Movements N = 4471 N = 42992 Continuation 0.740 0.316 0.425 56.48 Reversal 0.143 0.439 -0.295 -38.21 No change 0.116 0.246 -0.129 -19.47 Panel B: Periods with 7% upward and 3.5% downward price limits Upward Movements N = 2114 N = 7589 Continuation 0.659 0.309 0.350 29.21 Reversal 0.161 0.462 -0.301 -25.01 No change 0.180 0.229 -0.049 -4.80 Downward Movements N = 4359 N = 8428 Continuation 0.862 0.494 0.369 40.68 Reversal 0.072 0.383 -0.311 -37.28 No change 0.066 0.123 -0.057 -10.07

149 Table IV Cooling-off hypothesis: Single Limit Hits

A single limit hit occurs when a limit hit is followed by non-limit-hit transactions. Upper (Lower) limit hits occur when prices hit the upward (downward) price limits. Pseudo limit hits occur when prices hit the pseudo 3.5% downward price limits during Period 1 when actual price limits are 7%. Pre-Hits (Post-Hits) refers to the period 30 minutes prior to (after) the single limit hit. R is the mean of returns, V1 is the standard deviation of returns, V2 is the mean of absolute returns, and N is the sample size. Panel A (B) reports the return and volatility from both the 3-minute and the 5-minute return analyses based on transaction prices (bid-ask midpoints). A bid-ask midpoint is the average of the bid price and the ask price. Panel C reports the mean of relative spreads for each type of limit hits. The relative spread is defined as the bid-ask spread divided by the bid-ask midpoint. All figures reported are multiplied by 1000. >> (<<) indicates that the left-side figure is higher (lower) than the right-side figure at the 1% level of significance. > (<) indicates that the left-side figure is higher (lower) than the right-side figure at the 5% level of significance. Wilcoxon signed rank test is used to determine the level of significance.

Panel A: Transaction Prices 3-minute return analysis 5-minute return analysis Return and Volatility Pre-Hits Post-Hits Pre-Hits Post-Hits Period 1: Periods with 7% upward and 7% downward price limits Upper Limit Hits (N=118) R 0.3983 >> -0.3259 0.6639 >> -0.5431

V1 1.6317 1.5667 1.9643 1.8659

V2 1.0340 1.0364 1.3808 1.3847 Lower Limit Hits (N=143) R -0.7157 << 0.5414 -1.1928 << 0.9013

V1 3.3569 3.4969 4.2144 3.9736

V2 2.2017 2.0722 2.9353 2.5949 Pseudo Limit Hits (N=202) R 0.0282 -0.1223 0.0466 -0.2039

V1 5.2701 4.7803 6.5890 5.9899

V2 2.8853 2.5975 4.1788 3.7041 Period 2: Periods with 7% upward and 3.5% downward price limits Upper Limit Hits (N=62) R 0.8787 >> -0.2769 1.4646 >> -0.4614

V1 3.6914 >> 2.3930 4.5217 > 3.1066

V2 2.2616 >> 1.4642 3.1053 >> 2.1293 Lower Limit Hits (N=238) R -0.2903 << 0.1594 -0.4827 << 0.2655

V1 2.0171 1.8801 2.4277 2.2391

V2 1.1288 1.0827 1.5010 1.4282

150

Panel B: Bid-ask Midpoint 3-minute return analysis 5-minute return analysis Return and Volatility Pre-Hits Post-Hits Pre-Hits Post-Hits Period 1: Periods with 7% upward and 7% downward price limits Upper Limit Hits (N=31) R 0.4880 >> -0.7623 0.8133 >> -1.2705

V1 2.3426 < 3.0134 3.1317 3.5451

V2 1.3533 1.6317 2.0771 2.2642 Lower Limit Hits (N=67) R -1.1690 << 0.8662 -1.9484 << 1.4384

V1 4.5048 3.7463 5.7715 4.7064

V2 2.7800 2.2478 4.1423 3.1706 Pseudo Limit Hits (N=182) R -0.0410 0.0266 -0.0684 0.0459

V1 3.2173 3.2034 3.9324 4.0927

V2 1.7044 1.7688 2.3977 2.5660 Period 2: Periods with 7% upward and 3.5% downward price limits Upper Limit Hits (N=26) R 1.0226 >> -0.7115 1.7044 >> -1.1858

V1 4.8535 > 3.4506 5.7173 4.6499

V2 2.8110 > 2.0825 4.1422 3.1856 Lower Limit Hits (N=88) R -0.4638 << 0.4708 -0.7729 << 0.7842

V1 2.3820 2.7875 2.8717 3.3517

V2 1.1428 1.4623 1.6151 1.9184

Panel C: Spread Analysis Relative spread Pre-Hits Post-Hits Period 1: Periods with 7% upward and 7% downward price limits Upper Limit Hits (N=31) 6.1196 << 6.9006 Lower Limit Hits (N=67) 6.6736 6.2861 Pseudo Limit Hits (N=182) 8.1815 8.1926 Period 2: Periods with 7% upward and 3.5% downward price limits Upper Limit Hits (N=26) 5.6894 << 7.2423 Lower Limit Hits (N=88) 7.9950 8.0124

151 Table V Cooling-off hypothesis: Consecutive Limit Hits

A consecutive limit hit occurs when a limit hit is followed by further trades at the limit price. Upper (Lower) limit hits occur when prices hit the upward (downward) price limits. Pseudo limit hits occur when prices hit the pseudo 3.5% downward price limits during Period 1 when actual price limits are 7%. Pre- Hits (Post-Hits) refers to the period 30 minutes prior to (after) the consecutive limit hit. R is the mean of returns, V1 is the standard deviation of returns, V2 is the mean of absolute returns, and N is the sample size. Panel A (B) reports the return and volatility from both the 3-minute and the 5-minute return analyses based on transaction prices (bid-ask midpoints). A bid-ask midpoint is the average of the bid price and the ask price. Panel C reports the mean of relative spreads for each type of limit hits. The relative spread is defined as the bid-ask spread divided by the bid-ask midpoint. All figures reported are multiplied by 1000. >> (<<) indicates that the left-side figure is higher (lower) than the right-side figure at the 1% level of significance. > (<) indicates that the left-side figure is higher (lower) than the right-side figure at the 5% level of significance. Wilcoxon signed rank test is used to determine the level of significance.

V1 7.4748 >> 5.8039 9.1085 >> 6.7072

V2 4.8599 >> 3.4322 6.6792 >> 4.4696 Lower Limit Hits (N=73) R -1.6768 << 0.8633 -2.7693 << 1.4036

V1 6.0690 5.5990 7.8158 >> 6.7025

V2 4.0367 >> 3.4978 5.6827 >> 4.4097 Pseudo Limit Hits (N=188) R -0.1583 -0.0621 -0.2643 -0.1014

V1 5.3638 5.4315 6.4599 6.5119

V2 3.1867 3.3306 4.2856 4.3149 Period 2: Periods with 7% upward and 3.5% downward price limits Upper Limit Hits (N=43) R 2.2480 >> -0.8816 3.6450 >> -1.4420

V1 6.3728 5.8240 7.5904 6.7219

V2 4.5270 3.8124 6.1482 >> 4.7127 Lower Limit Hits (N=79) R -0.5787 << 0.4430 -0.9400 << 0.7275

V1 3.2999 > 2.5113 4.1017 > 3.1997

V2 1.8538 >> 1.3789 2.6104 >> 1.9431

152

V1 6.0377 >> 3.8467 7.7632 >> 4.7720

V2 3.7023 >> 2.1944 5.4156 >> 3.1013 Lower Limit Hits (N=59) R -1.4521 << 0.6110 -2.3987 << 0.9983

V1 4.3616 >> 3.2918 5.7606 >> 4.3748

V2 2.6990 >> 1.8570 4.0645 >> 2.6812 Pseudo Limit Hits (N=176) R -0.2420 -0.0722 -0.3996 -0.1184

V1 3.6263 4.0527 4.4683 5.1234

V2 2.0950 2.3681 3.0097 3.3631 Period 2: Periods with 7% upward and 3.5% downward price limits Upper Limit Hits (N=40) R 1.8471 >> -0.6215 3.0236 >> -1.0215

V1 4.6791 >> 3.6199 6.3065 >> 4.5991

V2 3.0389 >> 2.3388 4.7691 >> 3.2652 Lower Limit Hits (N=51) R -0.5047 << 0.1628 -0.8399 << 0.2639

V1 2.9013 >> 1.8206 3.6082 >> 2.1514

V2 1.4998 >> 0.9471 2.1131 >> 1.2626

Panel C: Spread Analysis Relative spread Pre-Hits Post-Hits Period 1: Periods with 7% upward and 7% downward price limits Upper Limit Hits (N=46) 4.5300 << 6.8566 Lower Limit Hits (N=59) 6.7229 >> 6.0832 Pseudo Limit Hits (N=176) 7.2175 7.3315 Period 2: Periods with 7% upward and 3.5% downward price limits Upper Limit Hits (N=40) 5.3758 < 5.8027 Lower Limit Hits (N=51) 7.5973 7.1410

153 Table VI Regression Analysis

This table reports the results of regression equations in which the dependent variable is the difference between Pre-Hit volatility and Post-Hit volatility for consecutive limit hits. A consecutive limit hit occurs when a limit hit is followed by further trades at the limit price. Pre-Hit (Post-Hit) refers to the period 30 minutes prior to (after) each consecutive limit hit. Panel A provides descriptive statistics of duration and transactions. Duration is the total time (in seconds) from the first to the last limit hit, while transaction refers to the number of limit hit transactions for each consecutive limit hit. Panel B reports the regression results of the regression: V(Pre)-V(Post) = α + β1 Duration + β2 Up + β3 Seven + ε as well as its simplified specification. V(Pre) is the pre-hit volatility multiplied by 1,000, while V(Post) is the post-hit volatility multiplied by 1,000. Up takes the value 1 for upper limit hits and takes the value 0 for lower limit hits. Upper (Lower) limit hits occur when prices hit the upward (downward) price limits. Seven takes the value 1 for periods with 7% upward and downward price limits and takes the value 0 for periods with 7% upward and 3.5% downward price limits. The numbers in parentheses are p-values.

Panel A: Descriptive Statistics Duration (in seconds) Transaction Mean 293 5 Median 131 3 Maximum 2504 75 Minimum 0 2

Panel B: Regressions 3-minute return 5-minute return Independent Variables [1] [2] [1] [2]

Constant ***0.80593 0.44802 ***1.55304 *0.92042 (0.0002) (0.1817) (0.0001) (0.0716) Duration **-0.00057 **-0.00053 ***-0.00109 ***-0.00103 (0.0200) (0.0320) (0.0037) (0.0067) Up 0.51852 0.7786 (0.1965) (0.2021) Seven 0.26444 0.57907 (0.4981) (0.3294)

R-square 0.0226 0.0319 0.0349 0.0462 Number of Observations 252 252 252 252

***Significance at the 1 percent level; **significance at the 5 percent level; *significance at the 10 percent level.

154 Table VII Magnet hypothesis: Return Autocorrelations

This table reports the average return autocorrelations from both the 3-minute return analysis and the 5-minute return analysis during Period 1 and Period 2. Results in Panel A are based on the transaction prices, while those in Panel B are based on the bid-ask midpoints. A bid-ask midpoint is the average of the bid price and the ask price. Magnet sample includes transactions whose prices hit the magnet prices. A magnet price is two ticks below the upper limit price or two ticks above the lower limit price. Control sample includes same stocks in the magnet sample, matched on time of day, day of week, and duration, during which no “magnet prices” are observed. Before (After) refers to the period 30 minutes prior to (after) the trading time of each observation. After-Before is equal to After minus Before. N is the sample size. P-value of After-Before is based on the Wilcoxon signed rank test.

Panel A: Transaction Prices 3-minute return analysis 5-minute return analysis Autocorrelations N Before After After-Before P-value N Before After After-Before P-value Period 1: Periods with 7% upward and 7% downward price limits Magnet Sample Upward movement 2855 -0.1153 -0.1954 -0.0801 0.0000 2846 -0.1458 -0.1930 -0.0472 0.0000 Downward movement 3574 -0.1652 -0.2019 -0.0368 0.0000 3534 -0.1899 -0.2199 -0.0301 0.0000

Control Sample Upward movement 2481 -0.2153 -0.2224 -0.0071 0.3160 2426 -0.2331 -0.2340 -0.0010 0.9890 Downward movement 2928 -0.2049 -0.2320 -0.0271 0.0000 2874 -0.2330 -0.2379 -0.0049 0.3370 Period 2: Periods with 7% upward and 3.5% downward price limits Magnet Sample Upward movement 803 -0.1284 -0.1836 -0.0552 0.0000 798 -0.1921 -0.2061 -0.0140 0.4650 Downward movement 1693 -0.1619 -0.1964 -0.0345 0.0000 1660 -0.2137 -0.2130 0.0006 0.9030

Control Sample Upward movement 597 -0.1851 -0.1762 0.0089 0.7000 592 -0.2180 -0.2101 0.0079 0.6400 Downward movement 1436 -0.1849 -0.2052 -0.0202 0.0440 1400 -0.2163 -0.2335 -0.0172 0.0720

155

Panel B: Bid-ask Midpoint 3-minute return analysis 5-minute return analysis Autocorrelations N Before After After-Before P-value N Before After After-Before P-value Period 1: Periods with 7% upward and 7% downward price limits Magnet Sample Upward movement 2679 -0.0490 -0.1124 -0.0634 0.0000 2661 -0.1046 -0.1550 -0.0505 0.0000 Downward movement 3161 -0.1064 -0.1439 -0.0375 0.0000 3131 -0.1455 -0.1823 -0.0368 0.0000

Control Sample Upward movement 2202 -0.1280 -0.1291 -0.0012 0.7580 2160 -0.1746 -0.1734 0.0012 0.7810 Downward movement 2554 -0.1256 -0.1418 -0.0163 0.0290 2507 -0.1803 -0.1955 -0.0152 0.0230 Period 2: Periods with 7% upward and 3.5% downward price limits Magnet Sample Upward movement 712 -0.0702 -0.1133 -0.0431 0.0020 706 -0.1421 -0.1650 -0.0229 0.0980 Downward movement 1410 -0.1179 -0.1374 -0.0195 0.0310 1388 -0.1746 -0.1854 -0.0108 0.3130

Control Sample Upward movement 533 -0.1003 -0.0990 0.0014 0.9760 529 -0.1626 -0.1597 0.0029 0.9510 Downward movement 1188 -0.1227 -0.1208 0.0018 0.9970 1167 -0.1911 -0.1816 0.0094 0.5270

156 Table VIII Magnet hypothesis: Trading Volume

This table reports the average trading volume of magnet sample and control sample during Period 1 and Period 2. Magnet sample includes transactions whose prices hit the magnet prices. A magnet price is two ticks below the upper limit price or two ticks above the lower limit price. Control sample includes same stocks in the magnet sample, matched on time of day, day of week, and duration, during which no “magnet prices” are observed. Before (After) refers to the period 30 minutes prior to (after) the trading time of each observation. Before+After is equal to Before plus After. Trading volume during each 30-minute period is scaled by its corresponding daily trading volume. N is the sample size. >> (<<) indicates that the left-side figure is higher (lower) than the right-side figure at the 1% level of significance. > (<) indicates that the left-side figure is higher (lower) than the right-side figure at the 5% level of significance. The superscript a (b) indicates that the figure is higher than that in the control sample with same type of movement at the 1% (5%) level of significance. Wilcoxon signed rank test is used to determine the level of significance.

Trading Volume N Before After Before+After Period 1: Periods with 7% upward and 7% downward price limits Magnet Sample Upward movement 2995 0.2232 >> 0.2024 0.4256a Downward movement 3869 0.1727 0.1698 0.3425a

Control Sample Upward movement 2732 0.1452 >> 0.1273 0.2725 Downward movement 3439 0.1391 >> 0.1230 0.2621 Period 2: Periods with 7% upward and 3.5% downward price limits Magnet Sample Upward movement 854 0.1948 >> 0.1720 0.3668a Downward movement 2205 0.1578 >> 0.1373 0.2951b

Control Sample Upward movement 763 0.1361 > 0.1282 0.2643 Downward movement 1776 0.1472 >> 0.1210 0.2682

157 Table IX Magnet hypothesis: Relative spreads

This table reports the mean of relative spreads of magnet sample and control sample during Period 1 and Period 2. Magnet sample includes transactions whose prices hit the magnet prices. A magnet price is two tick sizes below the upper limit price or two tick sizes above the lower limit price. Control sample includes same stocks in the magnet sample, matched on time of day, day of week, and duration, during which no “magnet prices” are observed. Before (After) refers to the period 30 minutes prior to (after) the trading time of each observation. Relative spread is defined as the bid-ask spread divided by the bid-ask midpoint. A bid-ask midpoint is the average of the bid price and the ask price. N is the sample size. All relative spreads reported are multiplied by 1000. >> (<<) indicates that the left-side figure is higher (lower) than the right-side figure at the 1% level of significance. > (<) indicates that the left-side figure is higher (lower) than the right-side figure at the 5% level of significance. Wilcoxon signed rank test is used to determine the level of significance.

Relative spreads N Before After Period 1: Periods with 7% upward and 7% downward price limits Magnet Sample Upward movement 2866 5.335 << 5.525 Downward movement 3582 7.304 7.308

Control Sample Upward movement 2549 5.691 5.688 Downward movement 3100 6.475 6.431 Period 2: Periods with 7% upward and 3.5% downward price limits Magnet Sample Upward movement 787 5.670 << 5.976 Downward movement 1929 7.115 << 7.210

Control Sample Upward movement 624 5.962 >> 5.790 Downward movement 1501 6.625 >> 6.446

158