1
Market Impact of Disposition Effect: Common Reference Points of Chinese Investors
Student Number: s2177153 Name: Wang Xuchao Study Program: Msc Finance Supervisor: Dr. Viola Angelini
University of Groningen, Faculty of Economics and Business
Abstract
This paper investigates market influence of disposition effect, using recent IPO aftermarket trading volume data in China. With the trading behavior investigation in the offer price level, I prove that the disposition effect is an important determinant of trading volume. The monthly maximum price is also discovered to be an important factor to affect the trading volume. The results suggest disposition effect is a prevailing anomaly and can influence the aggregate market behavior. I prove that the investors in emerging market share the similar reference points to make decision.
JEL classification: G10
Key words: Disposition effect; Trading Volume; IPO; China.
2
Introduction
An initial public offering (IPO) refers to the first sale of a company‟s stock to the public on a securities exchange. Initial public investors buy shares at the subscription price which implies they can define their returns in terms of the deviations from this identical price. Following the events of IPOs, the initial investors immediately obtain paper gains or losses when they have to decide whether to sell the shares straight away. In the past, the majority of the stocks around the world experienced positive initial returns. This is also known as „the IPO underpricing‟.
One of the interesting properties is that in most cases, the IPOs with positive initial returns are experiencing significant higher trading volumes than the IPOs with negative initial returns.
Standard finance theory is not able to explain this strange phenomenon. As the standard finance theory assumes rationality, investors should cut the losses decidedly and keep the profits for long. Alternatively, researchers use behavioral finance to explain the bias during the IPOs. Behavioral finance researchers consider that investors display several kinds of behavioral biases, which lead to irrational decisions. One of the behavioral biases is the disposition effect, as the investors tend to sell winner stocks early and hold on loser stocks. As the investors eagerly realize the initial gains, the trading volume is higher for the initially positive IPOs.
It is worthwhile to know to what extent the disposition effect influences the aggregate market behavior. In the studies of disposition effect, the relevant reference point is an important benchmark as the investors rely on it to define gains and losses. An investor compares his or her wealth with the target level of wealth, which is the reference point, to make the judgment.
If many investors set the same reference point to take action, the simultaneous behavior is significant enough to affect the market behavior. Because of the identical purchase price, the aftermarket performance of the IPOs offers a terrific environment to investigate the reference points for the disposition effect. Kaustia (2004) indicates IPO offer price is an important reference point for the U.S. investors, as the trading volume is influenced by disposition effect.
He also proves that the maximum and the minimum prices of the months are important reference points.
3
In this paper I attempt to shed light on the market impact of disposition effect among the Chinese investors by determining the existence of common reference points. For a sample of 77 firms from 2006 to 2011 which present negative initial return and exceed the offer price within one year‟s period, I document a significant increase in turnover of 29.3% in the offer price level. The results suggest that the investors sell more after a small gain with regard to the offer price.
For 983 firms which experienced the events of IPOs during the same observation period, the monthly highest price levels are important for the trading behavior of investors. The investors are more willing to trade when the prices surpass the highest prices during the previous month.
On the other hand, the investors‟ behavior bias is not significant for the monthly lowest price levels. My results extend the research of Kaustia (2004), as the investors show similar behavior bias in emerging market. The disposition effect is proved to have influence on the aggregate market behavior.
This paper is divided into four sections. Section 1 of this paper reviews the literature and introduces background for IPOs. The data collection and the research design are presented in section 2 and section 3, respectively. The results analysis, as well as the robustness check is presented in section 4. Section 5 provides a summary and conclusion of the research.
Literature Review
IPOs
Initial public offering is one of the most important events in financial markets all over the world. During the year of 2011, there were 1225 deals of IPOs worldwide, raising 169.9 billion U.S. dollars1. Asian companies have been the key drivers of the IPO recovery after the financial crisis.
The primary reason for a firm to go public is to raise additional equity capital. At the meantime, the founders of the firm and other initial shareholders gain opportunity to convert their wealth after the initial public offering. A firm can also earn publicity after the event of
1 See global IPO trends, 2012, Ernst & Young
4
initial public offering. There are also some costs of going public. Going public is expensive.
What‟s more, the firm has to disclose company information on a regular basis after going public. Ritter and Welch (2002) offer a detailed review of IPO activity.
There are several features of IPOs which are important for this study. Underpricing, which refers to the positive average initial return, has been observed in many countries (Loughran, Ritter, and Rydqvist, 1995). Yang, Wang and Jiang (2007) mention that the IPOs in the developing countries enjoy higher levels of underpricing comparing to the IPOs in the developed countries.
The selection of pricing mechanism plays an important role in IPOs. The efficiency of the IPO pricing mechanism has long been researched. Kucukkocaoglu and Sezgin (2012) summarize both theoretical models and empirical studies on the IPO pricing mechanism.
Book building, auctions, and fixed price offer are the most prevailing pricing methods of IPOs.
With book building, the underwriters involve road shows and take valuations from potential investors. In an auction, a price ranges is set for investors to reach a clearing price. The offer price is set first for the fixed price offers. They describe in detail the difference between those methods in this paper. Book building and auction are considered to be more superior over the fixed price method (see Benveniste and Spindt, 1989; Ritter, 1998).
The underwriter plays an important part in the event of IPO in many countries. Aggarwal (2000) describes that underwriters support the new issues in the U.S. market using stabilizing bids, aftermarket short covering, and penalty bids. Ellies, Michaely and O‟Hara (2000) discover that the underwriter‟s intervention on the market transaction behavior can be observed for months.
IPOs in the Chinese Market
The Chinese market is unique because of its rapidly changing features. Circumstances have changed tremendously during the last 20 years in the financial markets of this country. I introduce some recent articles here to summarize the market condition. Sohn et al. (2012) describe the characteristics of A-share IPOs in the Chinese markets. I highlight the following points with their help.
5
1. Significant IPO underpricing has been observed in this market (e.g. Chang et al., 2008;
Guo and Brooks, 2008 )
2. A-shares represent the tradable common shares for Chinese citizens; B-shares represent the tradable common shares for foreign investors; Government shares and legal entity shares are non-tradable; Employee shares are usually not tradable for one year.
3. The demand for IPO is very high for the domestic market. There are limited alternative investment opportunities in this country and the aggregate supply of new listing firms is insufficient. Those reasons can partly explain the high initial return of IPOs.
4. The pricing mechanism has changed in the year of 2010. The fixed price mechanism has been employed for a long time in which the offer price is determined by the underwriter and will not adjust until the listing date. The introduction of inquiry system in 2010 can be viewed as a major reform and cause more initial negative returns afterwards (see also Li, 2012). The new pricing mechanism is similar to the book building method.
5. Short selling was not allowed in this market until 2010. Investors can go short for 500 large securities in the market and it has tiny effects on IPO stocks. Taking long position dominates the market.
6. Until now, there is no option market in both stock exchanges.
7. Every year, there are up to hundreds of companies have their first issue in the A-share markets of China.
I emphasize that not only the potential behavior difference, but also the regulation difference can lead to different results for this research. The experimental design considers the unique market condition.
Disposition Effect
Standard finance theory is normative in nature and regards utility maximization as the core fundamental. A rational investor should maximize his or her utility by holding profits for long and cutting losses immediately. Standard theory finds it hard to explain the trading behavior during the IPO process. Researchers documented that the IPOs with positive initial returns are experiencing significant higher trading volumes than the IPOs with negative initial returns
6
(see Aggarwal, 2003; Chong, 2009; Reese Jr., 1998). Investors sell winner stocks early and hold loser stocks for longer period. This property can be summarized as disposition effect, which is one of the most important behavioral biases discovered in behavioral finance.
The most important theory in behavioral finance is prospect theory by Kahneman and Tversky (1979), which describes the choice under uncertainty. It is the fundamental of the disposition effect theories. Decision makers face several choices relative to a reference point, and they valuate the choices by an S-shaped valuation function. This utility function reflects risk aversion in the gain region and risk loving in the loss region as it is concave in the domain of gain, and convex in the domain of loss.
Shefrin and Statman (1985) provide the pioneer work on the disposition effect. They focus on behavioral theories on the investment decision making, especially the prospect theory. Under the behavioral framework, they explain the willingness to realize gains and reluctance to realize losses among investors. It is in the opposite of the standard theory. As introduced above, in those theories which assume rationality, investors should cut the losses decidedly and keep the profits for long. One another important point is that the disposition effect can be considered as the judgment against the purchase price. The viewpoint provides important intuition that the purchase price should be one of the essential reference points.
Odean (1998) analyzes household trading activities with individual accounts in a U.S.
brokerage to test the disposition effect hypothesis. He distinguishes „realized gain‟, „realized loss‟, „paper gain‟, „paper loss‟ and computes the proportions of gains realized and losses realized. He discovers strong evidence to support the disposition effect by proving significant differences between proportion of gains realized and losses realized. He shows that an investor has a greater tendency to sell stocks which are winners rather than losses with respect to the purchase price. The method is one of the major contributions for measuring the disposition effect, which was employed by several later researches.
Weber and Camerer (1998) design experiments to investigate if their subjects have tendency to sell more shares when price rises than it falls and the results show disposition effect. The purchase price and the previous period price are both important reference points according to their experiment. One important feature is that automatically selling largely reduces the disposition effect. Because subjects sell less when the shares have negative return with respect
7
to the purchase price in the experiments, it is reasonable to assume that the trading volume is higher for the stocks above the purchase price.
Grinblatt and Keloharju (2001) employ Logit regression method which enables them to distinguish different investor features and market characteristics. In this comprehensive study, they discover investors‟ reluctance to realize losses except in December and the relevance of historical return patterns. Sophisticated investors lay less emphasis on the past return patterns while the less sophisticated investors, particularly the domestic investors, tend to sell more with good past return patterns. Monthly high and low price is found to be relevant for the trading behavior. This intuition is very important for this study.
Though the professional traders are classified as sophisticated investors, they also display disposition effect. Locke and Mann (2005) examine the performance of professional traders and try to see if they tend to exhibit the disposition effect. The full-time professionals also hold on their losers significantly longer than gains. The traders who offset their losses slowly are less likely to be successful. Jin and Scherbina (2011) employ data on mutual fund holdings when the mutual funds replace managers. New managers save money over the following six month for the funds because they tend to sell the losers which the old managers are reluctant to sell.
Disposition Effect in China
There are a number of researches investigating the disposition effects in China. Feng and Seasholes (2005) provide a detailed analysis of the disposition effect on Chinese investors.
The paper is the winner of 2006 GSAM Quant Best Paper Prize in the Review of Finance. In this paper, they use the direct data from a brokerage which contain transactions and stock holdings information of investors and try to track the investors‟ investing career. Following the same methodology of Grinblatt and Keloharju (2001), they discover that sophistication and trading experience together can eliminate the phenomenon of holding losers too long.
Other characteristics also exert influence, including whether the investor diversify his or her portfolio from the start, the gender, and the age.
Chen et al. (2007) also use the Chinese brokerage account information to investigate the
8
decision making on investments. The investors are discovered to suffer from several behavioral biases including disposition effects, overconfident, and representativeness bias.
The disposition effect is considered to be stronger than the U.S. investors. Both papers use the data from a certain brokerage to represent the trading behavior. The degree of overall effect of disposition effect is not detailed elaborated.
Visaltanachoti, Luo, and Lu (2007) employing the regression method to research the disposition effect in this country with the aggregate market data from 1996 to 2003. By examining the cross-sectional results of the average holding period of the aggregate market, they imply that the investors present a strong disposition effect in A-share market for domestic investors, and the effect is not significant in B-share market for foreign investors.
Market influence of disposition effect
Several of researches highlight the impact of the disposition effect on the market-wide prospective. Lakonishok and Smidt (1986) find strong evidence that winner stocks have higher turnover than the loser stocks. It is one of the earliest articles addresses this phenomenon after the proposition of disposition effect though it focuses mainly on the impact of taxation. Ferris, Haugen, and Makhija (1988) provide strong empirical evidence to support disposition effect. They found that the trading volume of winners exceeds trading volume of losers in all the observation period. The historical trading volume is also very important to determine future trading volume in the specific price levels. Statman, Thorley, and Vorkink (2006) discover that both the individual stock turnover and market-wide turnover are positively correlated with lagged returns for months.
The essential reference for this thesis is given by Kaustia (2004) who conducts the test for the market-wide disposition effect using IPOs. According to the disposition effect, investors have tendency to sell winners early, which means they may trade more after the stock price cross their reference points from below. He builds up winners‟ portfolio and losers‟ portfolio and measure the development of trading volume after 20 trading days of IPO. He observes that the trading volume increases significantly when the stock price exceeds the offer price for the first time. If investors set the maximum price of the months as reference points, the exceeding
9
of those reference points can also lead to trading volume increase. With this assumption, he finds the monthly high price levels are significant reference points. The monthly low price levels are also positively significant in his research. He discovers that the reaching of monthly maximum price produces stronger effects than the reaching of monthly minimum price. He summarizes that the disposition effect exerts significant influence on the aggregate market behavior in the United States.
The most important issue I address in this thesis is the determination of the common reference points. As I summarized above, the previous price and the purchase price can be regarded as important reference points (Shefrin and Statman, 1985; Weber and Camerer, 1998). Monthly high and low prices are found to be relevant for the trading behavior (Grinblatt and Keloharju, 2001). As an important category of reference points, the monthly low price levels are also included in my thesis. According to the disposition effect, investors have tendency to sell winners early and ride losers for long. When the stock price has the tendency to drop below the reference points, the investors are urged to realize their gains in the fear of loss. However, when the stock price is below the reference points, the investors are predicted to hold their stocks. Because the investors are predicted to sell before loss in a declining market, it is another kind of tendency of selling winners early. So the positive coefficient of reaching the monthly minimum price levels in Kaustia (2004) can be explained by the disposition effect, which suggests the tendency of selling exceeds the tendency of holding in the same day with his observations. This can also explain why the reaching of monthly maximum price produces stronger effects than the reaching of monthly minimum price. To be more specific, Grinblatt and Keloharju (2001) discover that the monthly low price levels are significant for household investors, while insignificant for financial institutions and corporations. There is also a mixture of momentum and contrarian behavior in this price level.
Finally, whether the disposition effect has the same market-wide impact for different countries may potentially be influenced by the investor nature. For example, Yates, Lee, and Shinotsuka (1996) discover that Asian people exhibit more overconfidence than Western people (see also Yates, Lee, and Bush, 1997), except for the Japanese and Singaporeans.
In this thesis, I focus on the research question whether the Chinese investors display
10
disposition effect in three reference points which also has the similar market-wide influence as observed in developed countries.
Based on the discussions above, I can form two major hypotheses about the reference points for this study:
Hypothesis one: Trading volume will increase when the stocks with negative initial returns cross the offer price from below for the first time.
Hypothesis two: Trading volume will increase when the stocks reach new high and low price levels during the past month.
Data
I collected 992 IPOs taking place in Chinese stock market between June 2006 and December 2011. The IPO samples are obtained from two sources, including China securities registration and settlement statistical yearbooks and Dazhihui's stock terminal2. Table I provides the summary statistics of IPO firms.
Table I
Summary of Initial public offering China A-share, 2006-2011
Year 2006 2007 2008 2009 2010 2011
Number of companies 65 126 77 99 346 279
Total Value of Stock3 484.1 345.5 53.9 86.3 396.8 70.6
The IPO numbers are unbalanced because of two moratoriums from 2005. From the middle of 2005 to the middle of 2006, the regulator imposed a suspension on new IPOs in order to convert some non-tradable shares into tradable shares. From the second half of 2008 to the beginning of 2009, IPO activities suspended again, which was primarily due to low credit and liquidity availability in the market during the financial crisis. The country's Second Board market came into existence in the end of 2009 after a long wait.
I examine each firm carefully. The stock exchanges can impose special treatment to a firm
2 Software of Shanghai Dazhihui Software Develop Co., Ltd, listed company in Shanghai stock exchange
3 Billion RMB
11
with abnormal financial conditions, which usually means abnormal financial behavior or consecutive losses. After eliminating 9 cases which have been under special treatment or have insufficient trading days, the sample size reduces to 983.
For each firm, data of 253 trading days from the first day of IPO is collected. The data includes individual stock return, individual stock turnover, market turnover and market return.
I describe them below.
1. Individual stock return. The total return since IPO has been employed in this paper. It is reasonable to consider that the investors take dividends and stock splits into account. I fix the first day closing price as the benchmark and adjust dividends and stock splits backwards. For most of the firms they don‟t provide dividends in their first year of listing.
In the meantime, stock splits are also uncommon during the first year of trading. The return index is close to the price index in this study. Using the closing price is to some extent biased because the stock price can possibly cross the certain level and go back intraday. Kaustia (2004) argued that in most cases the results are the same with closing price method and intraday method. I define crossing as the closing price exceed a certain price level in this study. The observation period selected from the 253 trading days will be specified in the section of Methodology.
2. Individual stock turnover. A number of researches use individual turnover to measure volume (e.g. Richardson, Sefcik and Thompson, 1986; Stickel and Verrechia, 1994).
Individual stock turnover rate has been used to describe the trading volume in this study for each stock. By definition, turnover rate is calculated by the formula: Turnover rate = Total number of shares traded /Number of shares outstanding. I get the turnover rate data directly from the Dazhihui's stock terminal.
3. Market turnover and Market return. Lo and Wang (2000) summarize the literature on trading activities and several methods are elaborated. Campbell, Grossman and Wang (1993) use aggregate turnover to measure market volume. They argue that because the number of shares traded and the number of shares outstanding grow together, the use of turnover is more proper though it does not eliminate entirely the low-frequency variation.
The historical market return index can be obtained directly from the stock terminal. The total number of shares traded is obtained by the stock terminal and the aggregate number
12
of shares outstanding is acquired from the China securities registration and settlement statistical monthly books.
The summary statistics of the 983 IPOs are presented in table 1. The firms which have negative initial returns are considered as „Losers‟ while the firms with positive initial returns are labeled as „Winners‟. There is one firm that has zero initial return and is labeled as
„Neutral‟. The first day turnover of the winners is particularly high, which means that on average about 75.98% of the initial investors who share the same cost are left. This characteristic is undesirable for this study, because the late-comers possess various reference prices and they are accounted for a large proportion. Therefore, I focus on the losers which have mean initial turnover rate of 29.70%. The losers‟ subsample is formulated based on the first day return and the ex-post performance. The first time crossing offer price is taken into consideration. To be more specific, the losers whose stock price crosses the offer price between day 2 and day 253 from below are included in the losers‟ subsample. There are 77 companies getting the nod.
Table II
Summary statistics of IPOs for all firms, winners, losers, and losers’ subsample Offer year Initial Return Offer price Turnover All firms, N=983
Mean 2009.4 66.60% 23.12 71.12%
Median 2010 40.80% 20 76.02%
St. deviation 1.57 80.60% 15.45 18.38%
Winners, N=880
Mean 2009.2 75.14% 22.1 75.98%
Median 2010 47.12% 19.6 77.72%
St. deviation 1.58 81.01% 15.03 11.97%
Losers, N=102
Mean 2010.7 -6.61% 32.09 29.70%
Median 2011 -5.61% 29 28.22%
St. deviation 0.44 4.15% 16.06 8.80%
Neutrals, N=1
Mean 2006 0 2.8 22.12%
Losers‟ Subsample, negative initial return, exceed the offer price from below between days 2 and 253, N=77
Mean 2010.6 -5.80% 30.31 30.30%
Median 2011 -5.35% 28 28.50%
St. deviation 0.7 3.39% 15.5 8.85%
13
I pay attention to the mean offer year of those samples. The mean offer year of the winners‟
portfolio is 2009.2 while the mean offer year of the losers‟ portfolio is 2010.6. It is until recent years that the IPOs in this country begin to have more negative initial returns. I have 26 losers in the year 2010 and 76 losers in the year of 2011. From the year 2000 till the year of 2009, there were only 6 firms having negative initial returns in this market.
The observation period, which is from June 2006 to December 2011, witness the peak of the aggregate market in the second half of 2007. After that the investors bore a bear market for a long time until now. Most of the IPOs investigated in this study were taking place in the falling market. The summary statistics of the first day and one year market-adjusted returns are presented below.
Table III
Descriptive statistics of IPO returns of ‘Winners’ and ‘Losers’
Market adjusted; W-„Winners‟; L-„Winners‟; 2006-2011
Variables Initial return (W) Initial return (L) 1 year return (W) 1 year return (L)
Mean 0.75 -0.07 0.47 -0.10
Variance 0.65 0.00 0.78 0.07
Loughran and Ritter (1995) report that initial public offerings have poor long-run investment performance. I also observe this in my samples. The majority of the IPO firms‟ one year performances are worse than the initial returns.
Methodology
As introduced before, I define the „Winners‟ as the firms with positive initial returns while the firms with negative initial returns are labeled as „Losers‟. The observation period is divided into two parts. I compare the „Winners‟ and the „Losers‟ using the first day trading volume.
The 2nd day to the 253th day trading volume is employed to determine reference points.
There are several reasons why I separate the observation period. As the Chinese exchanges impose a dailyup and down price boundary of ten percent on individual shares after the first trading day, the first day is without price limits and usually has different characteristics which may be hard to explain with my explanatory variables. In the U.S. stock markets, the
14
underwriter can use several options to stabilize the price in the case of IPO. Aggarwal (2000) describes underwriters‟ support activities in the United States. This is not the case for China as the stabilize bids and options are still under trial use. The underwriters‟ influence is small after the listing. Sohn et al. (2012) review the relationship between underpricing and reputation of underwriter, arguing there is insufficient evidence to support the assumption. Su and Bangassa (2011) discover inadequate influence of underwriter reputation on the degree of underpricing in China. On the other hand, they find a positive correlation between underwriter reputation and the long term stock performance. As I mentioned in the Literature Review, many of the A-share IPOs are substantially undervalued at the offer, while the investors overvalue them on the listing day (Sohn et at., 2012). To summarize, I consider the first day is the only special day after IPOs. It is better to study the first day characteristics separately.
When I compare the first day trading volume, the Student‟s t-test is employed to see if there is significant difference between the trading volume of the „Winners‟ and „Losers‟. The F-test of equality of variances will be conducted first to test the population variance.
The hypotheses are H0: 𝜎12⁄𝜎22= 1; H1: 𝜎12⁄𝜎22≠ 1
The test statistics is 𝑠12/𝑠22, and the rejection region is F > 𝐹𝛼
2,𝑉1,𝑉2 or F < 𝐹1−𝛼
2,𝑉1,𝑉2
If the variances are equal: t = [(𝑥1− 𝑥2) − (𝜇1− 𝜇2)]/√𝑠𝑝2(1 𝑛⁄ 1+ 1 𝑛⁄ ) 2 If the variances are unequal: t = [(𝑥1− 𝑥2) − (𝜇1− 𝜇2)]/√(𝑠12⁄𝑛1+ 𝑠22⁄ ) 𝑛2
Where 𝜇1− 𝜇2 is the difference between the population means, 𝑥1− 𝑥2 is the difference between sample means, 𝑠𝑝2 is the pooled variance estimator, v is the degree of freedom, 𝑛 represents sample size, and 𝑠2 indicates sample variance.
If the trading volume of „Winners‟ is significantly higher, it can give evidence to support the existence of the disposition effect.
The most important objective of this research is to find common reference points of investors.
When I try to investigate the trading volume after the first day, the two step regression method
15
introduced by Kaustia (2004) is employed. As the first step regression attempts to explain the daily trading volume for the stocks with several explanatory variables, the residuals contain the properties which are not affected by those potential confounding variables. The method uses the residuals as the new dependent variable in the second step to neutralize the influence of the confounders in the first step. The second step model runs entering different price levels which help me to determine the reference points. So in my analysis, the second step is more important with the residuals from the first step. Because the trading volume is especially high for the „Winners‟, I use only the stocks which have negative initial returns and cross the offer price levels within one year to study the influence of offer price. This specific subsample is called losers‟ subsample as described in the section of Data. I use the entire sample to investigate the influence of monthly new record high and low price levels.
Then I come to explain the model of the first step regression. The first step regressions estimate the daily turnover for one year period from the second day of IPO. The regression model can be specified as:
Daily Turnoverit = Constant + Market turnoverit*X1 + Volume (-1)it*X2 + Max [R, 0]it*X3 – MIN [R, 0]it*X4 + Return (-1)it*X5 + Volatilityit*X6 + Timeit*X7 + Time squaredit*X8 + eit
I explain the variables below:
1. Logarithm of daily turnover and market turnover. I employ the turnover rate to measure the trading volume. The trading volume data of the aggregate market are acquired from the stock terminal directly. Using the method of Smith Bamber, Barron and Stober (1999), logarithm has been applied to the turnover rates to eliminate skewness.
2. Logarithm of lagged stock turnover. Volume (-1) means the logarithm stock turnover rate lagged one trading day. Gallant, Rossi and Tauchen (1992) find that daily trading volume displays serial correlation. Since the study starts in the second day, I include the one day lag return in this thesis.
3. Logarithm of return. „Max [R, 0]‟ represents the return if positive while „–MIN [R, 0]‟
shows the absolute value of return if negative. For the relationship between contemporary relation between return and trading volume, Karpoff (1987) summarizes previous findings that most of the literatures suggest positive relationship between price and volume. The variables „Max [R, 0]‟ and „–MIN [R, 0]‟ enables us to separate the positive effects and
16
negative effects of returns.
4. Logarithm of lagged return. Return (-1) means the stock return lagged one trading day.
Recent studies focus not only on the contemporary relation between return and trading volume, but also causality between those two terms. Rogalski (1978) provide empirical evidence to support the idea that volume and price change per se are positively correlated.
Since the study starts in the second day, I include the one day lag return in this thesis.
5. Volatility. In this study, the volatility is calculated by daily stock return squared.
Lamoureux and Lastrapes (1991) discover positive volume and volatility correlation with GARCH-type volatility features. Gallant, Rossi and Tauchen (1992) find the daily volume and volatility are positively correlated.
6. Time and time squared. Time is calculated by the month relative to the IPO data. The presence of seasonality has been investigated in a lot of stock markets. Boudreaux (1995) studies seasonality in seven countries and discover three of them confirm a monthly effect.
Recent research by Eleftherios (2009) includes 55 Stock market indices of 51 countries, which rejected all the month-of-the-year effects worldwide. Kaustia (2004) include both variables to detrend the turnover series.
The residuals of the first step regression are employed as the dependent variable of the second step regression. The regression model can be specified as:
Residualsit = Constant + New record high dummyit*X1 + New record low dummyit*X2 + First time crossing dummiesit*X3 + All times crossing dummiesit*X4 + Price range dummiesit*X5 + ejt
There are four sets of dummy variables in the second step regression. I describe each of them below.
1. New record high and low level dummies. As discussed in the Literature Review, the reaching of record high or low points might be important. I have dummies for the new record high and new record low for the previous month. The dummies are calculated as follows: for the first month (21 trading days) from the second day after IPO, the highest point of return index is set two 1. After the first month, the dummy variable is set to 1 if the return index exceeds the highest points during the previous 21 days. All the other days
17
are set to 0. The new record low dummy is calculated using the same methodology for the lowest points.
2. First time crossing dummies. As I mentioned in the section of Literature Review, it is reasonable to consider the purchase price as one of the important reference points for the investors. In the event of IPOs, the offer price can be considered as a common purchase price for the initial investors. Turnover of the securities should increase after IPO when stock price reaching the reference point from below for the first time, because investors are unwilling to realize losses when the stock price below their benchmarks. I set levels from 95% to 150% with respect to the offer price with 5% intervals as most of the firms have higher initial return than 90%. If the stock return index crosses those levels, e.g., 95%, 100%, 105%, from below for the 1st time, I set the dummy variable as 1; otherwise the dummy variable presents 0. There is exactly one event of crossing for every company in the losers‟ subsample for the level of 100%, and this level is the most important one in the experimental design because the disposition effect is supposed to be at the strongest.
Investors may require a „break even‟ return which includes transaction costs and return requirement against fixed income instruments. The argument gives rise to the potential importance of 105% or even higher price level. That‟s why so many price levels are included in this dummy group.
3. All times crossing dummies. The all times crossing dummies make up two subgroups.
The first subgroup has the same levels with the first time crossing dummies, while ignoring the first time crossing information. I set the dummy variable to 1 every time the return index crosses the specific level from below, after the first time. Otherwise the dummy variable displays 0. The second subgroup contains levels from 70% to 90% with an interval of 5%. Those levels are not included in the first time crossing dummies and contain also the first time crossing information. The purpose of setting those levels is to broaden the range. I set the dummy variable to 1 every time the return index crosses the specific level from below, otherwise 0. The reason why I set the all times crossing dummies is that 1)I want to highlight the importance of the first time crossing, 2)I want to see if the price levels matter during the entire observation period.
4. Price range dummies. To be more specific, those are the return index range dummies.
18
According to the information given in the Literature Review, the price ranges may also be relevant for this study. If the stock price falls into a specific range, I set 1 for the dummy variable. For example, the return index falls into the range [105%, 110%] means that it is equal or bigger than the price level of 105% relative to the offer price, but is smaller than the price level of 110%. In order to avoid the dummy variable trap, the benchmark range [100%, 105%] is excluded.
Though the Chinese stock markets have high turnover rates, and IPO performances are often inspected by initial investors, I cannot assume that the investors can react to the crossing event timely. One additional test has been designed with lagged response from one week to four weeks. The test is only for the lagged response of the first time crossing events and the monthly record high or low levels. For the second hypothesis, I only arrange the monthly high and low dummies in the second step. The experimental design is the same with the test of hypothesis one.
Results
In this section, I first present the first day results for all the firms. Then I employ the losers‟
subsample to investigate the first hypothesis to see if investors display disposition effects with the reference point around the offer price. Finally I use all the samples from the second day to test the second hypothesis to see if the investors regard the reaching of monthly record high or low stock price as important reference points.
The first day of IPO is always under the limelight because the positive first day price spike often occurs. It is well-documented that majority of the IPOs earn positive initial returns, not only in this emerging market. One of the earliest ones comes from Ibbotson (1975). In order to investigate the first day trading volume, firms are categorized into two portfolios, which are „Winners‟ and „Losers‟. In this situation, the reference price is the offer price. The first day turnover rates of two portfolios are compared, which determines whether investors want to realize their gains early when the price increases. I reject the null hypothesis of equal variance using the F-test of equality of variances. The unequal variance t-test is employed. Table IV
19
presents T-test results for the first day turnover rate of winners‟ and losers‟ portfolios.
Table IV
T-test results for the first day turnover rate of winners’ and losers’ portfolios Equal variance not assumed; 2006-2011
Winners Losers
Group statistics
Mean 0.760 0.297
Variance 0.014 0.008
No. of observations 880 102
T-test for equality of means
Mean difference 0.463
t-statistics 47.985***
P-value 0.000
***significant at 1% level
The mean turnover rate of the winners‟ portfolio is 0.760 while the losers‟ portfolio achieves a mean turnover rate of 0.297. The t-value is 47.985 and is significant at 1% level. The results clearly show that the first day turnover rate of the winners‟ portfolio is significantly higher, indicating evident disposition effect among investors. The result is in line with Reese Jr.
(1998) as he indicates that positive return IPOs experience significantly higher trading volume than losers. The first day turnover and the initial return are plotted into the bar chart below in order to have a more intuitive feeling of the relationship.
Figure I
First day turnover rate and initial return
Bars represent the mean turnover rate of first day trading using first day return classification
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
Turnover rate
First day Return range
20
All the firms are included in the figure which is grouped with 10% intervals. The first interval of 70% to 80% contains only 3 firms and there is no firm has less than 70% first day return.
There are 55 firms in the last category. The turnover rate for the winners‟ portfolio is very high.
I then begin the two step regressions. I employ the losers‟ subsample to examine the first hypothesis to see if the trading volume increases when the loser stocks cross the offer price from below for the first time. To save the residuals for the second step regression which will be run with variables of interest, 8 potentially important but undesired variables are regressed in the first step as explained in Methodology. Panel data technique is employed with data containing both cross-sectional and time series elements. Redundant fixed effects test and correlated random effects test are done beforehand to determine the model. The tests confirm that cross-section fixed effects are necessary while the random effect model is not appropriate.
Cross-sectional fixed effects model is applied which controls for the average differences across stocks. Table V presents the results for the first step regression.
Table V
Results of the daily turnover regression for the loser subsample (first step) Cross-sectional fixed effects panel data;
The residuals of the first step are employed as the dependent variable for the second step;
2006-2011
Periods included: 252 Cross-sections included: 77
Total panel (balanced) observations: 19404
Variable Coefficient Std. Error t-Statistic Prob.
Constant -0.456 0.045 -10.204*** 0.000
Market Volume 0.140 0.010 14.261*** 0.000
Volume(-1) 0.714 0.004 180.409*** 0.000
Time -0.033 0.004 -9.334*** 0.000
Time Squared 0.001 0.000 3.709*** 0.000
MAX[R,0] 20.445 0.410 49.822*** 0.000
-MIN[R,0] 11.434 0.411 27.795*** 0.000
Volatility -41.941 5.244 -7.999*** 0.000
Return(-1) 2.511 0.106 23.779*** 0.000
Adjusted R-squared 0.827
***significant at 1% level
21
All the independent variables are significant at 1% in this regression. I observe negative coefficients for the constant, „Time‟ and „Volatility‟, while the other variables have positive coefficients. The adjusted coefficient of determination for the first step regression is 82.7%. In this study, it indicates the first step regression approximates the stock daily trading volume well. The Durbin-Watson value is 2.257 which can be interpreted as no serial correlation, and the residual plots are desirable. The residuals of the first step regression are employed for the second step regression. In order to make sure that the standard error estimations in the test are robust to heteroskedasticity, White cross-section standard errors & covariance is selected to provide a more reliable result. The results are presented in table VI.
Table VI
Result of the residual regression for the loser subsample (second step) Dependent variable: residuals of the first step regression.
Independent variables:
1. Dummy group one: the highest and the lowest price of the last month;
2. Dummy group two: the first time crossing certain price level from below;
3. Dummy group three: the entire times crossing certain price level from below, some begin with the second time cross to distinguish from the second dummy group;
4. Dummy group four: the price in a particular range with regard to the offer price.
White cross-section standard errors; 2006-2011 Periods included: 252
Cross-sections included: 77
Total panel (balanced) observations: 19404
Variable Coefficient Std. Error t-Statistic Prob.
One Month High -0.012 0.009 -1.362 0.173
One Month Low -0.081 0.009 -8.616*** 0.000
1st Cross 95% -0.028 0.067 -0.425 0.671
1st Cross 100% 0.257 0.084 3.060*** 0.002
1st Cross 105% -0.024 0.055 -0.435 0.664
1st Cross 110% -0.150 0.048 -3.139*** 0.002
1st Cross 115% -0.116 0.051 -2.295** 0.022
1st Cross 120% -0.249 0.058 -4.267*** 0.000
1st Cross 125% -0.218 0.062 -3.497*** 0.001
1st Cross 130% -0.227 0.079 -2.864*** 0.004
1st Cross 135% -0.119 0.068 -1.749* 0.080
1st Cross 140% -0.210 0.084 -2.508** 0.012
1st Cross 145% -0.138 0.090 -1.532 0.126
1st Cross 150% -0.553 0.215 -2.568** 0.010
22
Cross 70% (From the 1st) 0.073 0.043 1.685* 0.092
Cross 75% (From the 1st) 0.089 0.029 3.061*** 0.002
Cross 80% (From the 1st) 0.025 0.028 0.892 0.373
Cross 85% (From the 1st) 0.040 0.026 1.574 0.116
Cross 90% (From the 1st) 0.026 0.021 1.264 0.206
Cross 95% (From the 2nd) 0.011 0.024 0.483 0.629
Cross 100% (From the 2nd) 0.033 0.030 1.100 0.271
Cross 105% (From the 2nd) -0.073 0.031 -2.344** 0.019 Cross 110% (From the 2nd) -0.054 0.026 -2.045** 0.041
Cross 115% (From the 2nd) -0.062 0.034 -1.837* 0.066
Cross 120% (From the 2nd) -0.085 0.033 -2.564** 0.010
Cross 125% (From the 2nd) -0.046 0.051 -0.913 0.361
Cross 130% (From the 2nd) -0.020 0.052 -0.384 0.701
Cross 135% (From the 2nd) -0.151 0.055 -2.737*** 0.006
Cross 140% (From the 2nd) -0.116 0.075 -1.551 0.121
Cross 145% (From the 2nd) 0.049 0.069 0.710 0.478
Cross 150% (From the 2nd) -0.066 0.093 -0.713 0.476
RANGE [40%, 50%] -0.141 0.050 -2.845*** 0.004
RANGE [50%, 60%] -0.054 0.022 -2.392** 0.017
RANGE [60%, 70%] -0.085 0.014 -5.871*** 0.000
RANGE [70%, 75%] -0.110 0.019 -5.709*** 0.000
RANGE [75%, 80%] -0.102 0.016 -6.360*** 0.000
RANGE [80%, 85%] -0.096 0.015 -6.353*** 0.000
RANGE [85%, 90%] -0.093 0.015 -6.333*** 0.000
RANGE [90%, 95%] -0.071 0.015 -4.838*** 0.000
RANGE [95%, 100%] -0.020 0.014 -1.427 0.154
RANGE [105%, 110%] -0.008 0.016 -0.509 0.611
RANGE [110%, 115%] -0.016 0.016 -0.966 0.334
RANGE [115%, 120%] -0.031 0.018 -1.715* 0.086
RANGE [120%, 125%] -0.025 0.021 -1.194 0.233
RANGE [125%, 130%] -0.048 0.025 -1.917* 0.055
RANGE [130%, 140%] 0.011 0.024 0.452 0.651
RANGE [140%, 150%] -0.076 0.036 -2.085** 0.037
RANGE [150%, 160%] -0.032 0.041 -0.782 0.434
RANGE [160%, 180%] -0.085 0.029 -2.939*** 0.003
RANGE [180%, 200%] -0.054 0.032 -1.698* 0.090
RANGE>200% -0.061 0.044 -1.383 0.167
Constant 0.071 0.012 6.053 0.000
Adjusted R-squared 0.021
∗∗∗, ∗∗, and ∗ indicate significance at the 1%, 5%, and 10% levels, respectively.
