Margin trading and price efficiency: information content or price-adjustment speed?

2.1 A brief look at margin trading

Margin trading is permitted in most stock exchanges around the world. The Securities Exchange Act of 1934 allowed margin trading on exchange-listed securities; the Over-the-Counter Market Act of 1968 allowed margin trading on some over-the-counter (OTC) securities. In Asian countries, the Tokyo Stock Exchange launched margin transactions in 1951, the Taiwan Stock Exchange did so in 1974, and the Hong Kong Stock Exchange did so in January 1994. In March 2010, the CSRC lifted its ban on margin trading for the first time.

Margin-trading activities make up a large proportion of total trading amount. For example, NYSE customers borrowed US$377 billion to purchase securities in May 2013 (accounting for approximately 2.25 percent of the total market capitalisation of listed companies) and this number increased to US$461 billion in December 2015 (accounting for approximately 2.51 percent of the total market capitalisation).5 The margin-buying debt in China was RMB213 billion in May 2013 (accounting for approximately 0.87 percent of the total market capitalisation) and sharply increased to RMB1,174 billion at the end of December 2015 (accounting for approximately 2.81 percent of the total market capitalisation).6

2.2 Debate on the effect of margin trading on price efficiency

Many studies document a positive relationship between margin trading and stock liquidity (Seguin, 1990; Seguin and Jarrell, 1993; Chang et al., 2014). However, margin traders are often considered as potentially informative speculators and are blamed for increasing excess volatility of stock prices. In practice, margin requirements are officially changed frequently, especially when the market is unstable.7 For example, the Tokyo Stock Exchange changed margin requirements 63 times between 1970 and 1990. Likewise, margin requirements changed 23 times in the United States between 1934 and 1974 (Hardouvelis, 1990). The U.S. government claimed, ‘It has long been recognised that margin requirements, through leverage, affect the volume of speculative activity’ after the crash of October 1987 (U.S. Presidential Task Force on Market Mechanisms, U.S., 1988). Furthermore, regulatory bodies have tended to propose higher margin requirements to discourage speculators. Similarly, many professionals and researchers attribute the rapid changes in the 2015 Chinese stock market to margin-trading activities.

These arguments suggest that margin trading affects many aspects of the stock market, such as liquidity, stability and price efficiency. Numerous studies examine the effect of margin trading on price efficiency, but do not reach consensus. Some studies support the view that margin trading and lower margin requirements increase price efficiency. Seguin (1990) examined the effect of adding OTC issues to the list of marginable securities and suggested that margin eligibility adds value by increasing the flow of information and enhancing its depth. Alexander et al. (2004) documented that the information content of trades increases after the level of margin requirements decreases, suggesting that margin trading improves market quality. Chen et al. (2016) also suggested that prices become significantly more informative when stocks are allowed to be traded on margin in China.

In contrast, some researchers argue that margin trading may destabilise the market and decrease market efficiency (Chowdhry and Nanda, 1998; Rytchkov, 2014). Hardouvelis (1990), Hardouvelis and Peristiani (1992), and Hardouvelis and Theodossiou (2002) documented that higher or increasing margin requirements are associated with lower stock-price volatility, lower excess volatility and smaller deviations in stock prices from their fundamental values. Thurner et al. (2012) argued that when funds are allowed to leverage, a margin call resulting from a downward price fluctuation causes them to sell into an already falling market, amplifying the downward price movement.

2.3 Background of margin trading in China

Both margin trading and short selling were strictly prohibited in China at the opening of the Chinese securities market. The constraint was lifted as of 31 March 2010, when the CRSC launched a long-awaited pilot programme allowing 90 stocks on a designed list to be purchased on margin and/or sold short ‘to enlarge the supply and demand of funds and securities and to increase the trading volume to some extent, thus leading to active liquidity on the securities market’ (CSRC, 2010).8 Later, four major additional revisions of the pilot list occurred on 5 December 2011, 31 January 2013, 16 September 2013 and 22 September 2014.9 Table 1 illustrates the timeline of these reforms. The underlying security list was finally expanded to cover 900 stocks and 15 exchange-traded funds (ETFs) in September 2014, covering more than one-third of the total listed stocks in China.

According to the administrative rules promulgated by the CSRC, only ‘qualified’ investors can purchase stocks on margin or sell stocks short. Such investors must meet certain requirements. For example, in most securities companies, qualified investors must (i) possess assets of at least RMB500,000 and (ii) have a trading history of longer than 1.5 years. As in Japan and Taiwan, the margin-trading and short-selling business in mainland China seems to cater to individual investors rather than to institutions. However, the cost of margin trading is quite high in China. At the beginning of the pilot programme, the interest rate for margin trading, approximately 8.6 percent per year, is much higher than the yield of 1-year treasury bonds (ranging from 2.60 to 2.85 percent), making margin trading costly for investors. In contrast, D'avolio (2002) reports that the value-weighted loan fee is only 0.25 percent in the United States, and only 9 percent of stocks have a loan fee higher than 1 percent. Although margin trading is high cost and imposes strict requirements relative to investors’ qualifications, margin-buying debt is growing quickly, reaching RMB1,174 billion by the end of 2015, as shown in Figure 1.

China's margin-trading and short-selling programme has several significant features. First, only stocks on the designated list can be purchased on margin and/or sold short. This restriction provides an excellent opportunity to examine the effect of this pilot programme and the impact of margin-trading activities on price efficiency. Additionally, as Chinese investors are new to the short-selling mechanism, unsophisticated investors may choose to steer clear of short selling. Thus, margin trading is much more active than short selling (Li et al., 2016), even though stocks on the list are eligible for both. Figure 2 illustrates the monthly margin-trading and short-selling activities from April 2010 through December 2015. It reveals that margin trading has become increasingly popular over the years. For example, the cross-sectional average of the monthly margin-trading amount scaled by trading amount was 1.73 percent in December 2010 and increased sharply to 20.50 percent in December 2015. Figure 2 also shows that there is far less short-selling activity than margin-buying activity. The cross-sectional average of monthly short-selling volume scaled by trading volume was < 2.3 percent from April 2010 to December 2015.

3.1 Data

We retrieve information on the designated or revised stock list from the Shanghai Stock Exchange and Shenzhen Stock Exchange, and margin-trading amounts are available from March 2010 through December 2015.10 We match daily margin-trading amount data with the China Stock Market Trading Research database to obtain each stock's closing price, trading amount, market capitalisation, turnover, institutional ownership, and high-frequency trade and quote data. We include only domestic stocks in our analysis, but exclude eligible ETFs from the list.

3.2 Measures of price efficiency

The literature uses various price-efficiency measures (Seguin and Jarrell, 1993; Morck et al., 2000; Chordia et al., 2005; Boehmer and Wu, 2013; Chang et al., 2014; Chen et al., 2016). Some measures (e.g. pricing error, absolute value of quote midpoint return autocorrelations, idiosyncratic volatility and R²) reflect the information content of stock price, whereas other proxies (e.g. cross-autocorrelation between stock return and lagged market returns, price delay and trading volume) represent the speed of price adjustment to public news. However, there is very little evidence on which measure can fully capture information content or price-adjustment speed. Thus, we include most of these measures in this study to investigate the impact of margin trading on divergent aspects of price efficiency as thoroughly as possible.

3.2.2 Measures of price-adjustment speed

We use two different types of proxies to measure price-adjustment speed: price delay (Hou and Moskowitz, 2005) and the cross-autocorrelation between stock returns and lagged market returns (Bris et al., 2007). These two proxies capture the speed with which marketwide information is impounded into stock prices.

Introduced by Hou and Moskowitz (2005), price delay can be used to measure the speed of stock price adjustment to marketwide information.13 A smaller price delay implies that the stock price adjusts more quickly to marketwide information. As this paper covers only 5 years of data, using annul price-delay measure would limit the precision with which we can investigate the effect of margin trading on price delay. We follow Boehmer and Wu (2013) and create analogous monthly measures using daily data. We regress an individual stock's daily returns on contemporaneous and 4 days of lagged market returns over 1 calendar month. Specifically, we run the following regression:

(1)

where r_i,t is the return on stock i and R_m,t is the value-weighted market return on day t. If the stock responded immediately to public news, β_i,1 would be significantly different from zero, but none of the δ_i,−n would differ from zero. However, if stock i's price responded with a lag, some of the δ_i,−n would differ from zero. Based on this, we compute two price-delay measures. The first measure of price delay (denoted as Delay1) is computed as one minus the ratio of the R² from Regression (1) restricting δ_i,−n to zero over the R² from Regression (1) with no restrictions. To distinguish between shorter and longer lags, we also use another measure (denoted by Delay2):

(2)

Both of these two price-delay measures capture the portion of individual stock return variation that is explained by lagged market returns. A larger price delay implies that more price variation is captured by lagged returns, and it takes more time to incorporate marketwide information.

Bris et al. (2007) use daily cross-autocorrelation between current stock returns and lagged market returns as an alternative proxy for price-adjustment speed. The cross-autocorrelations are equivalent to estimating the regression coefficients of individual stock returns on lagged market returns, as in Hou and Moskowitz (2005). Thus, the cross-autocorrelation coefficient is an alternative way of standardising the measure of price delay and can be used to measure the speed of individual stock price adjustment to common factor information. We adopt their approach and use |CR| to denote this proxy. If the stock price adjusted quickly to marketwide information, there would be a small |CR|. Overall, a higher |CR| indicates a lower price-adjustment speed.

3.3 Regression models

3.3.1 Testing the effect of implementing margin trading on price efficiency

China launched the pilot programme to lift the ban on margin trading in March 2010 and, as mentioned earlier, revised the list on 5 December 2011, 31 January 2013, 16 September 2013 and 22 September 2014. These changes provide a quasinatural experiment, allowing us to use a DID approach to verify the effect of the implementation of margin trading on price efficiency. The based event dates are the dates on which margin trading is initiated or changed, including the date of programme initiation and the dates of the four major revisions of the eligible list.

Following Chang et al. (2007), we use 1 year before and after each event date as the pre- and postevent windows, respectively.14 For each event, stocks newly added to the list are considered to belong to the treatment group, and the control group is generated using the propensity scores matching (PSM) method. In this method, each treatment stock eligible for margin trading is matched with a unique noneligible stock that is traded on the same stocvk exchange, and the scores are computed by turnover, market capitalisation and stock price.15 Specifically, we estimate the following DID regression:

(3)

where PriceEfficiency_i,t is price-efficiency measure of stock i at event t, which can be the measure of information content or the proxy for price-adjustment speed; POST_i,t is a dummy variable that equals one after the based event date and zero otherwise; and LIST_i,t is the dummy variable for treated stocks that equals one if stock i is eligible for the programme and zero otherwise. In this specification, the coefficient of the interaction term between POST_i,t and LIST_i,t (i.e. β₃) is used to detect the effect of the pilot programme on price efficiency. A statistically significant β₃ implies that the implementation of margin trading does affect price efficiency.

Following prior studies (Bris et al., 2007; Boehmer and Kelley, 2009; Boehmer and Wu, 2013; Chang et al., 2014), we also consider several control variables (Controls_i,t) that may be associated with price efficiency. We include measures of share price, market capitalisation, turnover and institutional ownership as control variables in our model. Specifically, we use the natural logarithm of volume-weighted closing price (denoted as LnPrice) to control for stock-price variation that may affect price efficiency. As it is easier and more precise to value bigger and more active stocks, we also include both the market value of the stock (LnCap), calculated as the natural logarithm of individual stock's market value divided by 1 billion, and turnover (Turnover), measured by trading volume scaled by shares outstanding. Moreover, Boehmer and Kelley (2009) find that institutional ownership may be related to price efficiency. Therefore, we control for this effect using the fraction of shares owned by institutional investors at the end of the previous quarter (denoted as Insti).

3.3.2 Testing the effect of margin-trading activity on price efficiency

The two stock exchanges in China publish the daily margin-trading amount of each eligible stock. This information is available for each stock from March 2010 through December 2015, which makes it possible for us to further investigate how margin-trading activity affects the degree of information content and the speed of price adjustment. We regress monthly measures of price efficiency on the proxy for margin-trading activity and other control variables. Specifically, we use the following model to investigate the impact of margin-trading activity on price efficiency:

(4)

where PriceEfficiency_i,t is the efficiency measure of stock i at month t, which can be the measure of information content or the proxy for price-adjustment speed. Following Boehmer and Wu (2013), we compute the monthly margin-trading activity (Margin) as the monthly average of daily ratio of daily margin-buying amount to the stock's daily trading amount. In addition, we include control variables in Model (3) to control for the effect of other determinants. Furthermore, to address the concern that current efficiency may be affected by past price efficiency, we include the lagged dependent variable (i.e. PriceEfficiency_i,t−1) as a control variable in our regression model. As the relevant proxies of price efficiency and margin-trading activity are available or can be computed at the monthly frequency, these tests are quite powerful.

4.1 Descriptive statistics

We estimate high-frequency measures of price efficiency using quota data for each stock and compute low-frequency measures using daily data over a single calendar month. For brevity, we average all of the variables for each stock over pre- or postevent windows, yielding a sample with 3,360 observations.16 Panel A of Table 2 reports the summary statistics. Both our information-content measures (i.e. PriceErr, |AR15| and IVOL) and price-adjustment speed proxies (i.e. Delay1, Delay2 and |CR|) exhibit a positive mean value and substantial cross-sectional variation.

We then perform paired t-tests to examine the changes in price-efficiency measures and firm characteristics around these event dates. Panel B of Table 2 shows the comparison results. We find that after the bans are lifted, PriceErr and |AR15| for both the pilot and nonpilot stocks decrease significantly, whereas IVOL increases significantly. This indicates that, overall, stock prices deviate less from their fundamental values and incorporate more firm-specific information after the pilot programme. In contrast, the price delay (Delay1 and Delay2) decreases significantly for the pilot stocks, but increases significantly for the nonpilot stocks, suggesting that the price adjusts quicker for eligible stocks after the pilot programme. |CR| increases significantly for both the pilot and nonpilot stocks after this pilot programme.

We also present Pearson correlation coefficients between different efficiency measures in Panel C of Table 2. PriceErr is positively correlated with |AR15|, and the Pearson correlation coefficients between IVOL and the first type of information-content measures (PriceErr and |AR15|) are significantly negative, implying that these measures share a common component of information content. Moreover, the Pearson correlation coefficients between the different proxies of price-adjustment speed, ranging from 0.267 (between Delay1 and |CR|) to 0.965 (between Delay1 and Delay2), are significantly positive, indicating that these measures also have a common component of price-adjustment speed. Information content and price-adjustment speed seem to be negatively correlated in general. These findings emphasise the need to decompose price efficiency.

4.2 Regression results

To reduce the interference of outliers, we winsorise all of the variables in our estimation at the 1st and 99th percentiles.17 We use event dummies to control for time-fixed effects and use the robust ordinary least squares approach to estimate Model (3). Table 3 presents these estimation results. Columns 1 to 3 report the regression results for measures of information content (PriceErr, |AR15| and IVOL, respectively), whereas columns 4 to 6 present the regression results for the price-adjustment speed proxies (Delay1, Delay2 and |CR|, respectively).

As shown in Table 3, the coefficients of POST are statistically significantly negative in columns 1 and 2, but statistically significantly positive in Column 3, suggesting that the overall information content in the Chinese stock market increases after the pilot programme. The coefficients of POST are statistically significantly positive in columns 4 to 6, indicating a marketwide decrease in price-adjustment speed after the pilot programme.

Additionally, the coefficient of LIST in Column 1 is statistically significantly negative, implying that eligible stocks deviate less from the random walks and thus incorporate more information. The coefficients of LIST are statistically significantly positive in columns 4 and 5, suggesting that the eligible stocks may incorporate information more slowly.

The interaction term between POST and LIST has statistically significant coefficients in most of these specifications. The coefficient of the interaction term is statistically significantly positive in columns 1 and 2, and is statistically significantly negative in Column 3. These results indicate that implementing margin-trading pilot programme is associated with higher pricing error, greater absolute value of intraday return autocorrelations, and lower idiosyncratic volatility, after controlling for other potential determinants of information content. Overall, the implementation of the pilot programme lowers the information content of eligible stocks at both high- and low-frequency levels.

In contrast, the coefficient of the interaction term between POST and LIST is statistically significantly negative in columns 4 and 5 and is negative in Column 6, indicating that the implementation of the pilot programme in China is associated with smaller price delay and smaller cross-autocorrelation. This association suggests that the implementation of the pilot programme accelerates the speed of information incorporation.

As we divide price efficiency into information content and price-adjustment speed, it is clearer to separately identify the effect of the implementation of margin trading on the divergent aspects of price efficiency. We find that this implementation reduces information content, which is disadvantageous in price efficiency. However, it speeds up price adjustment, which is favourable for price efficiency. These findings provide a more integrated view of the impact of the pilot programme on price efficiency in China compared with findings in the literature.18

5.1 Summary statistics

Panel A of Table 4 reports the summary statistics of price efficiency and other control variables based on monthly panel data. The mean value of the monthly margin-trading activity proxy (i.e. Margin) is 16.48 percent, indicating that the relative margin-buying amount accounts for 16.48 percent of the total trading amount during our sample period. In addition, both our information-content measures and price-adjustment speed proxies exhibit positive mean values and substantial cross-sectional variation.

In Panel B of Table 4, we report cross-sectional Pearson correlation coefficients between different price-efficiency measures. The Pearson correlation coefficients between IVOL and the other two information-content measures (PriceErr and |AR15|) are statistically significantly negative, indicating that these three measures share a common component of information content. In addition, the Pearson correlation coefficients between different measures of price-adjustment speed are statistically significantly positive, implying that these measures also have a common component of price-adjustment speed. These results are in line with those in Table 2.

5.2 Regression results

We run a panel regression with month- and stock-fixed effects to estimate Model (4).19 Table 5 reports the regression results. Columns 1 to 3 present the regression results for measures of information content (PriceErr, |AR15| and IVOL, respectively), and columns 4–6 present the results of the price-adjustment speed proxies (Delay1, Delay2 and |CR|, respectively).

As shown in Table 5, the coefficients of Margin in columns 1 and 2 are statistically significantly positive, which indicates that increased margin-trading activities are associated with higher pricing error (PriceErr) and higher absolute value of intraday return autocorrelations (|AR15|). In addition, the coefficient of Margin in Column 3 is statistically significantly negative, implying that greater margin-trading activities are associated with lower idiosyncratic volatility (IVOL). These findings suggest that stock prices deviate more from their efficient prices and incorporate less firm-specific information when there are more margin-trading activities. Overall, more margin-trading activities are associated with less information content, so margin trading is not conducive to information content.

In sharp contrast, the coefficients of Margin in columns 4–6 are statistically significantly negative, which implies that increased margin-trading activities are associated with smaller price delay (both Delay1 and Delay2) and smaller cross-autocorrelation between stock return and lagged market return (|CR|). These findings suggest that pilot stocks with greater margin-trading activities incorporate marketwide information more quickly into prices than do those with fewer margin-trading activities. Hence, more margin-trading activities are associated with faster price adjustment, implying that margin trading is beneficial to price-adjustment speed.

Our findings based on margin-trading activities in this section highlight the divergent impacts of margin trading on different aspects of price efficiency. We find that more margin-trading activities are associated with less information content, which is also supported by Hardouvelis (1990) and Rytchkov (2014). In sharp contrast, we show that more margin-trading activities are associated with faster price-adjustment speed, which is in line with Seguin (1990) and Chang et al. (2014). These findings indicate that the debate on the effect of margin trading on price efficiency may be related to the different measures of price efficiency used in the literature. The cross-sectional result in this section is more persuasive than that in the previous section.

5.3 Alternative tests