Choice of Information Learning: Evidence from Revisiting Price Discovery during the 2006 Dual Listing Craze in China

Extant studies have proposed information choice models to explain the home bias puzzle. In this paper, we explore information asymmetries among local and non-local investors which are captured by market microstructure variables to explain the repercussions of their information choices. We examine the short-run price dynamics, co-integrating relations, and price discovery processes of cross-listed Chinese firms. Using high-frequency data over the year 2006 of dual listing craze in history, we find that the local A-share market contributes to the price discovery for 68% to 75% of the firms, which is far more than the contribution of the non-local H-share market. Our results support the learning choice model and suggest that local Chinese investors may be endowed with a small home information advantage and their learning choices magnify information asymmetry. As such, the local stock market incorporates new information faster and contributes more to the price discovery of dual-listed firms in mainland China.


Introduction
Why do investors overinvest in domestic stocks? Researchers extensively study the determinants of home bias, including turnover (Tesar & Werner, 1995), transaction costs (Glassman & Riddick, 2001), corporate governance (Dahlquist et al., 2003), information costs (Ahearne et al., 2004), and the real exchange rate Theoretical Economics Letters risks (Fidora et al., 2007) of foreign assets versus those of domestic assets 1 . Recently, Nieuwerburgh & Veldkamp (2009) use an information choice model to show that learning choice is one factor that causes home bias. In this paper, we apply the information share model of Hasbrouck (1995) to explore how information is incorporated into the prices of dual-listed firms. An overinvestment in domestic stocks, which is characterized by a high percentage of domestic assets over the total market capitalization, suggests that investors may have forgone the potential benefits of international diversification. Even after taking transaction costs into account, Li et al. (2003) show that the diversification benefit is substantial. The evidence against an international portfolio theory creates the home bias puzzle. Is this bias a result of a concentrated ownership structure? Most companies in emerging markets are family-owned or closely-held, and controlling shareholders may enjoy the significant private benefits of control and may be reluctant to sell their shares; however, Dahlquist et al. (2003) use a world float portfolio to adjust for concentrated ownership structures, and they show that a home bias still prevails.
Prior studies argue that home bias could be caused by a prior information advantage. A counter-argument to this explanation is that information asymmetry and, consequently, home bias could be eliminated when investors can learn (Nieuwerburgh & Veldkamp, 2009). However, in some countries, such as China, citizens can only invest in their own country's stock markets, and only after 2006 have channels been created to allow them to invest overseas. Restrictions that Chinese investors face in their local stock market lower their opportunity costs for acquiring costly information of local firms that are listed on China's stock markets. This creates a comparative advantage for Chinese investors in obtaining information about local firms, which amplifies the information asymmetry.
Chinese mainland investors have no learning choice, in contrast to non-local investors outside mainland China, who possess an abundance of learning choices. Because information learning is costly, it is plausible that non-local investors would choose to free-ride on local investors' information about Chinese stocks (and profit in non-local stock markets) because of 1) the non-local investors' initial information disadvantage and 2) the magnification of the repercussions of this comparative information disadvantage by local institutional features 2 . As Nieuwerburgh and Veldkamp (2009) suggest, home investors are more likely to have a higher learning index in comparison to non-local investors. Therefore, Chinese investors who are investing in local stock markets should learn more than non-local investors in non-local stock markets for dual-listed firms, thus amplifying the Chinese investors' information advantage. We, therefore, expect that the price discovery, which we investigate via the information share model, should primarily be driven by the local stock markets. 1 See Demirgüç-Kunt and Huizinga (1995), Noronha et al. (1996), and follow-up studies on approaching the home bias puzzle. 2 Our argument also alludes to the ramifications of information market creation, as modeled by Veldkamp (2006). Y. C. Zhang et al. DOI: 10.4236/tel.2021.116074 1167 Theoretical Economics Letters This paper explores the process of how information is incorporated into prices for each pair of 17 firms that has one legal entity traded in the A-share market of the Shanghai Stock Exchange (SSE) and the other traded in the H-share market of the Hong Kong Stock Exchange (HKEx) during the 2006 craze of dual listing. By employing the information share model (Hasbrouck, 1995), we find that, in both statistical and economic terms, the A-shares of SSE firms play a more dominant role in the price discovery than the H-shares of the Hong Kong market. The A-shares contribute 68% to 75% of the price discovery, whereas the H-shares of HKEx firms contributed only 2% to 11%. Our findings support the learning choice hypothesis (Nieuwerburgh & Veldkamp, 2009) in that home investors, on average, choose to learn more about and buy more home assets, whereas non-local investors restrain from competing in the information market. As such, mainland stock markets incorporate new information faster and markedly contribute to price discovery. We find consistent results for each month of the year in our sample, which suggests that our major result is not driven by a seasonality effect, such as the January effect.
The local market dominance of price discovery may be due to infrequent trading of stocks or differences in trading mechanisms. This unique setup of A-H stock markets enables us to analyze local and non-local investors while controlling for firm fundamentals and microstructure differences. Our findings of the dominance of the local market in the contribution of price discovery cannot be driven by non-synchronous trading as the trade size and turnover in both markets are similar. Also, the local market dominance is not driven by the heterogeneity of trading mechanisms because both exchanges are pure order-driven markets.
The novelty of this paper is twofold. One contribution is that we show that the local market dominates the price discovery process in 2006 during China's stock market boom period. We examine several explanations for the home bias puzzle during the period from January 2006 to December 2006, when China's stock market experienced a dramatic price appreciation. First, this boom period would serve as a natural field experiment setting for understanding the investors learning choices. Second, we describe how quickly information flows between different markets during this boom period. Third, our results show that the home investors move first, and their behavior influences the non-local investors. Our findings not only help with explaining the home bias puzzle but also have important implications for portfolio diversifications among A and H shares. This information-based model helps to enhance the understanding of the phenomenon and loosen the international capital flow constraints. The investors could learn both domestic and non-local information. The non-domestic inventors could learn information from home investors as well.
The other novelty of our paper lies in explaining the home bias puzzle by investigating learning choice in two of the largest order-driven emerging markets with dual-listed firms. We contribute to the price discovery and the learning choice literature as follows. First, we are among the first to investigate the price discovery from the historical and theoretical perspective in two of the largest DOI: 10.4236/tel.2021.116074 1168 Theoretical Economics Letters order-driven emerging markets with dual-listed firms. Second, our analysis of the A-and H-share markets demonstrates that local investors dominate the price discovery of these dual-listed firms. Third, we provide direct evidence of the ability of local investors in exploiting local information by measuring their contribution to price discovery while previous studies focus on their portfolio returns. Fourth, we uncover the unobservable learning choice of investors through price discovery. Our results describe how quickly information could flow from the local market to the non-local market, and how trading behaviors of local investors causally affect non-local investors. Hence, the home bias can be explained by learning choice.
The rest of the paper is organized as follows. Section 2 reviews the related literature. Section 3 describes the institutional background and data. Section 4 explains the empirical framework. Our results are presented in Section 5. Section 6 concludes the paper.

Price Discovery and Information Choice
Home bias models are based on international capital flow constraints when this strand of literature initiates. But, as the capital flow constraints decrease among countries, the home bias persists, thus information-based models are used to explain the phenomenon. These models usually assume that information is not flowing freely across markets; hence, investors may choose to learn only information about their own country's stocks. Nieuwerburgh and Veldkamp (2009) develop an information choice model to predict how investors choose to learn about domestic and non-local information. They demonstrate that home investors optimally choose to learn domestic information even if they can learn non-local information. This is because home investors have a comparative advantage in collecting domestic information. If two different investors have their initial information advantage, the information asymmetry between the local investor and the non-local investor would be further magnified because of the comparative advantage of the former in collecting information. If the information share of domestic shareholders is substantially higher than that of non-local investors, the non-local investors will choose not to learn the domestic information; therefore, our results would support this learning choice model.
However, international finance theories provide inconclusive evidence of why individuals tilt their portfolios toward local stocks. Grinblatt and Keloharju (2000) show that foreign investors in these markets can earn abnormal profits and seem to be more informed. Recently, Seasholes and Zhu (2010) find that local investors do not generate abnormal returns, therefore implying that there is no information advantage for domestic investors when they trade. Ivković and Weisbenner (2005) find that individual households can exploit locally available, non-public information to earn excess returns. Massa and Simonov (2006)  Many studies investigate cross-border listings and price discovery in U.S. equity markets, such as Hasbrouck (1995), Smith and Sofianos (1997), and Eun and Sabherwal (2003), Stoll and Whaley (1990) and Chan (1992) examine the causal relationships between financial markets to analyze price discovery 3 . The price discovery of cross-border listings has yet to be fully explored, however, because of differences in the institutional settings of exchanges in various regions, such as the trading hours, exchange rate differentials, transaction costs, and price limits of individual markets. Examining cross-border listed companies in the Hong Kong and mainland stock markets mitigates most of the above differences. While many researchers study causal relations and price discovery processes in developed markets, such as the New York Stock Exchange (NYSE), relatively few studies are devoted to emerging markets. Chui and Kwok (1998) and Chan et al. (2008) both examine the correlation between the local A-share and foreign B-share markets. Wong et al. (2009) study the informed trading and liquidity in the SSE. The unique institutional setting in Chinese stock markets-A-share, B-share, and H-share markets-provides a natural setting to research the price information for single security with several prices 4 . In particular, we focus on Chinese A-share and H-share markets because the B-share market has a particularly low liquidity relative to its A-and H-share counterparts.

Forming the Synchronized Price Series for A-and H-Shares
While Hong Kong and mainland China are in the same time zone, the trading hours for the SSE/SZSE and HKEx are different. The SSE/SZSE have regular trading hours that range from 9:30 am to 11:30 am and from 1:00 pm to 3:00 pm, whereas the HKEx has trading hours of 10:00 am to 12:30 pm and 2:30 pm to 3:55 pm. Therefore, the common regular trading hours for both are 10:00 am to 11:30 am and 2:30 pm to 3:00 pm. We included only the transaction records for the trading hours between 10:05 am and 11:25 am in the morning session and between 2:35 pm and 2:55 pm in the afternoon session because these are the common trading hours. Stoll and Whaley (1990) argue that the opening transaction in higher volume stocks is likely to reflect an opening auction of trading. For this reason, Chan et al. (2008) delete the first and last 15 minutes in both the morning and afternoon sessions. Likewise, in our sample, we excluded the first and last 5 minutes in both the morning and afternoon sessions and paired the remaining trades in the common trading hours (10:05 am to 11:25 am in the morning session and 2:35 pm to 2:55 pm in the afternoon session). For each stock, we form the two price series using the minimal span procedure, as described in Harris (1995). That is, we form synchronous pairs of transaction prices for each stock sample between 10:05:00 and 11:25:00 and between 14:35:00 and 14:55:00. The advantage of this method is that it matches trades that are closest by time, which better represents the paired series than using a fixed time interval. We normalize the two price series for each firm in the same currency (RMB  (1)" corresponds to the trade size as measured by the number of shares and is equal to the average number of shares in each trade. "Trade Size (2)" corresponds to the trade size as measured by the turnover and is equal to the average turnover of each trade. "Daily Volume" equals the average number of shares that are traded each day. "Daily Turnover" equals the average turnover that is traded in each day.  Chan et al. (2007) in testing co-integration by testing the stationarity of the price differential across the A-share and H-share prices for a stock j in period t. We use an augmented Dickey and Fuller (1981) unit root test and the Dickey et al. (1984)  . We use an augmented Dickey and Fuller (1981) unit root test and the Dickey et al. (1984) limiting distribution for seasonal unit roots.

Information Share
In this section, we examine the short-term dynamics of the price series in each firm-month observation regarding the co-integrating relations of the A-share and H-share price pairs and the price discovery process using the Hasbrouck (1995) information share method. Hasbrouck (1995) breaks down the common factor variance, and new information in each market on the common factors contributes to the variance ratio to define the price discovery. The effective prices of A-shares and H-shares are facilitated by effective information that is obtained from the two markets, and fluctuations therein are a result of newly incorporated information. The newly incorporated information may arise from the existing A-share market, from the H-share market, or simultaneously from both markets. If it is simultaneously generated from the two markets, then the price discovery for assets of the two markets is a joint decision based on the information. When newly incorporated information of the market impacts the asset prices, the market prices of assets contribute to the price discovery. Hasbrouck (1995) identifies a positive and statistically significant correlation between the NYSE contribution to price discovery and its market share by trading volume by employing the information share model. He finds that the information share of the NYSE is around 92.7% and that the NYSE's contribution is significantly positively correlated with its trading volume market share in medium-size trades but not with shares in small or large trades. Our paper presents an econometric model of a firm that is listed in both the A-share and H-share markets.
We employ the information share model used by Hasbrouck (1995) as the measure of the relative contribution of the A-share price and the H-share price to the common price discovery process. Hasbrouck's starting point is that if a security trades in two different markets, then its prices in the two markets should be co-integrated. The shared information is a nonlinear function of the VAR coefficients and the disturbance covariance matrix. Accordingly, this study adopts Hasbrouck's (1995) route of addressing the issue of statistical significance by considering cross-sectional estimates of the information share and related measures. In our study, two price variables are closely related to single security transaction prices in A-share and H-share markets. Assume that there are n price variables that are closely related to a single security, such as transaction prices in n markets. These variables can be collected in a (n × 1) column vector where t e is a zero-mean vector of serially uncorrelated disturbances with a covariance matrix Ω, and ψ is a polynomial in the lag operator. Suppose that there exists a non-stationary autoregressive representation of order k for the price levels: and . ( ) The correspondence between the ECM and the level VAR is given by ( ) 1 1 and for 1, 2, , 1. (1) captures systematic differences in the prices.
It enters into the model in a nonlinear way with other parameters including α .
( ) The common trends representation of the model (2) contains a single random walk term that is common to all prices. The increment t e ψ is the component of the price change that is permanently impounded into the security price and is presumably caused by new information. The variance of this term is ψ ψ ′ Ω .
The proportion of this for market j relative to the total variance is defined as market j's information share:

Empirical Results
The results of this paper can be summarized as follows. A-shares contribute more in the intra-day price formation. This is consistent with the view of other studies (Demirgüç-Kunt & Huizinga, 1995;Noronha et al., 1996), which contend that segmented markets are less efficient, especially for non-local investors, because of the information transmission. Our result supports the theoretical prediction of Nieuwerburgh and Veldkamp (2009) that the home market may have superior access to information.  (2)" corresponds to the trade size as measured by the turnover and is equal to the average turnover of each trade.
"Daily Volume" equals the average number of shares that are traded in each day.
"Daily Turnover" equals the average turnover that is traded in each day. Table 2 reports the result of the co-integration test of the two price series. We follow Chan et al. (2007) in testing co-integration by testing the stationarity of the price differential across the A-share and H-share prices. We use an augmented Dickey and Fuller (1981) unit root test and the Dickey et al. (1984) limiting distribution with seasonal unit roots to account for seasonality. Table 2 reports the number of co-integrated pairs that are significant at the 5% level out of the 17 cross-listed pairs for the month in question. We find that the price series of 133 pairs out of 199 are cointegrated. The result suggests that the two price series are cointegrated.

Price Discovery Analysis
The cross-sectional averages of the daily coefficients of the efficient price for each firm month are reported in Table 3. The common factor is the implicit efficient price of the security, and it is defined as the random walk component of the price from the markets, which is a measure of the information intensity of the efficient price process. The comparative analysis results in Table 3 show that in each month the A-share always exhibits a one-order-of-magnitude higher coefficient for the common factor.
Information share estimates in each month are reported in Table 4. Means, medians, and standard deviations (across the day) are reported by month. The A-share price possesses by far, the dominant information share (greater than 70% of the mean and 97% of the median for all months) of the price discovery.
The H-share price displays approximately a single-digit percentage. This supports our hypothesis that local investors dominate the price discovery process of

Liquidity Implications
We next address the fundamental question: how is our price discovery result important to market liquidity? O'Hara (2003) mentions that liquidity and price discovery are related but are not the same. We are among the first to investigate the relation between a firm's information share and its market liquidity by comparing market liquidity in high and low information share subsamples. Following the literature on market liquidity (Hasbrouck, 2009;Goyenko et al., 2009), we employ three measures of market liquidity in Table 5: 1) the Amihud illiquidity ratio (Amihud, 2002), which is defined as the monthly average of the daily price change per dollar of the trading volume (all figures are multiplied by 106 for normalization (Amihud, 2002)); 2) the percentage of zero-return trading days (Lesmond et al., 1999); 3) the percentage of zero-return trading days with a positive volume (Lesmond et al., 1999). Information shares are computed following Hasbrouck (1995).  Note: Significance at the 1% level is indicated by ***.
for firms in high and low information share subsamples from January to December 2006. For each month, a firm belongs to the high (low) information share group if its information share is higher (lower) than the cross-sectional median for the month in question. Satterthwaite t-statistics with unequal variances are estimated and reported in the last column. We find strong evidence that firms with high information share are more liquid. The average Amihud ratio is significantly smaller for firms with higher information share (e.g., the corresponding t-statistic equals 3.077) , implying that, the price impact per dollar traded is lower. We find the average percent of zero-return days is significantly higher for firms with high information share. The result using the percentage of zero-return days with positive volume is also significant at 1%. Overall our results suggest that firms with higher information share have more liquid stocks.

Conclusion
This paper provides evidence that the local market (i.e., China's A-share market) dominates the non-local market (i.e., the Hong Kong H-share market) in price discovery processes of dual-listed firms in Mainland China. The results support the theory of Nieuwerburgh and Veldkamp (2009) that home bias can be explained by their learning choice model when there is a comparative advantage in the local investors' information collection. Our results also support the finding by Veldkamp (2006) about the existence of markets of information and suggest that it may be more cost-efficient for local investors to collect information for local firms. This creates a comparative advantage for Mainland Chinese investors in obtaining information about dual-listed firms and trading their A-shares, which later creates an even larger information asymmetry. The dominance of the information shared in the local market suggests that local investors may choose to learn more than non-local investors for dual-listed stocks, thus amplifying local investors' information advantage. Our results further reconcile the mixed findings in the literature by providing a more direct test-we examine different contributions of local and non-local investors to the price discovery process, af-