The Stock Price Overreaction Effect: Evidence on the Dutch Stock Market

(1)

The Stock Price Overreaction Effect: Evidence on

the Dutch Stock Market

University of Groningen

Faculty of Economics and Business

Msc Finance

Jan Wicher G. Noorda

Groningen, september 2007

(2)

(3)

Throughout the last year of my study at the University of Groningen, I have been studying the Master of Science in Finance. This thesis is the final part of my master. The subject of my master thesis is ‘The Stockprice Overreaction Effect: Evidence on the Dutch Stock Market’.

With this preface, I would like to take the opportunity to thank a number of people for their help and support. In the first place my mentor, Drs. Marc Kramer, who was always willing to advice me when necessary and at all times on short notice. Eventhough I was very busy with my application for the Rabobank, and preparings for my first project in Hong Kong.

(4)

Abstract

(5)

1 Introduction ... 7

2 Theoretical Framework ... 10

2.1 Evidence on Long Term Overreaction Effect ... 10

2.2 Evidence on Short Term Overreaction Effect ... 12

2.3 International Evidence on Stock Price Overreaction ... 13

2.3.1 Overreaction Effects on Large Financial Markets Outside the U.S. ... 13

2.3.2 Overreaction Effects on European Markets... 14

2.3.3 Overreaction Effects in Emerging Markets... 14

2.3.4 Overreaction Effects Outside Stock Markets ... 15

2.4 Summary of the Discussed Literature and Hypothesis Formulation... 15

3 A Brief Review of the Dutch Stock Exchange... 17

3.1 AEX... 17

3.2 AMX... 18

3.3 AScX ... 18

4 Data and Methodology... 18

4.1 Data ... 19

4.1.1 Daily Data ... 19

4.1.2 Selection of the Gainers and Losers... 19

4.1.3 Stock Data ... 20

4.1.4 Firm Specific News... 21

4.1.5 Descriptive Statistics ... 22

4.2 Methodology ... 23

4.2.1 Event Study Methodology... 23

4.2.2 Day 0 Definition... 23

4.2.3 Event and Estimation Period... 24

4.2.4 Computation of Stock Returns... 24

4.2.5 Normal Performance Model... 25

4.2.6 Computation of Abnormal Returns ... 26

4.2.7 Aggregation of Abnormal Returns ... 26

4.2.8 Computation of the Cumulative Abnormal Returns ... 27

4.2.9 Regression of the News Categories... 27

(6)

5 Empirical Results ... 29

5.1 Did the Winners Overreact?... 29

5.1.1 AMS Total Share Sample... 29

5.1.2 AEX, AMX, AScX Share Sample... 31

5.2 Did the Losers Overreact?... 32

5.2.1 AMS Total Share Sample... 32

5.2.2 AEX, AMX, AScX Share Sample... 34

5.3 Differences Between the News Categories ... 36

5.3.1 The News Effects for Winners ... 36

5.3.2 The News Effects for Losers ... 36

5.4 Summary and Discussion of the Results ... 37

6 Conclusions ... 38

References ... 40

Appendices ... 44

Appendix I In- and Excluded Companies

Appendix II Characterizations of the News Events Appendix III Descriptive Statistics

Appendix IV Average Abnormal Returns at the Four per cent Level

Appendix V Cumulative Average Abnormal Returns at the Four per cent Level

(7)

1 Introduction

Maybe one of the most illustrious examples of investor overreaction can be found in the Dutch market. Where in the 1630’s a mania in tulip bulbs was at hand, and the price of a single bulb had risen to, an equivalent current price of, €60,000. The tulip bulbs were traded on local market exchanges that did not differ that much from current stock exchanges. By 1634 the fever hit the Dutch middle-class, everybody was trading these tulip bulbs expecting to make a big profit by buying low and selling high. In less than a month the price of a single tulip bulb went up twenty-fold. To keep trading available for every investor, option plans were formed. Now, even the smallest individual investor was able to take ‘advantage’ of this craze. Then the Dutch government started to develop regulation to help control the tulip bulb mania and a few informed investors started to cash their contract. After their swap for cash, the market for tulip bulbs started to plummet. It lasted six weeks before the same tulip bulb that had costed €60.000 at the height of the mania reached a devastating worth of less than one Euro. This market mania is a classic example of investors acting irrationally, by not valuing the goods by their underlying value. Greed is one of the two well know forces that drives the market, and made the prices of tulip bulbs go up. Fear is the other and can be held responsible for the declining prices of the tulip bulbs.1

In an efficient market, stock prices always reflect the best information about fundamental values. Prices change only because of good, sensible information meshed very well with theoretical trends in time (Schiller, 2003, p.83). However, from the 1970s onwards, when the efficient market hypothesis (EMH) was first stated by Fama (1970), this hypothesis has been questioned. For stocks, this EMH implicates that prices should reflect a rational forecast of the present value of future dividends payments. Therefore, when the hypothesis holds, share prices reflect all information and no one can earn excess returns, the market is said to be strong form efficient. When share prices adjust within an arbitrarily small, but finite amount of time and in an unbiased fashion to publicly available new information, the market is said to be semi-strong form efficient. The weak-form efficient market implies that no excess returns can be earned by using investment strategies based on historical share prices or other historical financial data.

In research literature over the past few decades, the stock market overreaction effect has been given wide attention. The stock market overreaction hypothesis states that when a stock price experiences a sharp increase or decrease, it usually reverses itself. Recent research has shown that stock markets overreact and suggest that, when individuals revise their beliefs, they tend to overweigh recent information and underweigh prior information (Otchere and Chan, 2003:157). This implies that investors do not trade rational, which supports the findings of March (1994). March concludes that individuals are ‘bounded (or limited) rational’ meaning that individuals are intendedly rational.

1_{For more information on the tulip bulb mania see Robert Sobel, The Big Board: A History of the New York}

(8)

University of Groningen J.W.G. Noorda 8

“Although decision makers try to be rational, they are constrained by limited cognitive capabilities and incomplete information, and thus their actions may be less than completely rational in spite of their best intentions and efforts”(March, 1994:9).

If the stock market overreaction hypothesis holds, and a contrarian investment strategy can be made based on past returns data, it would make the stock market inefficient. De Bondt and Thaler (1985) were the first to investigate the overreaction hypothesis in finance. They tracked various firms’ stock performance for a given interval, long term, and formed portfolios of winner and loser stocks based on their performance prior to the event date. When the portfolios reversed themselves after the event date, so winners (losers) perform poorly (well) in a statistical sense, there is reason to believe stock markets overreact. In contrast, Howe (1986) and Brown and Harlow (1988) found that investor overreaction, especially to bad news, tends to be a short term event. The authors argue that if this hypothesis holds, a contrarian strategy (buying previous ‘losers’ and selling previous ‘winners’) can be formed. If an investor would pursue such a strategy and buy (short-sell) previous losers (winners), he/she would be able to earn excess profits. This leaves the stock market inefficient, even in the weak-form.

Another point of interest are the developments and new technologies in market environments. More individuals have begun to actively participate in the stock market, in China millions of new investors have found their way to the stock market which rose by 160 per cent in one and a halve year. The internet also paved the way for many new investors, by providing more accurate and up to date information to this group of investors. It also made it possible to invest small amounts. Even banks and large investment companies formed diversified portfolios in which individual investors could invest at a low price. Another recent development is that of the investing pension funds, who have found ways to channel their money in the market.

Although the stock market overreaction effect has been well documented for the US and other well developed big markets, little research has been done for the Dutch market. In view of the above, this article will try to contribute to the existing literature on stock market overreaction by evaluating profitable opportunities, when taking advantage of the reversal, on the Dutch market. To this end, it will build on the works of Doeswijk (1997), who studied contrarian investments in the Dutch market, using Dutch data. However, contrary to his research, this paper will further examine a categorization in indexes. The categorization in indexes, the AEX, AMX (Amsterdam Midkap) and AScX (Amsterdam Small cap), will make it possible to perform comparative analyses. Not only between winners and losers, but also between the indices, that have different average market capitalization. Furthermore, this research will look at short term overreaction, in contrast with the research of Doeswijk (1997) who investigated a long term reversal for the Dutch market.

(9)

is expected that merger and acquisition news will affect gainers most and revenues and earnings news second. This can be explained by the idea that mergers and acquisitions news has less information leakage and contains the most ‘new’, unanticipated information. For losers the news of revenues and earnings is most associated with overreaction.

The presence of a mean reverting effect is also examined. People overreact more on negative news, therefore the loser firms show a more quick reversal than do stocks of winner firms. Nam et al. (2001) found this evidence persistent for the U.S. stock market, and in particular for the NYSE, AMEX and NASDAQ. They reported that negative returns reverse to positive returns quicker than positive returns to negative ones between 1926:01 and 1997:12.

In this study, a closer look will be taken at the Dutch stock market, in more detail a two-calendar-year data set from March 2005 to March 2007 will be investigated for stock market overreaction. This is to empirically explore the stock price overreaction effect on stocks that have experienced a percentage price increase or decrease of at least 4 per cent. A sensitivity analysis is performed to check if the main conclusions hold when the trigger is set to 5 per cent or 6 per cent. The main research question of this paper will therefore be:

Is the stock market overreaction effect persistent for stocks listed on the Dutch stock market? Of special interest is if the overreaction effect, if present, is economically significant. This would allow investors to earn abnormal returns by forming a contrarian investment strategy.

(10)

2 Theoretical

Framework

Motivated by findings in cognitive psychology, the first authors to examine the stock market overreaction hypothesis in finance were De Bondt and Thaler (1985). They build on the works of Kahneman and Tversky (1973), who argue that individuals tend to overweigh recent information and underweigh prior (base-rate) data. In their 1974 article, Tversky and Kahneman show that heuristics, used by individuals to make decisions under uncertainty, may cause ‘systematic errors’ (p.1124). Individuals do not consider prior probabilities when making their evaluations, but instead they base their probabilities of occurrence on how similar, or ‘representative’, an event is to their alleged notions. Recent research of Barber and Odean (2006) showed that many investors only consider purchasing stocks that have first caught their attention. Thus, preferences determine choices after attention has determined the choice set. This is in violation of Bayes’ rule, and can lead individual investors to drive perceived values beyond their correct valuations. This results in the empirical phenomenon of overreaction, which is discussed in this article.

There have been several studies that empirically tested for stock market overreaction and the related following reversal. Most of the literature has found evidence of overreaction in stock markets, but some have leveled criticisms against the work of De Bondt and Thalers (1985). The discussion of the overreaction literature will be presented based on the longitude of the effect, from long- to short term overreaction. First overreaction literature for the U.S is presented, where after overreaction on large financial markets outside U.S., Europe, Emerging markets and evidence outside stock markets is discussed. This section will conclude with an overview in the form of a table (Table I).

2.1 Evidence on Long Term Overreaction Effect

(11)

most of the excess returns are realized in January; at the start of the year the loser stocks portfolio earned excessively big positive excess returns. The authors argued that investors using a contrarian investment strategy of buying the losers and short selling the winners could earn a highly economically significant abnormal profit.

Chan (1988) constructs similar portfolios as De Bondt and Thaler (1985) and uses data from the CRSP files from December 1926 to December 1985 to construct winner and loser portfolios. He finds that the risks of losers and winners are not constant over time, which followed from pronounced changes in leverage, which is known as the ‘leverage hypothesis’. Since the equity beta of a firm is a function of both asset risk and leverage, a series of negative abnormal returns will increase the equity beta of a firm. Thereby increasing the expected return on the stock (assuming that the asset beta is positive and does not decrease substantially, and that the firm does not change its debt to fully offset the decline in the value of its equity).2 When these risk changes are controlled for, he finds only very small abnormal returns. These returns are economically insignificant and therefore there is no (strong) support of the stock market overreaction hypothesis.

Ball and Kothari (1989) follow this same logic and provide evidence that suggests that the negative serial correlation in relative returns is due largely to changing relative risks and thus changing expected returns. This evidence is therefore not consistent with the stock market overreaction effect, and although the serial correlation in portfolios’ abnormal returns is statistically significant, the magnitude of the abnormal returns is small and, as they believe, economically insignificant. They add to it that it varies over time. An interesting observation is the difference in the extreme portfolios of winners and losers. The extreme loser- minus extreme winner portfolio has a beta of 0.76, which can account for considerable differences in realized returns.

Zarowin (1990) reported that the reason why losers outperform winners cannot be contributed to risk differences alone, therefore he uses the size effect. The work on the size effect was first performed by Banz (1981). He shows that the size of a firm and the return on its common stock are inversely related. Zarowin states that loser firms are smaller than winner firms and claims the return reversal pattern is attributable to a firm-size effect, not an overreaction effect.

The findings reported by De Bondt and Thaler (1985) were re-examined by Chopra et al. in 1992. Using the same data, Chopra et al. also form loser and winner portfolios. A big difference though is that they use a multiple regression model to incorporate the correlated variables size and betas. Using this multiple regression, they find an economically significant overreaction effect of about 5 per cent per year. This overreaction effect, however, has a pronounced January seasonal. Furthermore, it is stronger for smaller companies than for larger companies.

More recently, Benou and Richi (2003) examined the long term reversal pattern for a sample of large U.S. firms. These firms experienced significant stock price declines of more than 20 per cent

(12)

during a specific month. Their data contained firms from the Standard & Poor 100 index from the period of may 1990 to may 2000. They report that “six and 12 months after their initial price decline, the stocks of large firms earn approximately 4 and 12 per cent more than expected.” The overreaction effect was most pronounced for the technology and manufacturing sub-samples, where technology stocks appear to experience the largest and strongest reversal.

2.2 Evidence on Short Term Overreaction Effect

Howe (1986) was the first to research the overreaction effect for the short term. His research approach, and that of Brown and Harlow (1988), differed considerably with that of later research. The authors used weekly and monthly data respectively, whereas later research used daily data for investigating the short term overreaction effect.

Howe (1986) finds evidence of stock market overreaction for both the AMEX and NYSE. In his study he uses 18 years of weekly stock return data from the CRSP database. Weekly data is used because it possesses statistical properties closer to those assumed for regression analysis. Inclusion in the sample meant an increase or decrease of 50 per cent in the event week. In his study, he finds strong support for the overreaction effect for both winners and losers. Over the subsequent ten weeks, the losers (winners) show returns of 13.8 per cent (-13.0 per cent). Howe (1986) in contrast to De Bondt and Thaler (1985) finds the effect of overreaction to be short term.

Brown and Harlow (1988) found support of the short term overreaction effect. They used the CRSP database to select NYSE firms from January 1946 to December 1982. The trigger event is defined as stocks with residual returns that lose or gain between 20 and 60 per cent. In addition, the event lengths range from one to six months. If a stock meets all these requirements, it is examined for overreaction. They conclude their results with: “taken together, the short- and long term responses to negative events suggest that the stock market generates economically significant one-month corrections. (p. 12)”

Atkins and Dyl (1990) were the first to use daily data. They put the emphasis on the bid-ask spread in explaining the reversal of large changes in stock prices. It is possible that the short term reversals accruing to losers may not be because of a reversal, but it may reflect a return of transaction prices to the bid-ask average. Atkins and Dyl use daily data of NYSE stocks, listed on the CRSP files from January 1975 to December 1984. The authors find evidence that the stock market overreacts, but the magnitude of the overreaction is small compared to the bid-ask spread. They interpret the finding as being consistent with a market that is efficient after transaction costs are considered.

(13)

for the day after and rises cumulatively to 2.2 per cent by the second day. They stress that this multiple day price adjustment is observed in a sample of firms that is largely devoid of methodological problems that might explain the reversal phenomenon. They find this long recovery period of the stock price reversal of major interest of this phenomenon. Because, they argue, such a slow recovery is inconsistent with the notion that market prices fully and quickly reflect relevant information.

Cox and Peterson (1994) examine stock returns following large one-day price declines and find that the bid-ask bounce and the degree of market liquidity explain short term price reversals. However, unlike Atkins and Dyl (1990) they do not find evidence consistent with the overreaction hypothesis, instead they observe that securities with large one-day price declines perform poorly over an extended time horizon.

Akhigbe et al. (1998) argued that the article of Atkins and Dyl (1990) does not reflect a round-trip transaction cost faced by investors reacting to the event. Akhigbe et al. re-examined the work of Atkins and Dyl (1990) and conclude that the there are significant stock price reversals, especially for loser portfolios. Nevertheless, profits, net of transaction costs, are not possible. The market therefore does not exhibit an overreaction to the announcement. They see this as providing support for the weak-form market efficiency.

Finally Ma et al. (2006) investigated the market overreaction effect for the stocks with the largest daily percentage increases or decreases in price reported in The Wall Street Journal between January 1996 and December 1997. They found evidence of the stock price overreaction effect for the NASDAQ gainers and losers, with reversals of -1.76 per cent and 4.5 per cent respectively. No such evidence is found for their NYSE samples of gainers and losers. The reversal of the stock returns occurs within a two-day post-event period. Furthermore, they found that both firm size and prior stock performance are statistically significant factors in determining the overreaction effect.

2.3 International Evidence on Stock Price Overreaction

In this sub-section, research performed on markets outside the U.S. will be discussed. Research has been conducted on large financial markets outside the U.S., which will be discussed first. Thereafter, the rest of Europe will be discussed, followed by stock markets of emerging countries. The research field on the overreaction effect is not limited to stock markets, therefore articles on overreaction outside stock markets will be discussed as well. In line with the previous section, a classification in long- to short term overreaction effect will be given.

2.3.1 Overreaction Effects on Large Financial Markets Outside the U.S.

(14)

documented long term overreaction on the Tokyo Stock Exchange. Baytas and Cakici (1999) test for long term overreaction in 7 industrialized countries and find no evidence of overreaction in the US. However, returns to long term contrarian strategies in other countries seem to be generally significant. The other countries include UK, Canada, Japan, Germany, France and Italy. Moreover, they find that in the majority of these countries, while returns to arbitrage portfolios based on price are higher than those based on size, the latter generally outperform the winner-loser arbitrage portfolios.

Research on short term overreaction for large financial markets outside the U.S. has been conducted by Brookfield (1993) and Otchere and Chan (2003). Brookfield indicated the London Traded Option Market does not overreact. As for Hong Kong, Otchere and Chan (2003) found evidence of overreaction, which is more pronounced for winners than for losers. However, abnormal profits obtained from exploiting such a phenomenon are economically insignificant after controlling for transaction costs. Their conclusion is that the Hong Kong Stock market is efficient in the weak form.

2.3.2 Overreaction Effects on European Markets

For long term effects, Mun et al. (1999) investigated the German and French stock markets using monthly data. They found that contrarian portfolio trading strategies to be successful on both markets. Their results are consistent with investor overreaction because higher returns are not correlated with increases in the risk coefficients.

For short term overreaction effects in Europe, Galariotis (2004) and Diacogiannis et al. (2005) investigated the Greek stock market. Both found overreaction on the Athens Stock Exchange (ASE). Galariotis (2004) reports that short term contrarian profits are present in the ASE. He used weekly stock data, and reported the findings as evidence of the short term overreaction effect. Diacogiannis et al. used an event study methodology in which the event is defined as an increase or decrease in the stock price that activates the price limit for one, two or three days. Their findings confirm the occurrence of short term overreactions on the ASE during the period under investigation.

2.3.3 Overreaction Effects in Emerging Markets

(15)

trading strategy. Wang et al. (2004) investigated the overreaction effect for the main Chinese stock markets in Shanghai and Shenzhen and report overreaction among domestic shareholders. Gunasekarage and Power (2005) examined the stock market overreaction hypothesis using monthly share returns for the equities traded on the Colombo Stock Exchange in Sri Lanka. Their findings support the overreaction hypothesis in this emerging market, and find that the observed overreaction is asymmetric; it is more pronounced for losers than for winners. Furthermore, they find that the month of the year seems to be affecting the overreaction findings.

2.3.4 Overreaction Effects Outside Stock Markets

The overreaction effect has been found outside stock markets as well. Using short term data, Fung et al. (2000) found evidence of overreaction for the index futures of US and Hong Kong. Vergin (2001) found that overreaction is persistent in point spreads among NFL betters. Larson and Madura (2001) examined exchange rate changes following extreme one-day fluctuations for currencies in industrialized and emerging markets. An overreaction effect for currencies in emerging markets and an underreaction effect for currencies in industrial markets are found.

2.4 Summary of the Discussed Literature and Hypothesis Formulation

Overall, there has been extensive research on the topic of overreaction. To give a more plain view of the articles discussed in this section, Table I is formed to present the differences in long- to short term overreaction. The first row, called effect, shows whether the examined overreaction effect was short, medium or long term. The second row describes the authors who performed the research, followed by the third and fourth row that describe the years for which the overreaction effect is studied. The next row specifies the stock market from which the data is retrieved. After that, a row indicates for the statistical significance, which is broken down in the categories of winners and losers. The same line of reasoning holds for the following row, which indicates whether the authors examined stock from small firms, large firms or both. The last row specifies whether the results are significant after transaction costs, also called economical significant.

A second sub hypothesis will research the difference in stock overreaction effect for winner and loser firms. This hypothesis relates to the research of Nam et al. (2001) who found that negative returns reverse to positive news quicker than do positive returns to negative ones. They assign this asymmetry to the mispricing behavior on the part of investors who overreact to certain news.

(16)

16 J.W.G. Noorda september 2007

Table I

Summary of the Overreaction Literature Sorted by Short- to Long term Effects

Effect Data Statistical Significant Size of the Firm Persistent after

Long Short Authors Year # Months Basis Market Winners Losers Large Small Transaction Costs?

x De Bondt & Thaler (1985) 01/1926 – 12/1982 684 Monthly U.S. (NYSE) Yes yes n.i.* n.i.*

x Chan (1988) 01/1926 – 12/1982 684 Monthly U.S. (NYSE) Yes yes no

x Ball & Kothari (1989) 01/1930 – 12/1979 600 Monthly U.S. (NYSE) No no no

x Zarowin (1990) 01/1933 – 12/1977 540 Monthly U.S. (NYSE) Yes no yes no

x Power, et al. (1991) 01/1973 – 12/1987 180 Monthly U.K. top 30 Yes yes n.i.* n.i.*

x Chopra, et al. (1992) 01/1926 - 12/1985 720 Monthly U.S. (NYSE) Yes yes yes

x Newton da Costa (1994) 01/1970 – 12/1989 240 Monthly Brazil (São Paulo SE) no Yes yes n.i.*

x Clare & Thomas (1995) 01/1955 – 12/1989 420 Monthly U.K. (London SE) Yes yes n.i.*

x Dissanaike (1997) 01/1962 – 12/1990 348 Monthly U.K. FT 500 index Yes yes yes

x Baytas & Cakici (1999) 01/1982 – 12/1991 120 Annual Industrialized (incl. U.S.) Yes yes no

x Mun, et al. (1999) 01/1991 – 12/1996 72 Monthly French & German SE Yes yes yes

x Zamri & Hussain (2001) 01/1986 – 12/1995 120 Monthly Malaysia (Kuala Lumpur SE) Yes yes yes

x Benou & Richi (2003) 05/1990 – 05/1999 120 Monthly U.S. (S&P 100) n.i.* Yes yes n.i.* yes

x Iihara, et al. (2004) 01/1975 – 12/1997 276 Monthly Japan (Tokyo SE) Yes yes n.i.*

x Wang, et al. (2004) 08/1994 – 08/1999 180 Monthly China Yes yes n.i.*

x Chiao & Hueng (2005) 01/1975 – 12/1999 300 Monthly Japan (Tokyo SE) Yes yes n.i.*

x Chiao, et al. (2005) 01/1975 – 12/1999 300 Monthly Japan (Tokyo SE) Yes yes n.i.*

x Gunasekarage & Power (2005) 01/1989 – 12/2003 180 Monthly Sri Lanka (Colombo SE) Yes yes n.i.*

medium Howe (1986) 01/1963 – 12/1981 228 Weekly U.S. (NYSE & AMEX) Yes yes n.i.* n.i.*

medium Brown & Harlow (1988) 01/1946 – 12/1982 444 Monthly U.S. (NYSE) Yes yes n.i.* yes medium Galariotis (2004) 01/1990 – 12/1999 120 Weekly Greece (Athens SE) Yes yes no

x Atkins & Dyl (1990) 01/1975 – 12/1984 120 Daily U.S. (NYSE) Yes yes n.i.* no

x Bremer & Sweeney (1991) 01/1962 – 12/1986 300 Daily U.S. (Fortune 500) n.i.* Yes yes n.i.* yes

x Brookfield (1993) 01/1989 – 01/1990 24 Daily U.K. No no no

x Cox & Peterson (1994) 01/1963 – 06/1990 330 Daily U.S. (NYSE) No no no

x Akhigbe, et al. (1998) 01/1992 – 12/1992 12 Daily U.S. (NYSE) No no no

x Otchere & Chan (2003) 03/1996 – 06/1998 27 Daily Hong Kong (HKSE) Yes yes no

x Diacogiannis, et al. (2005) 01/1995 – 12/1998 48 Daily Greece (Athens SE) Yes yes n.i.*

x Ma, et al. (2006) 01/1996 – 12/1997 24 Daily U.S. (NYSE & NASDAQ) Yes no yes n.i.*

► n.i.* stands for not investigated. Where this notation is used, the research did not specify for the examined variable.

(17)

Table I shows that all of the articles on short term overreaction, except for Brown and Harlow (1989), use daily data with a total sample between 1 and 27½ year. The evidence of overreaction after transaction costs is divided. Four articles find that contrarian strategies can be profitable and five papers do not find economically significant returns, only one research is undecided for the short term.

As can be seen, most research has been conducted on US stock, but little attention has been paid to European market. In addition, none of the studies investigated the overreaction phenomenon for the Dutch stock market.

This study attempts to address the gap in research literature by focusing on the Dutch stock market, the research hypothesis is therefore formulated as:

There is no stock market overreaction effect for stocks traded on the Dutch stock market. The following sub hypotheses will further specify the research theme:

Hsub1: Stock of small firms do not show a larger overreaction effect than do stock of large firms. Hsub2: Stock of winner firms do not show larger overreaction effect than do stock of loser firms

3 A Brief Review of the Dutch Stock Exchange

The Dutch equity market has recently (April 4th_{2007) become part of the NYSE Euronext, and is the} world’s biggest and most liquid stock market indices group. For the Dutch stock market, there are three distinct indexes, apart from the AMS ALL, which contains all equities traded on the Dutch market. All three indices (AEX, AMX and AScX) are designed to reflect general trends in the trading of shares listed on Euronext Amsterdam. It is made up of shares by the top 75 (25 per index) companies, listed on Euronext Amsterdam. This is done in such a way that it is suitable to serve as the underlying value for index-linked products such as derivatives.

Furthermore, the Amsterdam Exchanges-index (AEX), Amsterdam Midkap Index (AMX) and the Amsterdam Small cap index (AScX) only include shares issued by companies deemed to be representative of the Dutch equity market. 3

3.1 AEX

The AEX was compiled by the AEX-Optiebeurs in 1983. Over the years, the composition of this stock index has been changed a number of times, the last revision was on March 2nd_{2007. It is a weighted} index based on the prices of shares of the 25 leading companies, based on traded volumes, listed on Euronext Amsterdam, also called the blue chips. The AEX-index consists exclusively of shares issued

(18)

by companies that have been admitted to listing on Euronext Amsterdam’s Official Market. Companies with more than one listing on Euronext Amsterdam will not be included more than once.

On January the third was the start of the AEX at a base level of 45.38 points (100 points divided by the conversion rate of the Euro-Dutch guilder 2.203714_{). In case of a dividend payment the} index level is adjusted for several days to prevent the index from falling to fast on one single day. This differs from for instance the German equity index, the German DAX, which is not corrected for dividend yield. In order to correctly compare both indices, one has to add three per cent to the yearly yield of the AEX.5

3.2 AMX

The AMX is the equity index that represents the medium-sized shares traded on the Dutch equity market. The AMX was introduced on October 1995 and contains the shares from number 26 to 50 stocks in descending order of traded volume. The index is revised on the first trading day of March. However, between times it can be revised due to the splitting up of shares and the like.

3.3 AScX

The AScX is the latest index developed for Euronext Amsterdam and is part of the Amsterdam Index Range. As from March 2nd_{2005 the AScX index is calculated and disseminated. Euronext provides a} reconstructed history for the period since July 2000. The aim of the AScX is to increase visibility and encourage liquidity of Dutch small caps. Furthermore, it will provide investors and issuers with a tool to track this market segment. The AScX is a weighted index based on the traded volume of 25 small caps listed on Euronext Amsterdam. The shares from number 51 until 75 in terms of effective turnover traded on Euronext Amsterdam will be included in the AScX, taking into account free float and velocity criteria.

4 Data

and

Methodology

This research tests for the stock market overreaction effect for the large percentage gainers and losers of the Dutch market. Based on evidence provided in finance literature, the a priori expectations suggest that large reversals will follow a large gain or loss in the short term (Howe (1986), Liang and Mullineaux (1994), Ferri and Chung-ki (1996), Galariotis (2004), Diacogiannis et al. (2005) and Ma et al. (2006)). In this section first a description of the data is given. Second, the methodology will be explained and discussed.

4_{Source: www.xe.com date: 20/7/2007}

(19)

4.1 Data

The data that is used for this research is acquired from multiple sources. Dutch newspapers, internet sites (among others www.fd.nl) and online databases (Datastream) are used and the various sources have been combined into a single large database.

4.1.1 Daily Data

The reasoning behind the choice of daily data stems from the methodology used by Brown and Warner (1985). They report the following: “Although explicit recognition of the characteristics of daily data can sometimes be advantageous, for example in cases involving variance increases or unusually high autocorrelation, the characteristics of daily data generally present few difficulties in the context of event study methodologies. Furthermore, some of the paper’s results indicate a striking similarity between the empirical power of the event study procedures and the theoretical power implied by a few simple assumptions and ‘back of the envelope’ calculations. This reinforces the view that the use of daily data is straightforward.”

And as many recent research has used daily data in examining short term overreaction (Larson and Madura (2001), Akhigbe et al. (2002), Otchere and Chan (2003), Diacogiannis et al. (2005), Ma et al. (2006), Michayluk and Neuhauser (2006), and Schaub (2006)), this research will follow the same reasoning.

4.1.2 Selection of the Gainers and Losers

The sample firms are selected using Datastream daily Total Return Index (RI) data. This daily index data is preferred over the daily return data, because RI data is adjusted for dividends. This is important because otherwise the sample could be biased by showing large decreases that could be caused by a stock going ex-dividend. This was checked for the daily return data, and indeed 10 per cent of the loser events in the AEX and AMX were on ex-dividend days.

(20)

As noted by Bremer and Sweeney (1991) and Cox and Peterson (1994), using reported returns potentially has the following two biases. First, it is possible that equities with a very low price may have large positive (negative) returns. These large changes are then followed by reversals that actually only reflect oscillation between bid and ask prices. Therefore, only stocks with an average price of over €5 are included in the sample. Second, stocks with an average trading volume of less than 5.000 over the event period under investigation, are excluded from the sample.

These criteria were used to select stocks for the period between March 2005 and March 2007. The reasoning behind the start date of March 2005 is that this is the date when the AScX was founded. For the calculation of the estimation period of the AScX index, Euronext provides a reconstructed history for the index since July 2000.

All firms that were delisted, due to a merger or acquisition, during the sample period of March 2005 to March 2007 were checked for inclusion. However, only events are included to the extend that they meet the requirements of sufficient trading days surrounding the event date. The AMS all share index, which proxies for the market index, includes these shares as well. By doing this, the sample is checked for the survivorship bias. When this is not checked for, it often causes the results of studies to skew higher because only companies that were successful enough to survive until the end of the period are included (Brown et al., 1992).

Currently, July 2007, there are 139 firms traded on the Amsterdam stock market. After taking into account all the requirements of inclusion in the sample, 45 stocks of companies are excluded. The total sample consists of 708 observations. See Table II for the resulting sample of firms, divided in Total AMS, AEX, AMX and AScX gainers and losers.

Table II

AMS, AEX, AMX and AScX Gainers and Losers

Total AMS AEX AMX AScX

Gainer Loser Gainer Loser Gainer Loser Gainer Loser

Total 438 270 56 50 106 75 132 74

Table II shows that there is a big difference between the number of observations from the AEX on the one side and the AMX and AScX on the other side. This shows that for the years March 2005 to March 2007 the AScX has the most ‘big’ gainers. Furthermore, the number of losers is double for AMX and AScX in comparison with the AEX. Appendix I shows the list of firms in- and excluded in the sample.

4.1.3 Stock Data

(21)

trading days prior to the event date. This allowed the regression coefficients used in computing expected returns to be based on the daily returns for the previous half year, ending twenty days before the event (from -120 trading days to -21 trading days relative to the event date). It ends 20 trading days prior to the event date in order to remove bias to the regression coefficient due to the possible changes in return variability brought about by information leakage (see Schaub (2006:1166)). To measure the short term reversal effect, daily return data is used to twenty days after the event date. These data were available for the total observation period in Datastream.

Because the event study uses an estimation period that starts 120 trading days prior to the event date, and 20 trading days after the event date, the data set used contains daily total return data from September 2004 to April 2007.

4.1.4 Firm Specific News

Then, the website of Euronext, Het Financieel Dagblad and the NRC Handelsblad are examined, and checked for any firm-specific news. News of the firms listed on the AEX, AMX and the AScX is examined for a three-day window surrounding the event day. This is done to see if the gains or losses are associated with certain particular news, and to check differences between companies of various size. The news is categorized in six types of news and a seventh type that stands for no news, which is used when no news could be found on either website. The results of this categorization can be seen in Table III. For a complete overview of the firm-specific news Appendix II is added, these tables show the exact news per event date.

Table III

Categorization of the News for AEX, AMX and AScX Gainers and Losers Between March 2005 and March 2007

AEX AMX AScX

Gainer Loser Gainer Loser Gainer Loser Revenues & Earnings 19 34% 19 38% 23 22% 25 33% 25 19% 15 20% Mergers and Acquisitions 7 13% 1 2% 12 11% 1 1% 9 7% 0 0% Restructuring / strategic 5 9% 4 8% 5 5% 8 11% 11 8% 5 7% Analyst recommendation 10 18% 1 2% 16 15% 2 3% 18 14% 2 3% News from competition 6 11% 2 4% 8 8% 2 3% 5 4% 1 1% Other news 3 5% 18 36% 16 15% 11 15% 8 6% 5 7% None 6 11% 5 10% 26 25% 26 35% 56 42% 46 62%

Total 56 50 106 75 132 74

(22)

For the AMX it shows that a larger part of loser firms is associated with revenues and earnings news. Another noticeable difference is that of the analyst recommendation news, which is responsible for a big part of the winner firms, but which seems not to be a trigger for firms losing share value. Furthermore, an explanation for the evident increase in the ‘no news’ category could be because less information is known about smaller companies. Investors may tend to trade more on ‘rumors’, or ‘noise’ as less information is available to the public.

The AScX further confirms and extends the differences found between the AEX and AMX sample. The ‘no news’ category has increased to include more than half the observations for the loser firms. Furthermore, revenues and earnings news and analyst recommendations are the categories most associated with large price gainer firms.

4.1.5 Descriptive Statistics

The descriptive statistics of the main variables are presented in Appendix III. It shows the number of observations, each equity’s mean, median and standard deviation, as well as the maximum (minimum) stock price increase (decrease). For each equity that is included in the sample a test for normality is performed. This normality test consists of a skewness and kurtosis, together forming the ingredients for the Jarque-Bera test. It also shows the outcomes for the White’s test. Furthermore, when heteroskedasticity is found, a GARCH (1,1) test is performed. The last column shows the alpha and beta regression estimates for the whole sample period, all observations were included. For the regressions of the events, a subset of the data is used for the input of the market model.

For seven of the 94 examined companies there were less than 653 observations (the two year sample data set including the event window). An explanation for this could be that these firms were taken over, delisted or had their IPO on the Dutch exchange. The mean and median of all companies are near zero, which shows that on average the companies compensated their losses with gains. Because the mean is positive, there were slightly more positive days than negative ones. In the next two columns the minimum and maximum price change are displayed. The biggest daily increase of the observed sample was 25 per cent, the biggest loss reported in this sample 23.5 per cent.

The Jarque-Bera test (JB) shows that all 94 firms have a JB value above 5.9. The assumption that the stocks are normally distributed is therefore rejected. The Jarque-Bera test shows that the probability of being wrong when rejecting H0 (H0: residuals are normally distributed) is below 5 per cent for the regressions of the selected companies.

(23)

A common finding in time series regressions is that the residuals are correlated with their own lagged values. This serial correlation violates the standard assumption of regression theory that disturbances are not correlated with other disturbances. To measure the serial correlation in the residuals of the model, the Durbin-Watson statistic is used. The Durbin-Watson statistic for all companies is near two (between 1.75 and 2.26), which implies that there is little evidence of autocorrelation Therefore the null hypothesis cannot be rejected.

4.2 Methodology

This section describes the methodology used for testing for overreaction on the Dutch market. As described by Brown and Warner (1985), event type methodology is used here to test the short term overreaction hypothesis, and the extend to which the Dutch market is associated with the news announcements by analyzing excess returns. These excess returns are calculated using the market model approach, as done by Brown and Warner (1985). The stock returns are converted into series of simple returns. The sub-sections below will give reasoning for the chosen models and methodology. 4.2.1 Event Study Methodology

An event study, in finance research, is an analysis of whether there is a statistically significant reaction in financial markets to past occurrences of a given type of event, in this case a large change in stock price, which is hypothesized to affect public firms' market values.6_{It “examines the direction,} magnitude and speed of […] price reactions to various phenomenon” (Dombrow et al. 2000). Alternatively, as McWilliams and Siegel (1997) state: “An event study determines whether there is an ‘abnormal’ stock price effect associated with an unanticipated event. From this […] the researcher can infer the significance of the event”.

For a correct understanding and interpretation, some assumptions are made such as market efficiency (Fama 1970, McWilliams and Siegel 1997), the event has to be unanticipated and there should be no confounding effects (McWilliams and Siegel 1997). The assumption ‘event is unanticipated’ is of particular relevance here, as it states that abnormal returns are the result of a reaction.

Because event studies are subject to noise problems over extended time periods, this research analyses 41 trading days around the event day. These days are grouped in various time frames to perform a sensitivity analysis with different lengths of event windows.

4.2.2 Day 0 Definition

The day 0, or event day, is described best as the day where the price of a stock increases (declines) with more than 4 per cent. See for a more detailed specification section 4.1.2, which also gives reasoning for exclusion of stocks.

(24)

4.2.3 Event and Estimation Period

For stock price overreaction, a typical choice is 100 trading days before the start of the event period. The estimation period used in this research to compute the model parameters is from -120 trading days to -21 trading days relative to the event date. The event period captures the 40 trading days surrounding the event date (from -20 trading days to +20 trading days). All is summarized in Figure 1.

The event and estimation period are chosen so that it captures most variability of the security and insures statistical accuracy. It will always be a bit arbitrary to set a length of the event and estimation period, but the period used in this paper lies in the range of periods used in other event studies. For statistical accuracy, large estimation periods are observed. However, in order to obtain a standard normal distribution, the estimation period has to be at least thirty days (Campbell et al., 1997).

Figure I

Event and Estimation Period

Estimation

Period

Event

Period

0 +20

-20

-120

Kabir and Roosenboom (2003) used a period of 200 trading days when studying abnormal returns on the Dutch market. On the other hand, smaller estimation periods are also common in financial research. Asquith and Mullins (1986) used a 47-day estimation period and Atkins and Dyl (1990) used 300 randomly selected trading days. Similar to this research, Ma et al. (2006) used an estimation period of 100 trading days before the event period, as well did Cox and Peterson (1994) and Schnusenberg and Madura (2001).

4.2.4 Computation of Stock Returns

Before the obtained stock total return index data from Datastream can be used, the equities are converted into series of simple percentage returns. This is in accordance with the measurement used by ‘Euronext’, which sets the rules for the composition of the Dutch stock indices. The computation of the simple actual return is shown in equation 1.

t j

R , = ((Pj,t −Pj,t−1)/Pj,t−1)*100 (1) Where:

(25)

4.2.5 Normal Performance Model

According to Strong (1992), it is crucial for a successful application of the event study methodology to use a correct specification of the ‘normal’ return. The market model is the most common model in overreaction literature for the estimation of the normal return (Schaub (2006), Ma et al. (2006)). With the market model a measure is provided that considers both differences in risk in the sample and the effect of general market fluctuations during the period being studied. This approach is similar to Atkins and Dyl (1990). A GARCH regression is used, when a White’s test is insignificant, to estimate the parameters, αi, βi, of the model presented in equation 2, where t is -120 to -21 trading days.

t j

R , =

α

i +

β

iRm,t +

ε

j,t (2) Where:

Rj,t = the return earned by the jth_{security on day t,}

Rm,t = the return on the value weighted Dutch market for day t, αi, βi = the GARCH estimates of the market model, and

t j,

ε

= an error term.

Previous event studies use the OLS method to estimate the parameters, which inherently yields not robust results due to the assumption of homoskedasticity. Homoskedasticity assumes a constant variance of error terms var(εt) =σ2. The problem is that when the error terms are assumed homoskedastic (OLS), but are in reality heteroskedastic, the error term estimates might be wrong. Especially in the context of high frequency stock data, as the daily data used in this research, it is very unlikely that the variance of the error terms is constant over time.

GARCH does exactly this, it allows accounting for variance effects of the error terms over time, and allows the errors of daily stock return to be heteroskedastic. Therefore, in this study the market model is adjusted to fit the GARCH (1,1) specification, shown in equation 3 through 5.

t j R , =

α

i +

β

iRm,t +

ε

j,t (3) t j,

ε

= v_j,_t

σ

_j,_t vj,t ~ N(0,1) (4) = (5) 2 ,t j

σ

2 1 , 1 0

+

α

ε

jt−

α

Where:

Equation 3 = the main equation of the market model as described by equation 2, Equation 4 = the GARCH term that specifies the equation for the standard error, and

(26)

For this data set a White’s test is performed which provides a scientific approval of using the GARCH method. An overview of the results of this test can be found in Appendix III.

4.2.6 Computation of Abnormal Returns

After estimating the parameters

α

ˆ

_iand with the GARCH model, the parameters are used to calculate the abnormal return of each security j for day t during the period following the event day. This is computed as presented in equation 6, where t takes values between -20 and +20 trading days.

i

β

ˆ

t j AR, =

R

j,t

−

α

ˆ

i

−

β

ˆ

i

R

m,t (6) Where:

ARj,t = the abnormal or excess return on security j on day t, Rj,t = the return earned by the jth security on day t,

Rm,t = the return on the value weighted Dutch market for day t, and αi, βi = the GARCH estimates of the market model.

4.2.7 Aggregation of Abnormal Returns

Then, for a number of N events, an aggregation of the abnormal return is performed. This aggregation makes it possible to make a statistical analysis of the sample over the event period. The abnormal returns are averaged across N securities on each event day to form an average abnormal return over the interval t= -20 to +20. This is expressed in equation 7.

t j AAR , =

∑

(7) = N i t j AR N 1 , ) / 1 ( Where:

AARj,t = the Average Abnormal Return, ARj,t = the Abnormal Return, and

N = the number of firms that satisfy the ‘trigger’ demands.

In absence of abnormal stock price behavior, the expected value of AARj,t is zero. To test the significance of AARj,t, the t-test statistic as shown in equation 8 will be used. This robustness check is performed to assure that the observed abnormal returns are statistically different from zero (Strong, 1992). Below equation 8, the formal specification of the null hypothesis is given by equation 9 and 10.

(27)

The formal specification of the null hypothesis: H0 =

N

AAR

S

AAR

t j t j

/

)

(

, , = 0 (9) Ha =

N

AAR

S

AAR

t j t j

/

)

(

, , ≠ 0 (10) Where:

AARj,t = the Average Abnormal Return,

S(AARj,t) = the estimated standard deviation of the average abnormal returns for day t, and N = the number of events.

If the null hypothesis is rejected, the conclusion will be that the abnormal return is statistically different from zero. Hereafter, the cumulative abnormal returns are calculated according to the method described in section 4.2.8.

4.2.8 Computation of the Cumulative Abnormal Returns

To determine the cumulative effect, the a

verage abnormal return

s are accumulated over various sub periods of k days from t to t+k to form cumulative average abnormal returns (CAAR). The expected cumulative average abnormal return should be zero. The significance of CAARt,t+k is estimated using the test statistic, as shown in equation 12.

t

= t j AAR k t t k CAAR , ˆ 2 1 ,

σ

+ (12) Where: AAR = CAAR jt k + t t = k + t t,

∑

, τ (13) And: 2 21 120 ) ( 100 1 ˆ = AARt AAR = t 2 AAR

∑

− − −

σ

(14)

4.2.9 Regression of the News Categories

(28)

results found with the abnormal returns. The regression will also show the differences of impact between the different types of news.

The regression will be performed on the total winner and loser sample of stocks listed on the AEX, AMX and AScX. A dummy variable (D1 through D7) for each of the news categories will make the regression testable. Equation 15 will be used to estimate the regression:

i

it

=

D

R

β

1 1

+

β

2 2

+

β

3 3

+

β

4 4

+

β

5 5

+

β

6 6

+

β

7 7

+

ε

(15) Where:

D1 = the news related to Revenues & Earnings, D2 = the news related to Mergers and Acquisitions,

D3 = the news related to Restructuring or strategic changes, D4 = the news related to Analyst recommendations,

D5 = the news related to News from competition,

D6 = Other news, not represented by catgerories 1 through 5, and D7 = denotes the value of 1 when no news is found.

4.2.10 Size and Index Diversification Hypothesis

To mitigate the size difference between AEX, AMX and AScX-index stocks, the value weighted market index as used by Euronext is used as a proxy for market returns of the AEX, AMX and AScX gainers and losers in the model. The weight of each equity in the index depends on several factors. To be precise the price of the stock, the weighting factor of a stock and the number of stocks in the portfolio (25 stocks each). Making a distinction between the three indices allows the research to search for differences between high capitalization stocks (AEX) and low capitalization stock (AScX and AMX). It is expected that stock with low market capitalization are more volatile, and less information is known about the firm, therefore overreaction should be more persistent in these types of stock.

This is also explained by the index diversification hypothesis, as used by Schnusenberg and Madura (2001), which states that indices consisting primarily of large stocks should exhibit a smaller over- or underreaction than indices with relative small stocks, according to the size effect. This implies the following relation between the indices examined with respect to absolute magnitude of the over- underreaction:

AEX<AMX<AScX

(29)

5 Empirical

Results

This section describes the results of the event study’s empirical testing. The event study is performed on the resulting sample of 94 companies with a total of 708 events. Where possible, a distinction between the different Dutch indices will be given, to mitigate the size difference. For the purpose of a clear overview, the results are grouped as follows. First the positive events (winners) and second the negative events (losers) will be described. Thereafter the results of the news categories regression are reported. This section will conclude with a summary and discussion of the main findings.

5.1 Did the Winners Overreact?

This section will give an overview of the results of the positive (winner) events, discussed by the AAR, the CAAR. Furthermore, an overview and comparison of three Dutch indices is presented to distinct between high capitalization stock and low capitalization stock.

5.1.1 AMS Total Share Sample

The Cumulative Average Abnormal Returns earned by stocks that exhibited a price increase on day t=0 are shown in Figure II. The AARs are computed using the market model with the AMS Total Share index as a proxy for the market.

Figure II

Cumulative Abnormal Returns for 438 AMS Stocks that Exhibited a Large Increase in Price at t=0

-2 0 2 4 6 -10 -8 -6 -4 -2 0 2 4 6 8 10 12 14 16 18 20 Event Day A bno r m a l R e tur ns

(30)

Panel A of Table IV displays the average abnormal returns of 10 days surrounding the event day (t=0). It furthermore includes the t-test value, the significance value, as well as the ‘AR ratio’. This AR ratio shows the percentage of AARs, from the total sample of 438, which are positive. As expected the event date is highly significant (significantly different from zero), with an AR ratio of almost 100 per cent. And is the result of stocks being included in the sample.

As turned out, one event that initially showed a price increase over 4 per cent became negative after being transformed into an AR. A second significant result is found on day one. If the stock market would have overreacted, the expected return on day one is negative. Instead, the Average Abnormal Return on day 1 is positive with 0.24 per cent. This is statistically significant, the discussion whether it is economically significant is left to the section summary and discussion (5.4).

Table IV

Daily Abnormal Returns (per cent) for AMS Stock That Exhibit Large One Day Price Increases From March 2005 Trough February 2007 (N = 438)

Abnormal

Day (t) Return t-value P-value AR Ratio

Panel A. Average Abnormal returns

-5 -0.036 -0.373 0.709 46.12% -4 -0.015 -0.240 0.810 46.58% -3 -0.140 -2.025 0.044 42.47% -2 -0.076 -1.000 0.318 44.75% -1 0.099 1.245 0.214 52.51% 0 5.324** 44.898 0.000 99.77% 1 0.244* 2.000 0.046 48.63% 2 -0.131 -1.287 0.199 42.24% 3 -0.026 -0.256 0.798 47.72% 4 -0.101 -0.954 0.340 44.06% 5 -0.049 -0.514 0.608 48.63%

Panel B. Cumulative Average Abnormal Returns

-10 to -3 -0.367* -1.656 0.099 46.80% -2 to -1 0.024 0.197 0.844 48.63% 0 5.324** 44.898 0 99.77% 1 to 3 0.086 0.465 0.642 47.26% 1 to 10 -0.600* -1.903 0.058 45.89% 1 to 20 -0.345 -0.788 0.431 45.66% ** Significant at the 0.01 level (two-tailed tested)

* Significant at the 0.10 level (two-tailed tested)

(31)

5.1.2 AEX, AMX, AScX Share Sample

Here a comparison between the three different Dutch stocks indices for the winner events is presented. Figure III illustrates the cumulative abnormal returns for the three indices. The AMX and AScX show a similar pattern as does the AMS in Figure II. After the large price increase, stocks of the AMX and AScX have a second increase on day 1. This increase is significant for the AScX, but not for the AMX. The AEX has some small significant AARs on days 5, 6 and 9, but these are relatively small and go up and down. Just as the AMS shows in Figure II, after 10 trading days of up and down AARs, the AARs are on their initial level of the event day. so on average there is no reversal.

Figure III

Cumulative Average Abnormal Returns of the Winners for the Three Dutch Indices

-2 -1 0 1 2 3 4 5 6 7 -10 -8 -6 -4 -2 0 2 4 6 8 10 Eve nt Day Ab n o r m a l Re tu r n AEX AMX AScX

Table V shows the daily average abnormal returns for winner events on the AEX, AMX and AScX. Again, the large positive returns on the event day are the result of the large increase in price that caused the stocks to be included in the sample. For the AEX, six out of the 10 post event days are negative and sums up to an insignificant 0.6 per cent decrease. More or less the same holds for the AMX sample, where seven negative out of 10 AARs accumulate to an insignificant negative 0.55 decrease. The CAAR shows no significant returns for the AMX.

The AScX differs from these two, by showing significant AARs on day -1 and 1. A positive 0.304 per cent AAR on day -1 could indicate that some news already reached the market. However, the 0.684 AAR on day 1 is also very significant and could be a sign that not all news was adopted on the event day. The AScX eventually, after 10 trading days, shows a small, insignificant, 0.3 per cent CAAR, where a negative one is expected when overreaction is persistent. As is clearly not the case for the winner events of all three indices.

(32)

Table V

Daily Abnormal Returns (per cent) for AEX, AMX and AScX Stock That Exhibit Large One Day Price Increases Between March 2005 and March 2007

AEX (N=56) AMX (N=106) AScX (N=132) Day (t) AR t-value AR t-value AR t-value

Panel A. Average Abnormal returns

-5 -0.168 -1.011 -0.076 -0.576 -0.061 -0.512 -4 -0.136 -0.797 0.029 0.237 -0.089 -0.937 -3 -0.238 -1.512 -0.058 -0.449 -0.043 -0.327 -2 -0.114 -0.664 -0.155 -1.325 -0.051 -0.566 -1 0.147 0.908 0.154 1.199 0.304* 2.575 0 4.950** 16.018 4.872** 24.645 5.302** 32.165 1 -0.123 -0.464 0.413 1.654 0.684* 2.857 2 -0.010 -0.051 -0.185 -0.950 -0.136 -0.809 3 0.016 0.086 0.015 0.074 0.089 0.467 4 -0.171 -0.949 -0.261 -1.190 0.096 0.486 5 0.330* 2.331 -0.355* -2.333 0.002 0.011

-10 to -3 -1.360 ** -3.306 -0.577 -1.353 0.235 0.571 -2 to -1 0.032 0.133 -0.001 -0.006 0.253 1.652 0 4.950 ** 16.018 4.872** 24.645 5.302** 32.165 1 to 3 -0.116 -0.347 0.243 0.634 0.638* 1.964 1 to 10 -0.597 -0.844 -0.558 -0.890 0.321 0.554 1 to 20 -1.091 -1.014 -0.258 -0.272 0.861 1.158 ** Significant at the 0.01 level (two-tailed tested)

* Significant at the 0.10 level (two-tailed tested)

the total sample of winners dropped from 438 to 243 to 135 events. The sample size declined by over two-thirds, however, the patterns of the AARs and the magnitude continue to be found in this smaller sample. General conclusions were not altered by the choice of a different trigger value for the winners.

The outcomes of these tests, which were done for only the AMS Total Share sample, can be found in Appendix VI.

5.2 Did the Losers Overreact?

This section will give an overview of the results of the negative (loser) events, discussed by the AAR, the CAAR. Furthermore, an overview and comparison of three Dutch indices is presented to distinct between high capitalization stock and low capitalization stock.

5.2.1 AMS Total Share Sample

(33)

Figure IV

Cumulative Abnormal Returns For 270 AMS Stocks that Exhibited a Large Decrease in Price at t=0

-10 -8 -6 -4 -2 0 -10 -8 -6 -4 -2 0 2 4 6 8 10 12 14 16 18 20 Event Days A bno r m a l R e tur n

Figure IV also shows that the CAAR for days -10 to -1 is negative. Two out of the 10 trading days preceding the negative event are positive. Moreover, as Table VI shows in Panel B, these preceding days are cumulated significantly negative at the 5 per cent level with a total of over two per cent.

Table VI

Daily Abnormal Returns (per cent) for AMS Stock That Exhibit Large One Day Price Decreases Between March 2005 and March 2007 (N=270)

Abnormal

Day (t) Return t-value P-value AR Ratio Panel A. Average Abnormal returns

-5 -0.093 -0.901 0.369 42.96% -4 -0.221** -2.358 0.019 44.81% -3 -0.087 -0.879 0.380 50.00% -2 -0.204** -2.329 0.021 40.37% -1 -0.112 -1.188 0.236 42.96% 0 -5.260** -30.181 0.000 0.37% 1 0.035 0.179 0.858 54.07% 2 -0.219 -1.568 0.118 45.93% 3 -0.157 -1.170 0.243 42.96% 4 -0.323** -2.518 0.012 34.44% 5 -0.459** -3.747 0.000 40.00%

-10 to -3 -0.594* -2.224 0.027 45.92% -2 to -1 -0.316** -2.312 0.022 40.37% 0 -5.260** -30.18 0.000 0.37% 1 to 3 -0.340 -1.34 0.181 48.51% 1 to 10 -1.758** -4.312 0.000 37.41% 1 to 20 -3.139** -5.518 0.000 35.93% ** Significant at the 0.01 level (two-tailed tested)

(34)

Panel A of Table VI shows a very small and insignificant ‘reversal’ on day 1, after which the AARs continue to be negative. Individually, the AARs are significant at the 1 per cent level for trading days 4,5,6,9,11,17 and 19.

The CAAR for the total AMS negative event sample, is significant ate the 1 per cent level for the cumulated post event returns of period 1 to 10 and 1 to 20. This sums up to a cumulated 8.399 per cent decline, calculated from the event day on. After the large price decrease on t=0, over 3 per cent is added to the CAAR by trading day 20. The last column of Table VI, the AR Ratio, shows that over the whole period there are more negative returns than positive, except for day 1 in Panel A. The discussion whether the large momentum found for losers of the Dutch stock market is economically significant, is left to the section summary and discussion (5.4).

5.2.2 AEX, AMX, AScX Share Sample

In Figure V, the differences between the losers of the AEX, AMX and AScX are graphically presented. The pattern of the AScX resembles that of the AMS Total Share most, where mostly negative AAR are reported during the event window.

Figure V

Cumulative Average Abnormal Returns of the Losers for the Three Dutch Indices

-10 -8 -6 -4 -2 0 2 -10 -8 -6 -4 -2 0 2 4 6 8 10 Event Day Ab n o rm a l Re tu rn AEX AMX AScX