Herding Behavior between Chinese and Hong Kong Stock Market—Based on Shanghai-Hong Kong Stock Connect Policy

This paper investigates herding behavior between Chinese A-share and HK stock market based on Shanghai-Hong Kong Stock Connect. This research innovatively details the difference in herding behavior between two markets bidirectionally. By applying CCK model and quantile regression, we examine herding behavior between Chinese and HK stock market in asymmetric situa-tions from the perspective of net capital flow of Shanghai-Hong Kong Stock Connect. Our findings suggest that the impact of HK share on A share’s herding behavior increases after Shanghai-Hong Kong Stock Connect starts trading. To some extent, the direction of net capital flow of Shanghai-Hong Kong Stock Connect acts as a powerful indicator for investors, which skillful-ly deflects Chinese stock market. Besides, the way in which HK’s stock market influences that of Chinese mainland has undergone a transformation.


Introduction
Classical Economic theory assumes rational people and complete information.
However, there is asymmetry information and information incompleteness in real financial market. Human cognitive bias and bounded rationality give rise to various economic and financial anomalies. Therefore, understanding investors' decision-making behavior becomes a research concern. As the globalization of capital market, rising uncertainty tends to lead to increased herding behavior and sharp fluctuations of asset prices. This can affect the stabili- Christie and Huang (1995) firstly propose the return dispersion indicators (CSSD, cross-sectional standard deviation of returns) to measure herd behavior in the market.  Five or one percent in the lower and upper tail of market return distribution is defined to be extreme cases. Christie and Huang (1995) (henceforth CH model) suggest that when facing extreme market fluctuation, investors tend to ignore their personal judgment. Thus, their investment behavior tends to be consistent with aggregate market. In this way, volatility of individual stocks and market tends to converge. In extreme cases, the degree of market dispersion equals zero. When investors' investment behavior is rational, market dispersion is proportional to the market return. However, the CH model can only recognize herd behavior in extreme market conditions, the validity of the identification is limited.

Literature Review Ease of Use Empirical Studies on Herding Behavior
In addition, even in extreme market situations, investors may be faced with the same information set and hence have similar decision-making process.
In order to overcome these limitations of CH model, Chang et al. (2000) (henceforth CCK model) constructs the cross-section absolute deviation of returns indicator (CSAD) to effectively identify the herd behavior in stock market. This model is derived from the capital asset pricing model (CAPM).
According to Goyal and Santa-Clara (2003), there is a significant positive relationship between cross-sectional dispersion of individual stock earnings and market return. Hwang and Salmon (2004) provide evidence that when herd behavior occurs, the relationship between return and risk changes in stock market (i.e. the coefficient of CAPM model β changes). However, as CCK method assumes that the β remain constant over time under CAPM framework. Therefore, Hwang and Salmon (2004) construct the cross-sectional dispersion of beta coefficient indicator (HS model), taking account of dynamic relationship between market return and risk in a volatile market. Compared to CCK model, HS model enables to theoretically distinguish the real and pseudo herd effect in stock market. However, it hasn't been used widely due to computational complexity.
After Christie and Huang (1995) firstly proposed CSSD dispersion indicators and Chang et al. (2000) gave a theoretical analysis of testing herd behavior by CSAD dispersion indicators based on CAPM, an increasing amount of literature began to pay attention to herd behavior in global stock markets.
One strand of literature investigates herd effect on the institutional level such as behaviors of mutual fund. The other strand of research focuses on the herd behavior on the level of aggregate market. Chiang and Zheng (2010) confirms the existence of herd behavior in developed countries (except for the United States) and Asian markets and further discovers the asymmetry effect of herd behavior in both rising and declining markets. Herding in emerging markets has aroused research interests over the past decade. Demirer and Kutan (2006) employs the CH method to study herd behavior of China's stock market from the firm-level and industry-level perspective and found that herd behavior doesn't exist. Subsequently, Tan et al. (2008) provided evidence of herd behavior in China's stock market, using data of 87 dual listed stocks in Chinese A and B-share market from 1994 to 2003. The empirical result of this research is not consistent with Demirer and Kutan (2006). Compared to Tan et al. (2008),  further expanded the sample range by using data of all A and B-shares ranging from 1996 to 2007 and found that herding exists in both Shanghai and Shenzhen stock market. No evidence of herding was founded with Chinese B-share market, which is not consistent with Tan et al. (2008). In line with methodology of Tan et al. (2008), Lao and Singh (2011) investigate the herding behavior in Chinese and Indian stock markets and suggest that the crowd movement exists in both. Compared to the results of Tan et al. (2008), the magnitude of herd behavior is higher. Yao et al. (2014) investigate herding behavior from both industry-level and other sub-level perspectives by incorporating monthly data of EPS, market-to-book values into CCK model. It is founded that investors exhibit herd behavior in Chinese stock market, particularly in B-share market which is inconsistent with Luo and Schinckus (2015a) and Tan et al. (2008). Its possible explanation is multicollinearity of traditional CCK model. Although the CCK model (2000) has a solid theoretical basis, the high-level of autocorrelation of CSAD may lead to nonlinear relationship between dispersion indicators t CSAD and equally weighted market return , m t R . The two explanatory variables m R and 2 m R may also exhibit strong multicollinearity. In order to overcome this problem, Yao et al. (2014) Luo and Schinckus (2015a) studies herd behavior in asymmetric context and under extreme market conditions. Results suggest that investors' crowd movement of Chinese stock market in the bullish market is more obvious than that in rising market, and herding is more obvious in extreme cases.
The existing literature on herd behavior mainly adopts CCK model. However, empirical results of the herding behavior in China's stock market are inconsistent and greatly influenced by the choice of sample period. Moreover, previous studies mainly focus on the herd behavior in individual markets.
Although there are established studies on herd behavior in worlds' main stock market, relatively few studies have explored the cross-market herd effect, especially the bidirectional herd behavior between A-share and Hong Kong stock market. Luo and Schinckus (2015b)

Modified CCK Model
This paper employs CCK model proposed by Chang et al. (2000), using the cross-sectional absolute deviation (CSAD) indicator to detect the herd behavior. According theoretical analysis by Chang et al. (2000), CSAD and equally weighted market return N denotes the number of stocks in the market, , To examine cross-market contagion of herd behavior linkage between the A-share and HK stock market, we modifying CCK model proposed by Chang et al. (2000). The modified CCK model is in line with the methodology of Luo and Schinckus (2015a), which is shown as Equation (4) (4), the significant negative coefficient 3 γ indicates that herding occurs in the Chinese A-share market, whereas a significant positive 4 γ a certain extent of dependence of Chinese A-share market over Hong Kong stock market. A significant negative coefficient 5 γ suggests that herding behavior of Chinese A-share market is influenced by Hong Kong stock market. Similarly, the meaning of coefficients for Equation (5) can be deduced in the same manner.
As the launch of the Shanghai-Hong Kong Stock Connect, its net capital flow has become an important market reference for both institutional and individual investors, affecting investors' sentiment. Existing literature shows that there is asymmetry of herd behavior in stock market (Christie & Huang, 1995;Hwang & Salmon, 2004;Bekiros et al., 2017). The magnitude of herding amplifies generally in bearish market.
In order to investigate the asymmetry of the herding behavior in the market, we further modify CCK model by adding a dummy variable. The value of dummy variable is set up to the direction of net capital flow of Shanghai-Hong Kong Stock Connect. If net capital flows into Shanghai stock market (northbound), In Equation (6), a significant negative 3 γ indicates existing of herding behavior in Chinese A-share market under a northbound net capital flow of Shanghai-Hong Kong Stock Connect (i.e., the net capital flows into Shanghai stock market), otherwise the herd behavior does not exist. A significant negative 4 γ suggests that there is herd behavior in Chinese A-share market under a southbound net capital flow of Shanghai-Hong Kong Stock Connect (i.e., the net capital flows into HK Stock market). Similarly, the meaning of coefficients in Equation (7) can be deduced in a similar manner.
To further investigate the asymmetric cross-market contagion of herding, this paper divides the sample according to the capital flow direction Shanghai-Hong Kong Stock Connect. Equation (8) to Equation (11) is employed to detect the herd behavior under different market conditions.

Quantile Regression
Quantile regression model was pioneer by Koenker & Bassett (1978), which enables to estimate the variability in the effects of a set of explanatory variables X on explained variable Y across the distribution. Therefore, quantile regression enables to investigate the sensitivity of herd behavior under different quantile of stock return dispersion.
Consider a random variable y, its right continuous distribution function is denotes the probability and Loss function is defined as below, where I (Z) is an indication function and According to Koenker and Bassett (1978), weighted absolute deviation of α can be minimized only when Quantile regression can be estimated by weighted least absolute deviation (WLAD): where ( ) 0,1 τ ∈ and t ω is the weight function. Assume that the dependent variable Y is a linear expression of a matrix composed of k independent variables, and the regression expression of the τ quantile is where X, β are k × 1 order column vectors, and ( ) τ β is the estimator of the quantile regression coefficient.
For the conditional mean function , the parameter can be estimated as below Quantile regression estimates parameters by minimizing the weighted absolute error sum. The weight depends on the difference in quantile points. It enables to estimate the relationship between the explanatory variables and the explanatory variables at any particular quantile, which is more flexible than the general least squares estimation, and provides a better method for testing the herd behavior.

Data
Data of stock return is collected from TongDaxin Stock Trading Database, including closing prices for all listed companies in A-shares and H-shares and covering a period from 2006/01/05 to 2017/02/27 on a daily basis. TongDaxin is a professional stock trading platform in China, which is adopted by major Chinese securities companies. Stock data can be exported via TongDaxin. We divide the sample period into two segments for comparative analysis: 2006/01/05-2014/11/14 and 2014/11/17-2017/02/27, which represents the sample before and after the launch of the Shanghai-Hong Kong Stock Connect. Both the two sample periods of Chinese A-share market experienced a complete bull to bear transition. Table 1 presents the main statistical characteristics such as mean and standard deviation of dispersion indicator (CSAD) and equally weighted markets return in both A-share and HK stock markets. γ in the first period is larger than the second period, which means a greater herd behavior. In addition, 3 γ is neg-ative and 5 γ is positive in the whole sample period, which indicating the herd behavior of A-shares is affected by the H-shares, and there was a certain linkage between of the return dispersion in both A-and H-shares. CSAD means the average dispersion between a single stock return and the market equity return.   Table 4 shows that the absolute value of 3 γ increases as the quantiles However, considering the dependent variable (CSAD) in H-shares, 3 γ and 5 γ are significantly positive in the whole sample period (Table 2). It shows that as a relatively mature capital market, there is no herd behavior in H-share market. In addition, assuming that there is herd behavior in H-shares market, which is not affected by A-shares market, while 4 γ is significantly positive, it shows that there is a connection of CSAD between H-shares market and A-shares market ( 4 0.385 γ = before launch of Stock Connect, 4 0.866 γ = after launch of Stock Connect and 4 0.386 γ = during whole sample period).    Table 6 and Table 7 show that, with the increase of the quantiles, the absolute value of 5 γ increases after the launch of the Shanghai-Hong Kong stock connect (in Table 6 Table 7.

Herd Behavior for Dual-Listed Stocks within Chinese A or H-Share Market
Comparing the herd behavior for all stocks between A-& H-shares market (5.1.1) and that for dual-listed stocks in the Shanghai-Hong Kong stock connect, the result in Table 8 report some interesting information.

For the A-H dual-listed stocks, A-shares market does not show a herd beha-
vior before the launch of the Shanghai-Hong Kong Stock Connect ( 3 γ is significantly positive). In Table 8, 3 γ shows a positive value of 3.120 during the whole sample period from 2006/01/05 to 2014/11/14. From the perspective cross-market herd behavior, the significantly negative 5 γ (−2.496, in Table 8) shows that herding exist between the two markets during the whole sample period. This can be attributed to the fact of large market capitalizations of  Table 2. *** denotes significance at the level of 0.001, ** the level of 0.01 and * the level of 0.05.   γ a significantly negative of −1.814 in Table 9), which further confirms the results and analysis of 5.3.1 and 5.3.2. Note: The empirical analysis above is based on Equation (6) and Equation (7). According to Table 8, ere is no herd behavior within and between markets. But since the launch of the Shanghai-Hong Kong stock connect, 4 γ increases significantly from 0.569 to 0.977, which indicates an enhanced linkage between the two stock markets. Table 9 shows that only the A-share market has the herd behavior within the market, while there is no herd behavior within H-share market. For all stocks of A-shares and the Shanghai-Hong Kong stock connect, the absolute value of 4 γ ( 0 D = means the net flow of the Shanghai-Hong Kong stock connect is southbound) is slightly larger than 3 γ ( 1 D = means the net flow of the Shanghai-Hong Kong stock connect is northbound). In Table 4, the absolute value of 3 γ equals 4.093 and 4 γ equals 4.252. It means that herding of Chinese A-stock market is greater when there is a net capital outflow of Shanghai-Hong Kong Stock Connect. It is in accordance with the theory of loss aversion and expected utility proposed by Tversky & Kahneman (1979) in the Behavioral Finance. That is, when people face equal gains and losses, losses cause greater mood swings. Table 10 shows that 3 γ and 5 γ are significantly negative, when the fund of Shanghai-Hong Kong stock connect flow out of the A-share market, the herd behavior within the market is relatively strong. In Table 10, the absolute value of 3 γ increases from 4.778 to 5.296. But the herd behavior between the A-& H-shares market weakens, the absolute value of 5 γ decreases from 3.977 to 2.072 (Table 10).  Figure 1 describes the net capital inflow of Shanghai-Hong Kong Stock Connect after the launch of Shanghai-Hong Kong Stock Connect. Figure 2 presents the composite index of Shanghai stock market after launch of Shanghai-Hong Kong Stock Connect. Figure 1 and Figure 2 shows, when there are large net inflows for the Shanghai-Hong Kong Stock Connect, the A-share market moves wildly. On this occasion, the fund of Shanghai-Hong Kong stock connect flow out of the A-share market, investors in A-share market may panic. Reduce the attention to the external H-share market, the herd behavior within the market would increase. Luo & Schinckus (2015a) found that the herd behavior of A-shares in the riskier market was more pronounced than that of the rising market, our study is consistent with the studies of Luo & Schinckus (2015a).

Conclusion
In this paper, we expand the CCK model and use quantile-regression model to test herd behavior between A-& H-shares market. We make the test from the perspective of the whole market, listed stock in Shanghai-Hong Kong stock connect and A-& H-shares. The results show that there is a herd behavior before and after the launch of the Shanghai-Hong Kong Stock Connect, and there is a linkage between the two markets in return dispersion. In addition, the results of the quantile-regression show that, since the launch of the Shanghai-Hong Kong stock connect, the influence of HK stock market on herd behavior of Chinese stock market improves with the increase of the quantiles. During whole sample period (2006/01/05-2016/02/27), the influence of HK stock market on herd behavior of Chinese stock market decreases with the increase of the quantiles. Thus, we think the capital flow of Shanghai-Hong Kong stock connect has a well-leveraged effect on A-shares investor, and the mode of influence of HK stock market on Chinese A-shares market has changed.  For the dual-listed stocks in A-& H-shares market, A-shares market does not show a herd behavior before the launch of the Shanghai-Hong Kong Stock Connect. However, there is a cross-market herd behavior for Chinese A stock market. This can be attributed to the fact that market capitalization of dual-listed stocks is large, and thus investors are mainly composed of institutions.
As a relatively mature capital market, there is no herd behavior in H-share market. Even assuming that there is herd behavior in H-shares market, it is not affected by A-shares market. But there still has a linkage in the yield dispersion. For the asymmetry test of herd behavior in the market, the results show that the herd behavior of Chinese A-stock market is greater when there is a net capi-between the A-& H-shares market weakens. It is accordant with the theory of loss aversion.
The herding behavior of Chinese A-share and HK stock market can be extended in several directions. One possibility is to investigate the mechanism of how Shanghai-Hong Kong Stock Connect enhances the herding behavior in Chinese A-stock market, such as the changes of investor structure. Another possibility is to study transition of linkage mechanism of Chinese A-share and Hong Kong stock market after the launch of Stock Connect.

Conflicts of Interest
The author declares no conflicts of interest regarding the publication of this paper.