Herding in China Equity Market

Herding is an irrational behavior and low information cost strengthens herding. Banerjee (1992) defines herding as ‘everyone doing what everyone else is doing, even when their information suggests doing something different’. Prechter and Parker (2007) suggest that uncertainty about valuation may cause herding. Kultti and Miettinen (2006) set up a standard sequential decision model; they purpose that if the cost of the information about the predecessors’ actions is very expensive then all the agents will act according to their own signals. If observing is free one acts in herding behavior. Facing financial panic, investors may not have enough time to collect and analyze valuable information from many disorderly data. Investors may herd during financial panic.

2002 to 2006; world equity market grows only 1.87 times at the same period.Contrasting to the importance of China equity market in worldwide financial markets, there are scare researches discussing about herding behavior in China, one of the most prosperous emerging markets.Previous studies only discuss about the herding in dual-share.Zhou (2007) investigates the herding behavior in China's A and B markets and finds the existence of significant herding in both A and B share markets.Chong and Su (2006) examine the co-movement between the A shares and H shares of twenty-one cross-listed Chinese companies and find a small portion of sample have a co-movement.Contrasting to previous studies focusing on herding in dual share, this study aims to discuss about the market wide herding in China in Shanghai market and Shenzhen market.
The remainder of this paper is partitioned as follows:(1) the methodology and data that include herding measurement 、hypotheses and the source of data; (2) results of the empirical tests; (3) conclusions.Christie and Huang (1995) and Chang et al. (2000) propose that investors herd during periods of high market volatility.When herd exists, the returns of individual stock converge towards the returns on the aggregate market -market index.Thus, herding results in a smaller difference between the returns on the individual stock and the market index.We use two alternative measures of dispersion, CSSD and CSAD, to identify herding behavior.

Herding Measurement
The cross-sectional standard deviation (CSSD) method is proposed by Christie and Huang (1995) and be expressed as Chang et al. (2000) define the cross-sectional absolute deviation (CSAD) as In this study, R i , t is the return of stock i during time period t; R m , t is the return of market index during the same time period t; N t is the number of stock listed in equity market during time period t.Shanghai equity market and Shenzhen equity market have their own CSSD and CSAD values at time period t.Shanghai composite index and Shenzhen composite index are used as proxies to measure Shanghai equity market index and Shenzhen equity market index.

Test of Herding
Herding will be more prevalent during periods of market stress, which is defined in terms of extreme market returns.If individual is rational, individual asset should have different sensitivity to the market return.So, β 1 and β 2 be zero, or not significantly positive and negative, indicates that rational model is fit.Chang et al. (2000) argue that the model in Eq. ( 3) requires defining what is meant by market stress and they propose a nonlinear relationship between CSAD and market return as follows: If herd exists, then γ 2 will be significantly negative.Gleason et al. (2004) suggest dependent variables in Eqs.(3) and (4) could be swapped, which are expressed in the following equations: To test the turnover rate (traded volume/ total shares) effect on herding, we define High Turnover Standard Deviation (HTSD) 、Low Turnover Standard Deviation (LTSD)、High Turnover Absolute Deviation (HTAD) and Low Turnover Absolute Deviation (LTAD) as follows: Where t is time period, R i is return of stocks i. R t m, is market return.When a stock's turnover rate is higher than median value of turnover at the same time period (month) in the same stock market, this stock is classified into high turnover stock (R t h, ); otherwise, it is a low turnover stock(R t l , ).Nt is the number of stock at time period t.Shanghai market and Shenzhen market has its own four herding measures, HTSD、 LTSD、 HTAD and LTAD for each month.To verify the turnover effect on herding, these four herding measures are treated as dependent variable in Eqs.(3) to (6).

Herding Hypothesis
We hypothesize low turnover stock will have higher tendency to herd market return.The HTSD and HTAD are used to measure the degree of high turnover rate stocks disperse from market return; the LTSD and LTAD are used to measure the degree of low turnover rate stocks disperse from market return.Based on the statements of Gregroriou and Ioannidis (2006) and Avery and Zemsky (1998), low turnover stock is lacking sufficient information and the lack of information will lead low turnover stock more tender to herd market return than high turnover stock.Low dispersion from market return means higher tendency to herd market.If turnover rate effect exists, the dispersion measurements, SD, AD, should be higher when turnover rate is high.Hypothesis 1: The mean value of HTSD (HTAD) will be significantly higher than LTSD (LTAD).
Herding is information dissemination.During extreme market situation, noise traders do not know the value of new information and need to make decision in a short period; they will herd.Based on Kultti and Miettinen (2006), it takes no cost to observe the change of market return; investors will tend to herd market return during extreme market situation.Herding is an irrational behavior and does not follow traditional hypothesis that people are rational.Based on traditional market hypothesis such as the Capital Asset Pricing Model that investors are rational then herd will not exist.To test the existence of herding and turnover effect on herding, regression models, Equation 11 and Equation 12 are used.These two equations are similar to Equation 3 and Equation 4, but the dependent variable Y which is denoted as herding measures: CSSD, CSAD, HTSD, LTSD, HTAD and The percentage of upper or lower tail of market return distribution can be set up as we want, and present study uses 5%.
Similar to the regression models of many researchers such as Christie and Huang (1995) and Chang et al. (2000), the coefficient of β 1, β 2 or γ 2 are used to test herd.If herd exists in China equity market during extreme market situation, the value of CSSD or CSAD will become smaller, this means that β 1, β 2 or γ 2 will be significantly negative.The significantly negative β 1 means investors herd during extreme upward market situation.The significantly negative β 2 means investors herd during extreme downward market situation.The significantly negative γ 2 means investors herd during extreme upward and downward market situation.
Hypothesis 2: If β 1, β 2 or γ 2 are significantly negative when dependent variables are CSSD and CSAD, herd exists in China equity market.

Turnover Effect Hypothesis
To testify the turnover effect, the sample is divided into two groups based on the turnover value of the stocks in each market, Shanghai and Shenzhen.The value of HTSD,LTSD,HTAD and LTAD will be calculated for each month.If turnover effect exists, the value of LTSD or LTAD will become significantly smaller during extreme market situations, while HTSD or HTAD will not become significantly smaller during extreme market.This means that β 1, β 2 or γ 2 are significantly negative only when dependent variables are LTSD or LTAD; and β 1, β 2 or γ 2 are not significantly negative when dependent variables are HTSD or HTAD.

Asymmetric Reaction
To ground on Christie and Huang (1995) and Chang et al. (2000), markets' reactions towards good news and bad news would appear to be diverse; If the dispersion is higher in up market, relative to down market, it is because investors are more fear of the extreme movements in down market.Christie and Huang (1995) use the difference between β 1 and β 2 to measure asymmetric reaction.β 1 and β 2 are estimated based on Equation ( 11) and the dependent variable Y denotes as dispersion measures: CSSD, CSAD, HTSD, LTSD, HTAD and LTAD.Chang et al. (2000) suggest other models to verify the asymmetric reaction.Equation 13 is used when the monthly market return, t m R , , is greater and equal to zero; this is the upward market.Equation 14 is used when the monthly return, t m R , , is less than zero; this is the downward market.The ) is rejected, then the degree of herding appears to be asymmetric during up market and down market.

Data
We obtain monthly data of listed stocks and market index from China database of Taiwan economic journal (TEJ) for period January, 2004 to June, 2009 which covering current financial panic.Shanghai composite index and Shenzhen composite index are chosen as proxies of markets because these two indices are the longest existing index in respective market.

Descriptive statistics
Table 1 shows that mean value of Shenzhen market return, 1.77%, is higher than Shanghai market return, 1.456%.Shenzhen's higher market return accompanies with higher standard deviation value.During the 66 months' sampling periods, Shanghai has 40 months' market return greater than zero.As for Shenzhen, there is 41 months" market return greater than zero.The mean value and standard deviation value of Cross-Sectional Standard Deviation (CSSD)、HTSD and LTSD of Shenzhen market are all higher than Shanghai market.The mean value and standard deviation value of Cross-Sectional Absolute Deviation (CSAD)、HTAD and LTAD of Shenzhen market are closer to Shanghai market.
Table 1 also shows that mean value of absolute deviation (AD) is higher than mean value of standard deviation (SD) in both markets.Mean value of HTSD is higher than LTSD in both markets; mean value of HTAD is higher than LTAD in both markets.The mean value of HTSD and LTSD for Shanghai stock market is 0.032 and 0.019, respectively.The difference between mean value of HTSD and LTSD for Shanghai stock market is 0.013 and t value of paired mean test is 2.01, which is significant at 5%.The mean value of HTAD and LTAD for Shanghai stock market is 0.114 and 0.087, respectively.The difference between mean value of HTAD and LTAD for Shanghai stock market is 0.027 and t value of paired mean test is 3.34, which is significant at 1%.The mean value of HTSD and LTSD for Shenzhen stock market is 0.059 and 0.019, respectively.The difference between mean value of HTSD and LTSD for Shenzhen stock market is 0.033 and t value of paired mean test is 2.05, which is significant at 5%.The mean value of HTSD and LTSD for Shanghai stock market is 0.032 and 0.019, respectively.The difference between mean value of HTAD and LTAD for Shenzhen stock market is 0.037 and t value of paired mean test is 4.67, which is significant at 1%.The above significant mean difference between HTSD (HTAD) and LTSD (LTAD) means low turnover stock has significant low dispersion from market return.This support hypothesis 1 that low turnover stocks has a significant tendency to herd than high turnover stocks.

Herding test using Shanghai data
Model A in Table 2 show that there do not exist herding behavior in Shanghai equity market.Here, the β 1 and β 2 coefficients are not significantly negative when different dependent variables are used, indicating no convergence of the individual stocks returns to the Shanghai composite index return.Five of six regressions show significantly positive β 1 coefficient, Shanghai's stocks demonstrate higher dispersion during extreme upward market situation.2 show the same results, there do not exist herding behavior in Shanghai equity market.The γ 2 is not significantly negative.This result points to the absence of herding during periods of high market stress in Shanghai.

Model B in Table
The negative adjusted R square value is shown in Table 2 when Model B is implemented for HTSD.Based on the statement of Greene(1993), this is because X (independent variables)and Y (dependent variable) has a sample correlation of zero, but too many X are added into the regression model which will makes negative adjusted R square value.

Herding test using Shenzhen data
Model A in Table 3 show that there do not exist herding behavior in Shenzhen equity market.Here, the β 1 and β 2 coefficients are not significantly negative when different dependent variables are used, indicating no convergence of the individual stocks returns from the Shenzhen composite index return.One of the six regressions shows significantly positive β 1 coefficient, low turnover stocks demonstrate higher dispersion during extreme upward market situation.3 show the same results, there do not exist herding behavior in Shenzhen equity market.The γ 2 is not significantly negative.This result points to the absence of herding during periods of high market stress in Shenzhen.

Model B in Table
Again, there are several negative adjusted R square values are shown in Table 3.The explanation is that X (independent variables) and Y (dependent variable) has a sample correlation of zero, but too many X are added into the regression model.
The findings in 3.2 and 3.3 show that hypothesis 2 is not supported; there does not exist herding behavior in China equity market.

Turnover effect on herding
To evaluate the turnover effect on herding, we test hypothesis 3 to examine β 1, β 2 or γ 2 are significantly negative only when dependent variables are LTSD or LTAD.
The results partly support turnover effect on herding.From Table 2 and Table 3, we can notice that regressions with different dependent variables coincidently show that β 1, β 2 or γ 2 are not significantly negative.Turnover effect does not exist in full samples.
We further analyze Table 4 and Table 5 which samples are divided into upward (market return is positive) sample and downward (market return is negative) sample.For Shanghai market, there is forty monthly market returns greater than zero, the upward market situations.The largest market return value is 27.4% and the lowest market return value is 0.19% in the upward market situation.As for the Shanghai downward market situation, market returns range from -0.06% to -23.6%.For Shenzhen market, there is forty-one monthly market returns greater than zero, the upward market situations.During the upward market situation, market return ranges from 0.36% 21-31. Prechter, R. and W. Parker. (2007).The financial/economic dichotomy in social behavioral dynamics: socionomic perspective.Journal of Behavioral Finance, 8(2), 84-108. Zhou, H. (2007).Herding in dual-share stock markets: evidence from China.Journal of Emerging Markets, 12(2), 5-15.
the return on the market for time period t lies in the extreme upper tail of the returns distribution, , if the return on the market for time period t lies in the extreme lower tail of the returns distribution, This study adopts 5% to define extreme market upward and downward.If herd exists, CSSD t will be smaller during periods of market stress.Statistically significant negative values for β 1 and β 2 would indicate the presence of herding.

Table 1 .
Statistics data

Table 2 .
Herding and Turnover effect in Shanghai equity market , if the return on the market for time period t lies in the extreme upper 5% of the returns distribution, and D t L = 1, if the return on the market for time period t lies in the extreme lower 5% of the returns distribution.Dispersion has several different measurements such as CSSD, CSAD, HTSD, LTSD, HTAD and LTAD.

Table 3
Herding and turnover effect in Shenzhen equity market , if the return on the market for time period t lies in the extreme upper 5% of the returns distribution, and D t L = 1, if the return on the market for time period t lies in the extreme lower 5% of the returns distribution.Dispersion has several different measurements such as CSSD, CSAD, HTSD, LTSD, HTAD and LTAD.1.***Significance at 1% level.2. Value in parentheses is t value.

Table 4 .
Asymmetric test and turnover effect in Shanghai equity market Note: a. D t U = 1, if the return on the market for time period t lies in the extreme upper 5% of the returns distribution, and D t L = 1, if the return on