HERDING BEHAVIOUR IN EXTREME MARKET CONDITIONS: THE CASE OF THE CHINESE STOCK MARKET Yue Fu Faculty of Economics and Business Rijksuniversiteit University Supervisor: Marc Kramer

(1)

HERDING BEHAVIOUR IN EXTREME MARKET

CONDITIONS: THE CASE OF THE CHINESE STOCK

MARKET

Yue Fu

Faculty of Economics and Business Rijksuniversiteit University

(2)

Abstract

This paper examines the existence of herding behavior in extreme market conditions using daily data from the Chinese stock market. We test stock return dispersions for the existence of herding as suggested by Christie and Huang (1995) and Chang, et. al (2000). Consistent with prior studies, we find the evidence of herding in extreme market condition within Shanghai A-share markets that is dominated by domestic individual investors for the year 2005-2008. Herding occurs in both bull and bear markets. However, comparing the results for the upside and downside moves of the market, we see that return dispersions during extreme downside moves of the market are lower than those for upside moves, indicating that herding is stronger during periods of a decreasing market.

Beyond pointing out the existence of herding behavior, the factors influencing herding behavior are discussed. Our results indicate that market stress and announcing new regulation are important influencing factors. Herding behavior by investors is more pronounced under conditions of extreme low return of market and government announcing new regulation may enhances the herding behavior in the Chinese stock market.

(3)

Understanding the decision making process of investors has always been a major challenge to academics as well as practitioners. Traditional financial theory has suggested that financial markets are "informational efficient” and investors are rational. The former hypothesis indicates that the prices of traded underlying assets already reflect all known information and the "rational investor" hypothesis proposes that all investors have the ability to make rational judgments towards investment based on all the information from an efficient market. However there are many abnormal phenomena that appear in the financial markets that are very difficult to understand, and experts in behavioral finance try to explore these complex and provocative puzzles. In the research of Odean (1998), it was noted that traditional financial or economic theory focuses on the aspect of forecasting, and ignores the hypothesis which reflects the emotions or moods of the investors.

Analyzing two extreme psychological states of mind of investors can explain the reason that the actions of investors deflect a long way from the fundamental analysis of companies. The first reason is overconfidence, which means the investors believe that the precision of their knowledge about the value of a security is greater than it actually is; therefore they tend to trade more than rational investors. The other reason is a lack of confidence and many scholars also call this “herding behavior”. In this case the investors will tend to suppress the impact of their own private information and imitate the actions of others. Because of this herding behavior, much information related to securities may not be reflected in the current market price and this may cause an “overreaction” to a single resource.

(5)

speculation would hurt the health of economics. For example, the bubble of high-tech stock occurred in 2000 and the real estate crisis in Japan followed World War II. Herd behavior seems to be a plausible explanation for the misalignment of prices and fundamentals. It also distorts the market mechanism.

Bikhchandani and Sharma (2000) define herd behavior as when “investors and fund managers are portrayed as herds that charge into risky ventures without adequate information and appreciation of the risk-reward trade-offs and, at the first sign of trouble, flee to safer havens’. Several views have been expressed concerning why investors would tend to copy the actions of other investors. Firstly, individuals may have an intrinsic preference for conformity (Devenow and Welch, 1996). Secondly there is a view that suggests that others may know some information about the returns on a particular investment and their actions mirror or echo this information (Chari and Kehoe, 1999). Finally there is the principal–agent relationship, when money managers who invest for customers might imitate others as the results of incentives provided by reward, a compensation scheme, or their reputation. Bikhchandani and Sharma (2000) provide a comprehensive summary of this field of studies.

As for the whole stock market, Christie and Huang (1995) have concluded that the finding among empirical studies at that time was that there is no evidence to show herd behavior between the single stock and market return in the American stock market using the CSSD (cross-sectional standard deviation of returns) method. Chang et al, (2000) further reported the consistent result that the developed stock market does not display the phenomena of herd behavior; however in the emerging or developing markets it is more apparent.

(6)

In this research, two questions will be discussed. The first is whether the phenomenon of herd behavior exists in the stock market under extreme conditions (bull and bear market). The existence of this behavior will be examined in the Chinese stock market during a particular period. The Chinese stock market has been riding on a "roller coaster" from June 2005 to October 2008. It started to increase from the 1020.63 point, and advanced about 400 percent over the period 2005 -2007 and then plunged by more than 70 percent in 2008. The second question that will be considered in this research is what are the factors that influence herding behavior? The factors including market stress, depth of the market, volatility, daily returns, regulation announcements, and change of volume will be considered here.

(7)

II. Literature Review

Many aspects of herding behavior have been researched from different angles, including the theory of this concept, the research models and empirical research. This section briefly summarizes these previous studies.

II - i The Correlation Theory of Herding Behavior

Research regarding herding behavior was first mentioned in Keynes’ s book “The General Theory of Employment, Interest and Money”(Keynes 1936). He believed that judgments and decisions may not be made based on information but instead these decision-making processes may be driven by mass psychology. Since then scholars have published many articles to discuss this phenomenon over the past 70 years. However, the early published research that mainly explains this behavior focuses on the field of economics and psychology and further defines the concept of herding behaviors based on these fields. Banerjee (1992) thought that the behavior of following others and then making the same decision could be regarded as herding. This situation has occurred in many situations such as evaluating new technology, polls, and voting, etc.

As late as the 1990s, this behavior was considered in the financial field when market participants recognized the importance of herding which affected their investment strategies. Thus, scholars started to pay close attention to the financial field. Herd behavior in financial markets is a very special irrational phenomenon. Institutions and individual investors tend to mimic the decisions of others rather than follow their own beliefs and information. In 1999, Nofsinger and Sias found that individual investors rush to trade stocks in the same position (long or short) at the same time like a swarm of bees. Bikhchandani and Sharma (2000) define herd behavior as an obvious purpose by investors to mimic the behavior of other investors.

(8)

something that they don’t, so they still follow the herd and give up the rational analysis. Pursuing a trending market is a good example (Bikhchandani, Hirshleifer, and Welch, 1992). During the process of rational decision making, it is still possible to cause herding. According to the categories by Devenow (1996), there are three potential reasons for rational herding behavior, which are direct payoff externalities, principal-agent problems, and informational learning (cascades).

Direct payoff externalities

Payoff externalities may drive the decisions made by investors for the stocks that they acquire information, if a minimum number of investors have acquired private information and earn a profit; other investors will soon discover the same signal and mimic the activities of the first investors (Admati and Pfleiderer, 1998). This theory shows that the payoff for investors adopting an action increases the number of other investors adopting the same action. This is because share prices adjust to new publicly available information although not very rapidly in the depth market (high liquidity of the market); investors will therefore prefer to enter into market following other informed investors until the stock prices reflect all the information available. This herding problem becomes more serious when investors follow rumors or just imitate their neighbors’ simple behavior (Shiller, 1984).

Reputation and principal- agency problem

(9)

beaten path of conventionality even following group failure.

Informational externalities: Cascades

An informational cascade occurs when people observe the actions of others and then make the same choices that the others have made independently of their own private information signals. This concept was introduced by Bikhchandani et al. (1992) and Welch (1992). The theory demonstrates that “agents gain useful information from observing previous agents' decisions, to the point where they optimally and rationally completely ignore their own private information” (Devenow and Welch, 1996). When more and more people replicate this process, these behaviors are seen as flowing steadily downwards (like a waterfall) before finally causing the phenomena of herding. But this cascade can be reversed if only a small amount of information has been made public. Herding from cascades is mainly caused by individual investors because the disadvantage of an inherent lack of information encourages individual investors to speculate on the trades of agents that have inside information.

According to the above discussion, we can conclude that the herding from payoff externalities may occur in a group of institutional legal individuals who have the excellent ability to acquire information. Reputation and principal-agency problems are always reasons for fund managers’ herding. In order to pursuit their relatively high performance, they prefer to fall behind others than earn high returns with additional risk. And individual investors’ actions tend to cause herding from information cascades because of the absence of information. The next subsection will discuss models that measure herding behavior.

II - ii Models that Measure Herding Behavior

(10)

expectation. If the return of this financial commodity is similar to the return of market or the portfolios, we can deduce that the herding behavior of investors exists in the market. The cross-sectional standard deviation (CSSD) model and cross-sectional absolute deviation (CSAD) models are good examples. These methods are described below:

LSV model

LSV method was introduced by Lakonishok, Shleifer, and Vishny (1992). The authors based their criterion on the trades conducted by a group of fund managers, comparing the actual behaviors with an ideal behavior to test the existence of herding. However, this method has mainly been used to research the herding of institutions and therefore in this research this method is not given lengthy attention.

CSSD and CSAD models

These methods use the return dispersion to test herd behavior. Evidence of Herding would be seen with a lower CSSD and a lower or a less than proportional increase in the CSAD during periods of extreme market movements. The CSSD method was first used by Christie and Huang (1995). Following this Chang, Cheng, and Khorana (2000) developed the CSAD model based on a similar concept. The detailed explanations of models are discussed later.

Because the CSSD and CSAD have similar meanings, the research literature generally uses both of them. Gleason (2003 and 2004) has measured the existence of herding behavior in the ETF and the commodity futures in the US using both the CSSD and CSAD models. Hwang (2004) and Henker (2004) also use the same approaches to measure herding behavior.

II - iii Empirical Research of Herding Behavior

Measurement of herding behavior

(11)

and the futures markets, even expanding into the field of real estate. The second mainly focuses on the classification of traders, where these maybe classified as dealers, institutional investors and individual investors, etc. This research mainly tests whether herding behavior exists in the Chinese stock market, and so here we primarily review the literature of herding behavior in financial commodity markets.

The setting for herding behavior in the financial commodity market is always the stock market. Primary research has mainly paid close attention to the developed countries such as the US stock market. Christie and Huang (1995) found that there is no evidence to support herding behavior between simple stocks and market returns using the NYSE and AMEX stocks as an example. Henker (2004) also obtained a similar result finding that there is no phenomenon of herding behavior in the Australian stock market. Contrary to the evidence found from the developed countries, some recent research has seen evidence of herding in emerging markets. Chang, Cheng, and Khorana (2000) studied five stock markets including US, Japan, Hong Kong, South Korea and Taiwan. Their results indicated that the US Japan, and Hong Kong stock markets do not demonstrate this phenomena, but South Korea and Taiwan show the existence of herding behavior, the greater the maturity of the market the less significant the herding behavior that was seen.

Compared with studies of financial commodity market, there are fewer studies focused on financial derivatives. Gleason, Lee, and Mathur (2003) studied the thirteen commodity futures contracts traded on three European exchanges including food, corn, livestock and meats, etc. Their study period was from 1986 to 1998. They used the commodity index from KR-CRB (Knight Ridder-Commodity Research Bureau) as the benchmark to calculate the returns of the market, but they did not find any supporting evidence for herding behavior. The evidence from Gleason and Mathur (2004) who undertook empirical research of ETFs trading on the American Stock Exchange, also suggested that market participants in ETFs markets make investment choices rationally.

(12)

institutional investors hold large positions and they can influence the market more significantly (Lakonishok et al.1992, Oehler1998, Wermers 1999, Fong, Gallagher, Gardner and Swan 2004, Voronlova and Bohl 2003). All of these researchers have concluded that herd behavior is demonstrated significantly in emerging or immature markets like Portland, Portugal and China; however in areas such as US, Australia, Germany, they could not find much evidence to support the existence of herd behavior.

Considering research on individual investors or traders, Nofsinger and Sias (1999) have studied private investors using the LSV indicator from 1977-1996; Iiara, kato and Tokunaga (2001) tested this behavior in Japanese individual investors and overseas investors. And Chen (2002) studied the case of Taiwan during the period of 1995-2000 .The consistent results from all of these studies shows that herding behavior among institutional investors is more significant than that among individual investors.

According to the review of the above literature, we can conclude that herding in the developed countries is much weaker than that in the developing countries or emerging markets, and the herding by institutional investors is more significant than that of individual investors.

Factors influencing herd behavior

The factors that influence herding behavior may be divided into micro factors and macro factors. Micro factors may include the scale of company, classification of the industry, market stress, market daily returns, the volatility of the price, the expiry date of the future, important announcements, and the depth of the market. In contrast, the macro factors reveal more diversification, such as the yield spread of government bonds in a different month, the yield rate of the index, the rate of the company; the economic environment and the status of the market (bull or bear market).

(13)

relies on the maturity of the market, the more maturity in the market such as US, Japan, the lower the effect of market pressure. Besides, the herd behavior will present asymmetrical phenomena when the market returns in extreme values, and this has considerable basis between the lowest and highest by observing the coefficient of the variables.

Hwang and Salmon (2004) gave detailed explanations of the factors affecting herd behavior in the UK and South Korea. In their research, the authors not only tested the effects of the bull and bear markets, but also examined the SMB ("small market capitalization” minus “big”) and HML ( “high book-to-price ratio” minus “low"), with the representation of company scale and the value as an indicator of influencing factors. Moreover, there are other macro factors, such as the treasury bill interest spread (3 month treasury bill rate, 12 month treasury bill rate, and US 30 year treasury bond rate) and the difference between Moody’s AAA and BAA rated corporate bonds for the default spread. The two former indicators SMB and HML represent the risk during this period, and the other factors shows the prosperity of economics and credit risk. As a result, they found that micro factors can enhance the herding behavior in bull and bear markets, however macro factors appear useless.

According to the views of the classification of traders, in general, institutional managers that demonstrate herding mainly invest their funds in the middle or small companies (Lakonishok et al., 1992; Oehler, 1998; Wermers, 1999; Sias, 2004) and the high-tech companies or some high growth stocks (Oehler, 1998, Fong, Gallahger, Gardner and Swan, 2004). This evidence can be verified in the US, German, and Australian markets. Sias (2004) explains this phenomenon in that the information of small companies can’t be obtained easily by the traders and institutional managers may suspect others (buyer/seller) of holding the implicit information to make the decisions. As a consequence, the herding behavior caused by informational cascades occurs. Another reason for herding is that high-tech and high growth stocks in general have high volatility and high risk. When the time is ripe for investing, some institution investors can earn extra profit. As a result, their actions in the high-tech or high growth stock market also attract others causing herding.

(14)

The Chinese stock market was founded in the early 1990s; two official stock markets (Shanghai and Shenzhen) in China have expanded dramatically and become one of the leading markets in Asia. As of January 2009, it has more than 1600 firms listed. Despite the anticipation by Morgan Stanley that the Chinese stock market will grow rapidly, it may not be characterized by the depth and maturity of stock exchanges observed in developed countries, especially due to the poor legal structure and the weak rule of law in China. Institutional investors account for a large percentage of investors; individual investors face only a few alternatives and have to tolerate government involvement, such as with regulation changes and central bank intervention. These investors tend to speculate on the stock market, causing dramatic market volatility.

A number of recent studies have examined the herding behavior over different periods. Chui and Kwok (1998) found herding movements in the B-shares traded by foreign investors. Fung et al. (2000) reported that there was no herding behavior during the period from 1990-2000. However, Song et al. (1998) found significant evidence of this in the stock market. More recently, the Chinese stock market has provided an interesting setting for the analysis of herding behavior, especially during the period from July 2005 to October 2008. Between July 2005 and October 2007 (at a historic high) the Chinese stock market (represented by the Shanghai Composite Stock Index) increased by over 400%. The stock market crash that followed resulted in a fall in the Shanghai A-share index to 1861.47, i.e. a 70.64% loss.

(15)

III. Research Methodology

This research applies the CSSD and CSAD methods to measure the existence of herding behavior in the Chinese stock market during the special period of sharp upward and downward movements (2005.7-2008.10). After proving the existence of herding behavior, the possible influential factors that affect this behavior will be examined by building a multivariate regression model.

III - i The Measurement of Herding Behavior

CSSD (Cross-sectional standard deviation of return)

Using CSSD to test herding behavior was first proposed by Christie and Huang in 1995. The way to measure the level of dispersion using the CSSD method can be defined as follows: 2 1 ( ) 1 N it mt i t R R CSSD N = -=

-å

(3.1)

Where Rit is the return of Stock I in the time t. Rmt is the return of market or portfolio at time t. N is the number of stocks in the market or portfolio. This measure can be regarded as a proxy for the individual security return dispersion around the market average.

(16)

L U

T L t U t t

CSSD = +a b D +b D +e (3.2)

Dtu a dummy variable which is 1 when the return of the market for time period t lies in

the extreme upper tail of stock returns distribution and 0 otherwise. DtL is also a

dummy variable which is 1 when the return of market for time period t lies in the extreme lower tail of stock returns distribution, and 0 otherwise. Although somewhat arbitrary, in the literature, an extreme market return is defined as one that lies in the lower or upper one percent tail of the return distribution.

These dummy variables are designed to measure the difference in the investing behavior between extreme and normal conditions. As herd formation indicates conformity with market consensus, the presence of negative and statistically significant b (for downward markets) and_L b (for upward markets) coefficients _U would indicate herding by the market participants.

CSAD (Cross-sectional absolute deviation of return)

However if b and_L b coefficients are positive, we cannot determine the existence _U of herding behavior, instead only explain that we cannot find evidence to support herding under extreme conditions in the market. However Chang, Cheng and Khorana (2000) developed the cross-sectional absolute deviation (CSAD) model, which is based on the general quadratic relationship between CSADt and Rmt. This method

stems from the capital asset pricing model (CAPM) model; utilizing the cross-sectional absolute deviation to measure the consistency of investors’ behavior. Supposing that there are N stocks in the market, the daily return of stock i is Rit in

trading day t and the portfolio of the market return is Rmt, the cross-sectional absolute

deviation of the market in trading day t is expressed as:

1 1 N T it mt t CSAD R R N = =

å

- (3.3)

Based on the CAPM model by Sharp, the expected return is equal to the risk free rate plus the system risk premium,

[

( ) ( )

t i f i t m f

E R =R +b E R -R ù_û (3.4) And then the formula can be converted to

(

)

[

( ) ( ) 1 ( )

t i t m i t m f

(17)

Because the return of the market portfolio is larger than the risk-free rateR_f , we can take the absolute value of equation 3.5:

(

)

[

( ) ( ) 1 ( )

t i t m i t m f

E R -E R = b - E R -R ù_û (3.6) Then we take the sum of the average of equation 3.6:

(

)

[

1 1 1 1 ( ) ( ) 1 ( ) N N t i t m i t m f i i E R E R E R R N = N = b ù - = - - _û

å

(3.7) Then

(

)

[

1 1 ( ) 1 ( ) N i t m f i E CSAD E R R N = b ù =

å

- - _û (3.8) First-order and second order derivatives are used to obtain:

1 ( ) 1 1 ( ) N t i i t m E CSAD E R N = b ¶ ₌ -¶

å

(3.9) 2 2 ( ) 0 ( ) t t m E CSAD E R ¶ ₌ ¶ (3.10) 2 1 2 T mt mt t CSAD = +a g R +g R +e (3.11)

The relationship between CSADt and Rmt is used to detect herd behavior. According to

the result of the second order derivative, in the presence of herding the relationship between CSADt and the average market return is non-linear. A significantly negative

γ2 coefficient implies the existence of herding behavior. If market participants are more likely to herd during periods of large price movements, then there should be a less than proportional increase (or decrease) in the CSAD measure. In the absence of herding, the relationship is linear and increasing, that is the dispersion increases proportionately with the increasing returns of the market.

Moreover, the relationship between CSAD and market returns may be asymmetric. This hypothesis can be tested using two different models, Equa.3.12 is for upward market (bull market) and Equa.3.13 is for downward market (bear market):

2 1 2 up up up up up T mt mt t CSAD = +a g R +g R +e (3.12) 2 1 2

down down down down down

T mt mt t

CSAD = +a g R +g R + (3.13) e

The absolute values Rupmtand down

mt

(18)

the return, not with its sign. This also makes a comparison between 2

up

g andgdown₂ possible.

III - ii Factors Affecting Herding Behavior

Regression is an attempt to explain movements in a variable by reference to the movements of one or more other variables (Brooks, 2002). Using this method the factors that affect herd behavior can be identified. In this study market stress, market depth, regulation announcement, the volatility of stock price, daily market return, and the change of volume are the independent variables considered here.

The results from the CSAD or CSSD models are used as the response variables, and then the factors described in the previous paragraph will be used as the explanatory variables in the regression function. When the coefficient of the independent variable is negative and statistically significant, we can judge that this factor has an impact on herd behavior, and vice versa. The multiple regression model is expressed as:

t 0 1 2 3 4 5 6 7 (CSSD ) ( ) ( ) L U T t t t t t mt mt t t C SAD D a D Dep D R TV a a a a a s a a e = + + + + + + + + (3.14) Where D L andD U

are dummy variables which is 1 for market returns that lie in the extreme 5% lower and upper tail of the distribution of market returns and zero otherwise; Dep_t is the depth of the market; D_tT is a dummy which is one for markets that the non-tradable shares has been trader by owners (the result of government announcing this regulation) and zero otherwise; s_mt is the standard deviation of daily return; R_mt is daily return of the weighted stock price index and TV_t is the daily change of trading volume.

Influencing factors: a description

Market stress

(19)

and set up the critical points in the left and right tails of the returns under 5%. We use two dummy variables which indicate the range of lowest and highest returns. D

L

= 1 if the market return on day t lies in the extreme 5% lower tail of the distribution of market returns (and zero otherwise), and D

U

= 1 if it lies in the extreme 5% upper tail of the same distribution (and zero otherwise).

Depth of market

Market depth is the size of an order needed to move the market a given amount. If the market is deep, a large order is needed to change the price and investors can enter into or exit stock markets easily. For example, if the market for a stock is "deep", there will be a sufficient volume of pending orders on both the bid and ask side, preventing large trading from significantly moving the price. Along with the depth of the market becoming deeper, the liquidity of the market improves. The trading volume changes as the price changes by one unit:

Trading volume( current-previous period) Price variability(current-previous period)

marketdepth=

Based on the theory of market microstructure, some market characteristics such as a better market depth attract new investors to enter the stock market. Therefore, there is a good chance that herding may occur in a deep market.

Regulation announcement

According to the rules regarding the reform of the shareholder structure of listed companies, some non-tradable shares may be traded by their owners in the open market. As approximately two-thirds of outstanding shares are not publicly tradable, we consider the hypothesis that this announcement of lifting the ban on non-tradable shares may influence the psychology of investors and further trigger fluctuations in the stock market. In this model, there is a dummy variable Dr =1 when the owners trade their “Large Non-Tradable Shares”, and Zero otherwise.

Volatility of stock price

(20)

Hwang and Salmon (2004) found that increasing volatility enhances the herding in US and South Korea stock markets.

Daily market return

Daily market return is found by calculating the daily return of the weighted stock price index as an independent variable. Many prior studies have already confirmed the positive relationship between herding behavior and positive feedback trading of investors (De Long et al, 1990; Nofsinger and Sias 1999), which means these investors are basically trend chasers who buy securities when their prices have increased and sell stocks when their prices have decreased. So we can deduce that it is likely that daily market returns can cause herding behavior to occur.

Change of volume

(21)

1 ln( i ) i i S u S -=

IV. Data and Descriptive Statistics

IV - i Data Collecting

Christie and Huang (1995) have noted that “herding behavior is a very short-lived phenomenon”; and therefore we restrict our attention to analysis using daily data. The daily information from the stock of Shanghai stock exchange were obtained using software called “Straight Flush” over the period from June 22 2005 to October 18 2008. During this period, the index of the Shanghai Stock Exchange began to increase from 1073.86 to its historical peak of over 6000 in October 2007 and then dropped sharply to the low of 1861.47. In this period, we can distinguish two stages of the market. The first part (from June, 2005 to October 2007) was a bull market and the second part (from November, 2007 to October 2008) was a bear market. The sample includes 794 observations. Within this sample, there are 22 industries and 816 firms that are listed as A-shares on the Shanghai exchange. This analysis focuses on 814 firms except for the wood industry because there were only have two companies in this sector on the list.

The daily information consists of the stock price and the trading volume. The stock return for A-shares is calculated for a normal day as:

However if dividends exist, then the reinvestment returns are considered. Their calculation is complicated and is therefore explained in the Appendix 2. Other necessary information comes from reports of the State Statistical Bureau and China Securities Regulatory Commission (CSRC).

IV - ii Descriptive Statistics

(22)

Identify market trends

In the models of CSAD and CSSD, we not only test the data over the whole period, but also distinguish the bear and bull markets in order to further the research and the existence of the asymmetrical phenomena of herding behavior in the bull and bear markets. Therefore we distinguish the market trend via observing the range of upward and downward trends in the market index. Based on the theory of Fabzzi (1979), a positive monthly return for the market is defined as an up-market while a negative return is defined as a down-market. A monthly increase over 20% for the market can be regarded as a bull market, and in market conditions in which the prices of securities are falling (a downturn of 20% or more in multiple broad market indexes), and there is widespread pessimism leading the negative sentiment to become self-sustaining over at least a two-month period, this is considered an entry into a bear market.

Table 1 presents the descriptive statistics of the bull and bear markets. During this period, we can easily distinguish the bull and bear markets from the trend in the market index. From June 21, 2005 to October 16, 2007, the index of the market increases to the maximum and after that, the index decreases sharply from 6092.06 to 1771.82 in October 16, 2008. Obviously the first half part of this period is a bull market and the second part is a bear market.

Graph 1 shows the market trend. After distinguishing the market trend, the time that the daily index was a bull market was 541 days and it was a bear market for 253 days. The total period of the sample was 794 days.

Returns of the market, CSSD, CSAD

(23)

daily returns, measured by standard deviation, ranges from a low of 2.155% for Petroleum, Chemical Products, Plastics, Rubber to a high of 3.44% for Food, and Beverages. The minimum daily return and the maximum daily return are in the range of between -10% and 10% of approximately 9.542% because of the rules of Limiting up and Down (limitation of 10%) by the China Securities Regulatory Commission (CSRC) in China.

(24)

V. Empirical Results

This chapter is divided into two parts. Part one describes the results of measuring herding behavior using the CSSD and CSAD models. Part two considers the results that examine the factors that lead to herding behavior. In this research, the following factors were chosen as the focus of this research: Pressure of market, depth of market, the impact of regulation changes, the volatility of market, return of market, interest rate spreads between long-term and short-term, the supply of currency and growth of real estate price.

V - i Measuring Herding Behavior

In the empirical analysis, this research examined the presence of herding behavior that would lead to security returns not deviating far from the overall market return.

Test of CSSD

Christie and Huang (1995) have suggested that forecasted variances between rational asset pricing models and herding behavior models are significant. When individual returns herd around the market consensus, dispersions are predicted to be relatively low. In contrast, rational asset pricing models predict an increase in dispersion because individual returns are repelled away from the market return when stocks differ in their sensitivity to market movements. Two sets of dummy variables Du and Dl were used to identify the days that displayed extreme market movements. We use a 1% and 5% criteria for the lower (upper) tail of the market return distribution. In the regression, α means the dispersions of the sample excluding the part of the stock return coming from two dummy variables. βL and βU which should be positive in rational asset pricing models and negative if there is evidence of herding behavior.

(25)

of market stress induce decreased levels of dispersion as individual returns differ in their sensitivity to the market returns. So we should use the alternative CSAD model to test the presence of herding behavior.

Test of CSAD

CSAD is an alternative approach which stems from CAPM proposed by Chang et al (2000) which can measure the dispersion of stock returns and test whether the relationship between this CSAD indicator and the market returns is linear or not. According to Chang, Cheng, and Khorana (2000), in the presence of herding the relationship between CSADt and the average market return is non-linear. A

significantly negative coefficient γ2 implies the presence of herd behavior.

The table 6 describes the results from estimating the herding regression in which a negative value for the coefficient γ2 is consistent with herding. The results indicate that γ2 is significantly negative for the whole of the markets, suggesting that herding behavior exists in A-share markets on the Shanghai stock exchanges. This finding does not support the rational asset pricing model results that predict that periods of market stress induce increased levels of dispersion as individual returns differ in their sensitivity to the market return. However, we see that the relationship between return dispersions and market return will be negative. This can be interpreted as being consistent with prior studies, which have found evidence of herding many times in emerging markets.

To sum up, through the inclusion of the two models, we find that the results of the CSSD and CSAD models are different. As stated previously, the results from the CSSD do not support herding behavior but CSAD did. We compared the R-squared values to identify which model is preferred. The R-squared of CSSD is 0.003802 (5%) and 0.004185 (1%), however, the R-squared increases sharply to 0.073035 when using CSAD. Obviously, CSAD seems much better than CSSD when measuring stock return dispersions.

(26)

market. When the market is an upward market (bull market), the regressions yield statistically significant (5% level) and negative γ2 coefficients, indicating that herding takes place in these markets. A similar situation is seen in the downward market (bear market). However, comparing the coefficient values for the upside and downside moves of the market, we see that return dispersions during extreme downside moves of the market are lower than those for upside moves. This result suggests that herding is stronger during periods of a decreasing market, indicating a “flight to safety” for the market consensus in bad times.

Having found the evidence of herding behavior using the whole market data, we next examine the return dispersions for all industries separately with the exception of the Wood Industry. Table 8 reports the results for all industries. γ2 coefficients ranges from a low of -2.94 for Real Estate and a high of 4.65 for Finance, Insurance. Consistent with prior studies using the whole market data, over half of γ2coefficients (11 industries) are significantly negative, indicating that herding takes place in these industries, especially in Real Estate Industry.

V - ii Influencing Factors Affecting Herding Behavior

Tests of herd formation using the CSSD for the empirical research provide no evidence to support herding behavior in Chinese Shanghai markets, but the CSAD model has demonstrated that this behavior has been shown to exist. So the CSAD seems a better indicator compared to the CSSD. In this section, we use the CSAD as the proxy variable of herding behavior and build a regression to further investigate the influencing factors that cause herding behavior. In the first subsection, we test the autocorrelation of the influencing factors. Following this we further examine the potential impact on herding behavior with respect to 6 potentially influential factors.

Autocorrelation of influencing factors

(27)

matrix. From the matrix, we can see that herding behavior may be more pronounced during periods of market stress.

Table 9 reports the results of correlation matrix. It can be seen that market stress for the lower return, market depth, and regulation announcements have a negative relationship with the CSAD indicator especially the regulation announcements which has a significantly negative correlation. When the market pressure on the extreme short position or some new regulation announcements increases, the CSAD will decrease, indicating that herd formation will exist in this market. And then we can use VIF (variance-inflation factor) to test the autocorrelation of the influential factors in the regression model. The detailed explanation of VIF is discussed in the Appendix 3. Multivariate Regression Analysis

Table 11 shows the estimation results for the regression models. For the analysis of the regression model as a whole, the F-test is significant at the 5% level and the adjusted R-square is 0.487. The results of the variance-inflation factor (VIF) are far less than 10, so we can conclude that the model examined here has no autocorrelation among the independent variables.

The coefficients of Dl and regulation announcement were found here to be statistically significant and negative for Shanghai stock markets, indicating that the cross-sectional absolute deviations are significantly smaller when these factors increase or occurs. Although the coefficient of market depth is negative, it is not significant and cannot therefore be regarded as the driving factor for herding behavior. The coefficients for the other three factors including volatility, Du, daily return and change of volume are positive, but only volatility and daily returns were found to be significant. So it can only be said that they are not influential factors, however, there does not appear to be any supporting evidence that Du and change of volume cause or enhance the herding.

DL and DU are the dummy variables that correspond to the market pressure. Our

(28)

possible increase of herding, which indicates that herding behavior is enhanced when the market returns are very low. The reason for this can may be explained by the fact that the investors are risk averse, and so when the market return is low; they tend to discard their positions of stocks and quickly protect the profit they hold. In contrast, in a situation where there are high market returns, they tend to think carefully and make more rational investment choices.

As part of the regulation announcements, we examined the impact of lifting the ban on trading “Large and Small Non-Tradable Shares” on the stock market. The coefficient of this factor was -0.0177 which is significant under 5%, indicating there is an evidence to support the hypothesis, which was referring to the description of influencing factors. This result proves that many commentators tend to speculate that the large-scale shaking of the Shanghai stock index in year 2008 may have been caused by this factor. In consequence, investors who suffered psychic anxieties, dumped stock soon afterwards and these actives would therefore have enhanced herding behavior. So we can deduce that lifting the ban on trading non-tradable shares is very likely to cause herding behavior to occur. Individual investors are basically trend chasers who sell securities when they hear that the initial share owners have sold their shares and buy stocks when these owners have bought their shares at the secondary market

The coefficient of the depth of market is close to Zero (-3.07E-08) and it is not significant under 5%, this result shows that this factor does not affect herding behavior. Because the Chinese stock market belongs to a shallow market (Zhang and Liang 2006), the market is not deep enough to compare to Western stock markets, and therefore a little trading form large institutions may have a great influence on the market price. So we can deduce that the depth factor does not enhance herding behavior. This view is different from other research that has examined western stock markets.

(29)

(30)

VI. Conclusion

VI - i Conclusions

This research has examined the existence of herd behavior in the Shanghai A-share stock market during both bull and bear markets. The sample used data from June 21, 2005 which was regarded as the starting point of the Chinese bull market to October 18, 2008 when the stock market had hit the bottom over a period of 794 trading days. Results based on the results from the CSAD model indicate that herding behavior is indeed present although the CSSD model does not support this finding. These results are similar to those reported by Tan et al. (2008). This research then considers the market stress, the depth of market, regulation announcement, volatility of stock price, market return and change of volume as independent variables in the regression model to test the correlation between these factors and the indicator of herding behavior (CSAD). The main conclusions of this research are as follows:

In the Shanghai A-share stock market, the volatilities of stocks are not determined in rational asset pricing models, which are based upon perfect information and symmetrical information. Investors are more likely to ignore their private rational information and go with the market consensus during periods of market stress (wide range of index of market). Consistent with prior studies, we find evidence of herding behavior using the CSAD model and company-level data from the Shanghai Stock Exchanges. After distinguishing between the bull and bear markets, this phenomenon still significantly exists.

(31)

VI - ii Further Research

The following are suggestions for future work:

As the CSAD model when measuring herding behavior is based on the CAPM model, a specification error when estimating the coefficients may occur, while this model may also overlook the impact of market volatility on the model itself. Therefore future work could use GARCH models or other quantitative methods to overcome these problems in this model.

(32)

VII. References

Admati, A. R., and P. Pfleiderer, 1998, Forcing Firms to Talk: Financial Disclosure Regulation and Externalities, Working Paper, Stanford University.

Banerjee, A., 1992, A Simple Model of Herd Behavior, Quarterly Journal of Economics, Vol. 107, 797–8187

Bikhchandani, Sushil, David Hirshleifer, and Ivo Welch, 1992, A Theory of Fads, Fashion, Custom and Cultural Change as Informational Cascades, Journal of Political Economy, Vol. 100, 992–1026.

Bikhchandani Sushil, and Sunil. Sharma, 2001, Herd Behavior in Financial Markets: A Review, IMF Staff Papers, 2001, 47(3), 279-310.

Chang, Eric, Joseph Cheng, and Ajay Khorana, 2000, An Examination of Herd Behavior in Equity Markets: An Empirical Perspective, Journal of Banking and Finance, No.24: 1651-1679 Chari, V.V. and Patrick Kehoe, 1999, Financial Crises as Herds, Federal Reserve Bank of

Minneapolis.

Chen, Y. M., 2002, Domestic investors’ herding behavior in reaction to foreign trading, Working

paper, San Francisco State University.

Christie, William G., and Roger D. Huang, 1995, Following the Pied Piper: Do Individual Returns Herd around the Market? Financial Analysts Journal, July-August, 31–37.

Chui, A.C.W. and Kwok, C.C.Y., 1998, Cross-autocorrelation between A-shares and B shares in the Chinese stock market. Financial Res. 21, pp. 333–335.

Devenow Andrea and Ivo Welch, 1996, Rational Herding in Financial Economics, European

Economic Review, Vol. 40, pp. 603–15.

De Long, Bradford, Andrei Schleifer, Lawrence Summers, and Robert Waldman,1990, Positive Feedback Investment Strategies and Destabilizing Rational Speculation, Journal of Finance, Vol. 45, pp. 379–95.

Fung, H.G., Lee, W., Leung, W., 2000, Segmentation of the A- and B-share Chinese equity markets, Journal of Financial Research, Vol. 23, 179–195.

J. Fernald and J.H. Roger, 2002, Puzzles in the Chinese stock market, Review of Economics and

Statistics, Vol. 84, 416–432

Fung, H.G., Lee, W., and Leung, W., 2000, Segmentation of the A- and B-share Chinese equity markets, Journal of Financial Research, Vol. 23, 179–195.

Gleason, K. C., Mathur, I. and Peterson, M. A., 2004, Analysis of Intraday Herding Behaviour Among the Sector ETFs, Journal of Empirical Finance, Vol. 11, 681-694.

Gleason, K.C., Lee, C.I., Mathur, I., 2003, Herding behavior in European futures markets, Finance

Letters, Vol. 1, No. 1, 5–8.

Henker, J., T. and Mitsios, A., 2006, Do Investors Herd Intraday in Australian Equities?,

International Journal of Managerial Finance, Vol. 2, No. 3, pp. 196-219 (24).

Hwang, Soosung & Salmon, Mark, 2004, Market Stress and Herding, Journal of Empirical

Finance, Vol. 11, 585-616

Keynes, J. M., 1936, The general theory of employment, interest, and money, Harcourt, Brace,

and Co., New York

(33)

Journal of Financial Economics, Vol. 32, 23–43.

Lakonishok, J., A. Shleifer, R. Thaler, and R.Vishny, 1991, Window dressing by pension fund managers, American Economic Review Papers and Proceeding Vol.81, 227-231

D. Long, J. Payne and C. Feng, 1999, Information transmission in the Shanghai equity market,

Journal of Financial Research, Vol. 22, 29–45..

Nofsinger, John R. and Richard W. Sias, 1999, Herding and Feedback Trading by Institutional and Individual Investors, Journal of Finance, Vol. 54, pp. 2263–95.

Odean, T., 1998, Are investors reluctant to realize their losses, Journal of Finance, Vol.53, No5, 1775-1798

Oehler, A., 1998, Do mutual fund specializing in German stocks herd? Finanzmarktund Portfolio

Management, No.4, pp452-465

Poon, M. Firth and H.G. Fung, 1998, Asset pricing in segmented capital markets: preliminary evidence from China-domiciled companies? Pacific-Basin Finance Journal, Vol. 6, 307–319 Poon, W. Fung, H.G., 2000, Red chips or H-shares: which China-backed securities process

information the fastest? Journal of Multinational Financial Management, Vol. 10, 315–343. Scharfstein, David S., and Jeremy Stein, 1990, Herd Behavior and Investment, American

Economic Review, Vo. l80, No. 3, 465-479.

Shiller, R. J., 1984, Stock prices and social dynamics, Brookings Papers on Economic Activity, Vol. 2, 457-498.

Sias, R. W., 2004, Institutional Herding, The review of financial studies, Vol.12, No.1: 165-206 Song, H., Liu, X., and Romilly, P., 1998. Stock returns and volatility: an empirical study of

Chinese stock markets. International Review of Applied Economics, Vol. 12, 129–139.

D. Su and B. Fleisher, 1999, Why does return volatility differ in Chinese stock markets?,

Pacific-Basin Finance Journal, pp. 557–586

D. Su and B. Fleisher, 1998, Risk, return and regulation in Chinese stock markets, Journal of

Economics and Business, Vol.50, 239–256.

Wermers, R., 1999, Mutual Fund Herding and the Impact on Stock Price, The Journal of Finance, Vol. 54, No.2, 581-622

Zhang, Jikang and Liang Yuanyuan, 2006, The Institutional and Structural Problems of Chinas Foreign Exchange Market and Implications for the New Exchange Rate Regime, China: An

(34)

VIII. Appendix

V III- i Tables and Graphs

Table 1: Descriptive statistics of the Bull and Bear Markets

This table distinguishes the market trend. The discrimination is based on the monthly increase/decrease rate (20% is the standard) for the market.

Period of research Index increase (decrease) The growth (drop)rate Status of market 07/21/2005-10/16/2007 1020.63-6092.06 496.89% Bull market 10/17/2007-10/28/2008 6036.38-1771.82 -70.64% Bear market Graph 1: Market trend of Shanghai A-share market

The Graph shows the trend of market during the period from June 22 2005 to October 18 2008. The market index began to increase from 1073.86 to its historical peak of over 6000 in October 2007 and then dropped sharply to the low of 1861.47.

Market Trend 0 2000 4000 6000 8000 07/21/2₀₀₅03/08/2₀₀₆10/18/2₀₀₆06/04/2₀₀₇01/09/2₀₀₈08/19/2₀₀₈ day Inde x

Table 2: The summary statistics of Chinese stock market characteristics

(35)

Fishing and Hunting

Mining 24 794 0.107% 2.926% 46.456% -10.534% 9.298%

Food ,Beverage 37 794 0.122% 3.447% 54.713% -9.066% 9.190%

Textile, Apparel, Leather 36 794 0.052% 2.449% 38.880% -9.757% 9.251%

Wood Product 2 794 -0.038% 2.779% 44.110% -9.658% 8.950%

Paper, Printing 18 794 0.015% 2.582% 40.981% -10.089% 9.515%

Petroleum, Chemical

Product, Plastics, Rubber 80 794 0.142% 2.155% 34.210% -8.820% 8.125%

Electrical Equipment 28 794 0.007% 2.444% 38.805% -9.629% 9.443%

Metal, Nonmetallic Mineral

Product 74 794 0.075% 2.580% 40.955% -9.529% 8.966%

Machinery,Equipment,Meter 87 794 0.099% 2.318% 36.804% -8.467% 9.122%

Medicine, biologic Product 55 794 0.139% 2.275% 36.119% -8.826% 7.929%

Eletricity,Gas,Water

Supply 40 794 0.047% 2.627% 41.705% -10.183% 9.340%

Construction 22 794 0.072% 2.798% 44.421% -10.412% 9.274%

Transprot,Storage 47 794 0.032% 2.333% 37.040% -9.505% 9.120%

Information,Technology 52 794 0.018% 2.370% 37.623% -9.658% 9.029%

Wholesale and Retail Trade 62 794 0.153% 2.317% 36.775% -8.864% 9.299%

Finance, Insurance 20 794 0.098% 3.115% 49.444% -10.213% 9.549% Real Estate 36 794 0.128% 2.963% 47.039% -10.109% 9.506% Social Services 19 794 0.103% 2.655% 42.142% -9.593% 9.354% Transmission Culture 8 794 0.049% 3.431% 54.459% -10.554% 9.523% Conglomerate 45 794 0.034% 2.383% 37.834% -8.591% 8.137% Total 816 794 0.069% 2.116% 33.592% -9.261% 9.033%

Table 3: The statistic description of CSSD

This table lists descriptive statistics for an average number of companies from each industry with the exception of wood industry, the number of observations, the daily Cross-Sectional Standard Deviations (CSSDt), standard deviations and min/max value. The daily data range is from 7/22/2005 to 10/28/2008. It should be noted that the number of stocks in an industry do not stay constant every day, so the number of returns used to calculate the CSSDs varies over time.

Industry Firms number Observation Mean (%) Standard

deviation (%) Min% Max%

Mining 24 794 2.93% 1.07% 0.75% 7.50%

Food ,Beverage 37 794 3.02% 3.43% 0.57% 48.59%

Textile, Apparel, Leather 36 794 3.00% 1.36% 0.83% 18.17%

Paper, Printing 18 794 2.87% 2.73% 0.84% 52.79%

Petroleum, Chemical

Product, Plastics, Rubber 80 794 3.16% 10.36% 0.74% 290.95%

Electrical Equipment 28 794 2.94% 1.28% 1.14% 24.26%

Product 74 794 2.59% 0.96% 0.81% 8.78%

Machinery,Equipment,Meter 87 794 3.06% 4.99% 0.99% 134.07%

Medicine, biologic Product 55 794 2.85% 1.43% 0.98% 34.58%

(36)

Supply

Construction 22 794 2.34% 0.91% 0.68% 12.00%

Transprot,Storage 47 794 2.81% 1.21% 0.69% 15.16%

Information,Technology 52 794 2.56% 1.62% 0.90% 29.58%

Wholesale and Retail Trade 62 794 2.79% 0.90% 0.94% 8.75%

Finance, Insurance 20 794 3.02% 1.97% 1.00% 51.21% Real Estate 36 794 3.07% 4.34% 0.48% 97.88% Social Services 19 794 2.97% 1.07% 0.92% 9.34% Transmission , Culture 8 794 2.84% 2.05% 0.69% 50.47% Conglomerate 45 794 3.49% 4.57% 0.53% 123.84% Total 816 794 2.82% 1.83% 0.98% 45.71%

Table 4 The statistical description of CSAD

This table lists descriptive statistics for an average number of companies from each industry with the exception of wood industry, the number of observations, the daily Cross-Sectional absolute Deviations (CSADt), standard deviations and min/max value. The daily data range is from 7/22/2005 to

10/28/2008. It should be noted that the number of stocks in an industry do not stay constant every day, so the number of returns used to calculate the CSADs varies over time.

Industry Firms number Observation Mean (%) Standard

deviation (%) Min% Max%

Mining 24 794 2.07% 1.18% 0.42% 13.26%

Food ,Beverage 37 794 2.24% 0.86% 0.78% 7.43%

Textile, Apparel, Leather 36 794 2.03% 0.87% 0.63% 11.90%

Paper, Printing 18 794 2.12% 2.68% 0.55% 73.78%

Petroleum, Chemical

Product, Plastics, Rubber 80 794 2.17% 0.73% 0.73% 5.97%

Electrical Equipment 28 794 1.92% 0.74% 0.59% 5.96%

Product 74 794 2.12% 0.97% 0.71% 19.05%

Machinery,Equipment,Meter 87 794 2.08% 0.69% 0.66% 6.97%

Medicine, biologic Product 55 794 1.97% 0.65% 0.65% 5.20%

Supply 40 794 1.70% 0.64% 0.58% 4.13%

Construction 22 794 2.10% 0.92% 0.50% 7.40%

Transprot,Storage 47 794 1.83% 0.65% 0.64% 6.33%

Information,Technology 52 794 2.09% 0.69% 0.69% 6.18%

Wholesale and Retail Trade 62 794 2.28% 0.87% 0.74% 10.36%

(37)

Table 5: Results from CSSD Model

This table reports results of the following regression for Shanghai A-share stock market.

L U

T L t U t t

CSSD = +a b D +b D +e , where DU/Lt =1, the market returns lie in the 5% and 1% regions,

DU/Lt = 0, otherwise. βs are the coefficients of independent variables. αis the intercept term. The sample period is from 7/26/2005 to 10/28/2008. Numbers in parentheses are standard errors; Prob. is the probability of t-statistics. The results represent statistical insignificance at the 5%, levels.

Model 1 α βL βU

Market return in the extreme upper/lower 5% of the return distribution

0.034131 -0.004509 -0.00539

Std. Error (-0.001819) (-0.007816) (0.007816)

Prob. 0 0.5645 0.4908

Market return in the extreme upper/lower 1% of the return distribution

0.033862 -0.012386 -0.00785

Std. Error (0.001744) (0.015916) (0.015916)

Prob. 0 0.4372 0.6225

Table 6 Results of CSAD model

This table reports results of the following regression for Shanghai A-share stock market

2 1 2

T m t m t t

C SA D =a g+ R +g R +e t, where Rm,t is the equally weighted portfolio return at time t.

CSADt is the equally weighted cross sectional absolute deviation.βs are the coefficients of

independent variables. αis the intercept term. The sample period is from 7/26/2005 to 10/28/2008.

Numbers in parentheses are standard errors; Prob. is the probability of t-statistics. The results represent statistical insignificance at 5% level.

Model 2 α γ1 γ2

The whole market 0.018548 0.237003 -2.41595

Std. Error 0.000391 0.035446 0.552965

Prob. 0 0 0

Table 7 Results of CSAD model in bull and bear markets

This table reports results of the following regressions. Former one is for upward market (bull market) and later one is for downward market (bear market):

2 1 2 up up up up up T mt mt t CSAD = +a g R +g R +e and 2 1 2

down down down down down

T mt mt t

CSAD = +a g R +g R +e , where Rup/downmt is the equally weighted

(38)

market is from 10/17/2007 to 10/28/2008. Numbers in parentheses are standard errors; Prob. is the probability of t-statistics. The results represent statistical insignificance at the 5% level.

Model 2 α γ1 γ2 upward market 0.018152 0.242604 -1.60565 Std. Error 0.000504 0.051173 0.910115 Prob. 0 0 0.00783 Model 2 α γ1 γ2 downward market 0.020031 0.159456 -1.9688 Std. Error 0.000559 0.041933 0.583053 Prob. 0 0.0002 0.0009

Table 8 Results of CSAD model using industries-level data

Regression coefficients are for

2 1 2

T mt mt t

CSAD = +a g R +g R +e

using industries-level data.

T-Statistic significance at 5%

Return dispersions Intercept item Absolute of return Square of return

Industry

Firm

No. α γ1 t-Statistic γ2 t-Statistic

the whole market 814 0.018152 0.242604 -1.605645

Agriculture, Forestry,

Fishing and Hunting 24 0.017229 0.333057 7.664264 -2.772147 -4.89509

Mining 24 0.016838 0.114605 0.0188 1.559791 0.009

Food ,Beverage 37 0.017932 0.306363 6.833477 -2.098951 -3.48802

Textile, Apparel, Leather 36 0.018718 -0.01077 -0.276986 2.292917 4.755396

paper, Printing 18 0.018481 0.049335 2.726411 1.36107 44.39895 Petroleum,Chemical Product, Plastics,Rubber 80 0.018964 0.111017 2.996369 -0.896378 1.721167 Electrical Equipment 28 0.017146 0.098139 2.566466 0.066723 0.130748 Metal,Nonmetallic Mineral Product 74 0.021435 -0.178815 -4.646393 4.511481 10.00056 Machinery,Equipment,Meter 87 0.018226 0.174732 4.839428 -1.138403 -2.34823

Medicine, biologic Product 55 0.017168 0.16638 4.674644 -0.923344 -1.85191

Supply 40 0.013584 0.259806 8.362902 -2.246136 -5.50651

Construction 22 0.017236 0.276766 5.723725 -2.483246 -3.86728

Transprot,Storage 47 0.015184 0.229155 6.944704 -1.714001 -3.69333

Information,Technology 52 0.018313 0.160605 4.351585 -0.823455 -1.60799

Wholesale and Retail Trade 62 0.018999 0.241099 5.022238 -1.105684 -1.60607

Finance,Insurance 20 0.020479 -0.118643 -2.681888 4.653113 12.09283

Real Estate 36 0.01851 0.284961 7.029592 -2.944572 -5.52526

Social Services 19 0.018383 0.137539 2.951114 0.105961 0.178732

Transmission , Culture 8 0.023732 0.01302 0.817245 2.124415 27.84304

(39)

Table 9 Correlation matrix of influencing factors

This table presents correlations between CSAD and influencing factors. DL andDUare dummy

variables which is 1 for market returns that lie in the extreme 1% lower and upper tail of the

distribution of market returns and zero otherwise; DEPTH is the depth of the market; BAN is a dummy which is one for markets that the non-tradable shares has been trader by owners and zero otherwise; VOLATILITY is the standard deviation of daily return; DAILY_RETURN is daily return of the weighted stock price index and CHANGE OF VOLUME is the daily change of trading volume.

CSAD_

MARKET DL DU DEPTH BAN

DAILY_ RETURN VOLA- TALITY CHANGE_OF _VOLUME CSAD_MARKET 1.0000 DL -0.4175 1.0000 DU 0.2744 -0.0531 1.0000 DEPTH -0.0027 0.0136 0.0156 1.0000 BAN -0.3566 0.0893 0.0518 0.0247 1.0000 DAILY_RETURN 0.6271 -0.5866 0.4752 0.0032 -0.1511 1.0000 VOLATALITY 0.1031 0.0574 -0.0057 0.0229 -0.0168 -0.0896 1.0000 CHANGE_OF_VOLUME 0.1155 0.1023 0.2454 0.0470 -0.0038 0.1968 -0.0263 1.0000

Table 10 The results of regression model for testing influencing factors that affecting herding behavior.

Regression coefficients are for t 0 1 2 3 4 5

6 7 (CSSD ) ( ) ( ) L U T t t t t t mt mt t t CSAD D a D Dep D R TV a a a a a s a a e = + + + + + + + + . The

sample period is from 7/26/2005 to 10/28/2008. Prob. is the probability of t-statistics. VIF is results of variance-inflation factor, which is to test the autocorrelation of the influential factors in the regression model. The results represent statistical insignificance at 5% level.

Independent variables Coefficient Std. Error t-Statistic Prob. VIF

C 0.00889 0.001083 8.211903 0

DL -0.00857 0.003177 -2.69736 0.0071 1.80

DU 0.002658 0.002889 0.920167 0.3578 1.49

DEPTH -3.07E-08 6.24E-07 -0.04924 0.9607 1.00

BAN -0.017698 0.001726 -10.2521 0 1.05

DAILY_RETURN 0.510216 0.037907 13.4596 0 1.01

VOLATALITY 0.196596 0.033107 5.93821 0 2.39

CHANGE_OF_VOLUME 0.00163 0.00264 0.617307 0.5372 1.14

Adjusted R-squared 0.486687 S.D. dependent var 0.020388

Log likelihood 2233.011 Hannan-Quinn criter. -5.58645

(40)

V III- ii Daily Returns for A Share with Cash Dividends Reinvested (RETA)

DESCRIPTION:

1. Item represents daily stock returns with cash dividends reinvested.

2. If any data element used in calculation is unavailable, a missing value is stored in this field.

3. RETA includes all distributions. 4. RETA is calculated as follows:

, 1 1 1 1 2 2 1 ' ' t t t n t t t P f d r P_- P_- k p k p = + -+ + Where 1 2 1 1 1 1 2 2 (1 ) ' ' t t t k k p f P k p k p -+ -+ = + + t

f = cumulative adjustment factor at day t,

t

d = cash distribution amount at day t,

t

p = closing price at day t.

(41)

V III- iii Variance Inflation Factor（VIF）

In statistics, the variance inflation factor (VIF) quantifies the severity of

multicollinearity in an ordinary least squares regression analysis. It provides an index that measures how much the variance of an estimated regression coefficient (the square of the estimate's standard deviation) is increased because of collinearity.

The VIF can be calculated in three steps: Step one

Calculate k different VIFs, one for each Xi by first running an ordinary least square regression that has Xi as a function of all the other explanatory variables in the first equation.

If i = 1, for example, the equation would be

Where c0 is a constant and e is the error term.

Step two

Then, calculate the VIF factor for with the following formula:

Where R2iis the coefficient of determination of the regression equation in step one.

Step three

HERDING BEHAVIOUR IN EXTREME MARKET CONDITIONS: THE CASE OF THE CHINESE STOCK MARKET Yue Fu Faculty of Economics and Business Rijksuniversiteit University Supervisor: Marc Kramer