Against the tide: the commencement of short selling and margin trading in mainland China

1. Introduction

On 31 March 2010, the China Securities Regulatory Commission (CSRC) permitted margin trading and securities lending for the first time.1 The launch of short selling in China occurred at a time when market regulators around the world were restricting short-selling activities, especially on financial sector stocks.2 This allows us to document the impact of a country ‘swimming against the tide’ of international regulation. We examine the effect of margin trading and short sales on relative prices, liquidity, and volatility, using data from Mainland China and Hong Kong.

We contribute to the literature in several ways.3 Firstly, the dual introduction of short selling and margin trading allows us to assess the relative impact of each. We find that the impact of allowing short sales is the stronger of the two. It seems that it is the heightened risk of trading against informed investors that is driving this result. Secondly, we document the introduction of short selling and margin trading on liquidity in an emerging market. These issues have received relatively little attention in emerging markets compared with developed markets. Emerging markets evidence may be different to that from developed markets due to factors such as weaker investor protection (e.g. Morck et al., 2000). Contrary to both the regulator's intention and evidence from developed markets, this change in regulation resulted in decreased liquidity. This implies that investors avoid the stocks the regulation relates to. Thirdly, there is only limited evidence on the impact of short sales and margin trading on stock returns in emerging markets. Lee and Yoo (1993) find no relationship between margin requirements and stock return volatility in Korea and Taiwan, while Lamba and Ariff (2006) find return increases following the relaxation of short-sale constraints in Malaysia.

We use two matching procedures. The first involves matching eligible A-share (pilot programme companies) with ineligible A-share companies (referred to as the M-shares hereafter in the paper) which have similar characteristics, while the second uses the pilot programme cross-listed H-shares. Our main results are robust to both matching methods and the length of pre- and postperiod used. We show the short-selling effect dominates the margin trading effect. In the postperiod, the prices of A-shares eligible for short selling and margin trading decrease more than those of noneligible M-shares (or matched H-shares). As a consequence, the A-M (A-H) premium4 becomes smaller in the postperiod. These results are consistent with the theoretical models of Miller (1977) and Chen et al. (2002) which suggest short-sale restrictions tend to inflate prices. However, in contrast to these models, which imply that it is the actions of short sellers that drive prices lower, our results suggest that, consistent with Ausubel (1990), it is the risk of trading against informed investors, in this case short sellers, that lead to the price reduction. Our results are also in accordance with Chan et al. (2010) who show the premium of Chinese-listed stocks over their dual listed Hong Kong equivalents becomes relatively larger during market declines, for stocks where H-shares can be short sold. While the reduction in margin requirements might be expected to increase the relative prices (and premiums) of eligible A-shares, Chinese investors with positive information could already trade prior to the regulation change by borrowing with credit cards and house mortgage agreements (Tarantino, 2008).

We also explore the impact of margin trading and short selling of the eligible A-shares on the trading value ratio between A-share and M-share (H-share) trading value. On announcing the plans to launch the pilot programme, the CSRC's chief motivation is ‘to enlarge the supply and demand of funds and securities and increase the trading volume, thus leading to active liquidity’.5 This motivation is consistent with the finding of most empirical studies. Kolasinksi et al. (2009) and Boehmer et al. (2009) document liquidity declines in stocks which have short-sale bans imposed. Moreover, margin trading papers such as Seguin (1990) and Hardouvelis and Peristiani (1992) find that lower (higher) margin requirements lead to an increase (decrease) in liquidity.

In contrast to CSRC's motivation and the prior literature, we find both absolute and relative liquidity decreases for the pilot programme stocks. Liquidity in Chinese market in general declined in the period we consider, but we show additional declines in pilot stock liquidity. This is not unexpected as the asymmetric information risk increases for the 89 pilot programme stocks enabling insiders with either negative or positive insider information to exploit the asymmetric information more effectively through short selling and margin trading, respectively.6 Ausubel (1990) argues that ‘if ‘outsiders’ expect ‘insiders’ to take advantage of them in trading, outsiders will reduce their investment’ (p. 1022).7 Similarly, uninformed investors may seek to reduce their risk of losses to informed investors by only trading to meet their liquidity requirements (Admati and Pfleiderer, 1988). As the pilot programme is effective for only 89 stocks (approximately 6.7 per cent of all A-shares), uninformed investors may be able to find appropriate substitute nonpilot stocks to invest in thereby reducing their asymmetric information risk.

Chakravarty et al. (1998) suggest that insider trading is common in China. Therefore, our results point to ‘outsiders’ being aware of these facts and directing their trading activity away from the pilot programme stocks as a result. We also find that the frequency and level of short sales are very low. The mean pilot stock has short-sales activity on just 8 per cent of days, and, on average, short-sales activity relates to just 0.01 per cent of volume. This indicates that the additional trading volume directly associated with short-selling and margin trading activity is dwarfed by the reduction in normal trading activity for these pilot stocks. Further, the decline in premium of pilot stocks is more likely due to short-sale risk rather than short-sale activity.

We find some evidence that the spreads of pilot securities increase, relative to peer firms, after the introduction of margin trading and short selling. This exists in the A-H results but not in the A-M results. This result contrasts with Autore et al. (2011), Beber and Pagano (2013) and Boehmer et al. (2009), who each show that short-sales bans during the global financial crisis lead to wider bid-ask spreads. However, our spread results are consistent with our trading value results. A decrease in trading value and increased spreads both point to a decrease in overall liquidity for the pilot stocks following the introduction of short selling and margin trading.

Previous literature has not formed a consensus on the relation between short selling and/or margin trading and volatility. Scheinkman and Xiong (2003) suggest short-sale constraints may result in increased price volatility. However, Chang et al. (2007) find higher volatility in individual stocks when short selling is practised. Seguin (1990) finds a decrease in volatility when margin trading is allowed, but Hardouvelis and Peristiani (1992) find lower volatility in stocks when margin requirements are increased. Our volatility results are less consistent than those for the other variable; however, the majority of evidence points to a reduction in volatility.

From a public policy perspective, our findings suggest that the introduction of margin trading and short selling aids in the price discovery process. In particular, short selling allows for the first time in China investors with a pessimistic outlook for a firm to have their views incorporated into stocks prices. However, the regulation change does not boost the liquidity of the pilot programme securities, as anticipated by the regulators. In fact, liquidity as measured by spread and daily trading value deteriorate for pilot stocks.

The rest of this paper is organized as follows: Section 2 contains background on short-sales and margin trading regulations in China; our hypotheses are motivated and stated in Section 3; the Data and Methodology are outlined in Section 4; the results are discussed in Section 5; and Section 6 concludes the paper.

2. Chinese margin trading and short-sales regulation

The purpose of introducing the pilot programme of margin trading and short sales by the China Securities Regulatory Commission (CSRC) is ‘to integrate more information into securities prices. So investors can conduct securities lending or margin trading when the stock price is high or low, thus forming more proper stock prices’.8 The change in regulations to allow margin trading and short selling were discussed by the CSRC as early as 2006 (Bryan et al., 2010). In October 2008, the CSRC announced that there would soon be a margin trading and securities lending trial. However, it was not until 8 January 2010 that China's State Council gave the ‘in principle’ approval to introduce margin trading and security lending on a trial basis (Bryan et al., 2010). Then on 12 February 2010, the CSRC announced the first details of the stocks that would be part of a pilot programme for margin trading and short-selling operations (Bryan et al., 2010). Under the programme, CSRC approved 90 blue-chip securities for margin trading and securities lending including 50 from the Shanghai Stock Exchange (SSE) and 40 from the Shenzhen Stock Exchange (SZSE). The pilot programme was formally launched on 31 March 2010 (Bryan et al., 2010). The implementation rules are given in Appendix I, and the list of 90 stocks targeted for margin trading and short selling are reported in Appendix II.

3. Hypotheses development

The first hypothesis relates to the effect of short selling and margin trading on the prices of stocks in the pilot programme relative to the price of peer stocks listed in China and cross-listed stocks in Hong Kong. We measure the A-M (A-H) ‘premium’ as the difference in A-share and M-share (H-share) prices, divided by the A-share price. It is well documented that Chinese-listed firms trade at a premium to their Hong Kong-listed equivalents (e.g. Chan et al., 2010). The A-H premium is therefore expected to be positive. For the sake of consistency, we refer to the difference between A and M prices (relative to the A price) as a ‘premium’. However, as the M companies are simply Chinese-listed firms with similar characteristics to their A company counterparts, this ‘premium’ may be negative. This does not affect our analysis as we are focused on the change in premium following the regulation and the hypothesized direction of the change should be the same in both the A-M and A-H samples.

The introduction of short sales is expected to result in a decrease in the price of pilot companies relative to their peers. The theoretical models of Miller (1977) and Chen et al. (2002) show that short-sale constraints result in overvaluation as pessimistic investors are prevented from entering the market. The empirical work of Jones and Lamont (2002), Ofek and Richardson (2003), Chang et al. (2007) and Chan et al. (2010) also finds that short-sale constraints lead to higher relative prices. Hence, removing these constraints (i.e. allowing short sales) on pilot stocks should be expected to results in a decrease in their price relative to peer firms.

Conversely, allowing margin trading permits cash-constrained investors with a positive view on a stock price to trade. This implies the price of pilot firms should be higher, relative to their peers, following the regulation change. Largay (1973) and Hardouvelis and Peristiani (1992) both find that the margin size requirement is negatively related to price movements. In the case of China, margin requirements on pilot stocks declined from 100 per cent to a lesser amount so the relative price response of pilot firms should be positive. However, the margin buying effect on pilot stock prices may be limited in case of China, as the optimistic investors can invest in shares by borrowing against house and other assets before the implementation (Tarantino, 2008). Based on the literature discussed above, our two hypotheses are:

However, empirical results from emerging markets may vary from their developed market equivalents (e.g. Morck et al., 2000). Following the implementation of pilot programme, less-informed investors are likely to be reluctant to invest in pilot stocks if, as suggested by Asubel (1990), they believe there is more chance of trading against informed investors. The regulators in China set high entry requirements for investors to participate in pilot margin trading and short selling (Wang, 2011). This suggests it is likely that it is mostly better informed institutional investors who short sell and margin trade. The risks to uninformed investors are exacerbated by the well-documented insider trading and investor protection issues in China (e.g. Chakravarty et al., 1998). Given the above, it is possible that, consistent with the findings of Ausubel (1990), liquidity declines in pilot programme stocks.

4. Data and methodology

Our study covers the period from 1 October 2009 through 30 June 2010. This is split into a 3-month preperiod of October 1–December 31 and a 3-month postperiod of March 31–June 30.9 We decide against using the 3-month preperiod of 1 January to 30 March 2010 due to the contaminating announcement of the launch of pilot programme on 12 February 2010. Henan Shuanghui Investment and Development Co., Ltd did not trade from 22 March 2010 to 26 November 2010 and therefore is excluded. The final sample therefore includes 89 pilot stocks that were part of the margin trading and short-selling pilot programme. Fifty are listed on the Shanghai Stock Exchange (SSE), and 39 are listed on the Shenzhen Stock Exchange (SZSE). We also conduct robustness tests around period length and start points, which we discuss in more detail in the results section. Some firms were removed from the pilot short-selling and margin trading programme after 30 June 2010 so longer postperiods involve a reduction in stock numbers. Data on closing stock prices, trading volume, bid-ask spread, high/low volatility, market capitalization and number of shares outstanding at daily level for A-shares and H-shares are obtained from Thomson Reuters Datastream. The stock market index, including SSE A-share Index, and currency exchange rate between Chinese RMB and Hong Kong Dollar are also obtained from Thomson Reuters Datastream. Further, we obtain the daily short-selling and margin trading data from Chinese Securities Market and Accounting Research (CSMAR).

As mentioned in the introduction, we match the 89 pilot programme firms to peer companies using two distinct approaches. The first match is to peer Chinese-listed A-shares. One of the advantages of studying the launch of the pilot programme is that only 90 blue-chip stocks are included in the pilot programme leaving a large number of stocks with similar characteristics to match with. The first matching procedure is similar to the approach Boulton and Braga-Alves (2010) use. We first require matched candidates to belong to the same industry. For all SSE- and SZSE-listed firms that meet the industry requirement, we calculate the mean market value of equity, closing stock price, volatility of daily returns and daily turnover between 1 October 2008 and 30 September 2009. The match for each pilot programme stock is determined by finding the peer firm that minimizes the following equation:

(1)

where the factor refers to the mean market value of equity, closing stock price, volatility of daily returns and daily turnover between 1 October 2008 and 30 September 2009. Eligible (matched) firms are those in (not in) the pilot programme. Matches are performed without replacement, resulting in a unique combination for each eligible firm.10 The difference between the means of pilot and matched sample factors is insignificantly different from zero except market value which is significantly different at 1 per cent level. All data for the matching procedure are obtained from Thomson Reuters Datastream.11

The second matching procedure involves sourcing data for the Hong Kong-listed share of each of the 89 pilot programme stocks. Using data from the Website of Hong Kong Stock Exchange (HKEx), we find that 26 of the 89 companies have shares listed in Hong Kong. The matching procedures each have their strengths and weaknesses. The first matching procedure has the advantage of using firms trading on the same exchange; however, the disadvantage is that it is not always possible to find a matched firm of a similar size.12 The second matching procedure uses prices of the same company; however, the exchange is different. These two approaches therefore complement each other.

We investigate whether the regulation resulted in changes in four variables. We calculate each variable each day using both M and H control firms, but all formulae below are based on M-shares. To calculate premium between A- and H-shares, we converted HKG dollar H-share prices to Chinese RMB. Following Chan et al. (2010), premium is calculated as:

(2)

Following the methodology of Gagnon and Witmer (2009), we construct the A-M value ratio as:

(3)

In accordance with Boulton and Braga-Alves (2010), the difference in spread is calculated as13 :

(4)

We follow Helmes et al. (2010) and calculate the volatility difference as14 :

(5)

5. Empirical results

5.2 Short-sales and margin trading activity summary statistics

We present statistics relating to the level and frequency of short-sales and margin trading activity in Table 2. The short-sales (margin trading) ratio is the ratio of short sales (margin trading) to total volume. This is calculated each day for each stock. A time series average is then computed for each stock, and summary statistics are based on the stock average numbers. The proportions relate to the fraction of days there is short-sales or margin trading activity, respectively, in each stock. Again, statistics are based on the stock numbers.

Table 2. Short-sale and margin trading activity summary statistics

Variable	Mean	Median	Minimum	Maximum	Std. Deviation
Short-sale ratio	0.0001	0.0000	0.0000	0.0019	0.0003
Margin trading ratio	0.0021	0.0018	0.0003	0.0084	0.0014
Short-sale proportion	0.0794	0.0000	0.0000	0.8305	0.1563
Margin trading proportion	0.6629	0.6833	0.2333	0.9322	0.1659

• These statistics relate to the stocks that were allowed to be sold short and traded on margin between 31 March 2010 and 30 June 2010. The short-sale (margin trading) ratio relate to the short-sale (margin trading) volume divided by total volume. This time series average is calculated for each stock, and cross-sectional statistics are then calculated. The short-sale (margin trading) proportion is the proportion of days when there is short sales (margin trading) activity. Statistics are calculated across all stocks.

The Table 2 results indicate short sales and margin trades only account for a small fraction of the volume traded in each pilot stock in the 3-month period following the regulation change. Short sales contribute just 0.01 per cent of the total volume for the average stock. Margin trades contribute just 0.2 per cent on average. This lack of activity is reasonably consistent across stocks. The stock with the most short sales (margin trading) activity has just 0.19 per cent (0.84 per cent) of volume driven by this activity. The proportion of days with short-sales activity is just 7.9 per cent for the average stock. The equivalent number is considerably higher for margin trading (66.3 per cent), which indicates a reasonably high frequency. Figure 1 shows the proportion of average daily volume that relates to short sales and margin trading in the 3-month period following the regulation change. While margin trading activities show growth over the period, short sales do not and clearly the trades are small and relatively inconsequential when compared with total volume.

These results clearly indicate that the decline in premium in Table 1 is not due to the short sellers driving prices down. Rather, it appears to be a result of market participants being concerned about the potential for short sellers to drive prices lower. This concern appears to manifest itself in investors reducing their trading in pilot stocks, which combined with the increased asymmetric information risk results in an increase in spreads. We address these issues in the next section on multivariate analysis.

5.3 Multivariate analysis

A multivariate fixed effect panel regression framework is employed to test whether the introduction of margin purchases and securities lending has any effect on eligible A-shares in mainland China.

5.3.1 Premium

To test hypothesis 1, we estimate a number of regression models to examine how the premium changes between eligible A-shares and noneligible M-shares (i.e. changes in A-M premium) and also between a subsample of A-shares and shortable H-shares (i.e. changes in A-H premium) from preperiod to postperiod. To control for fixed effects and autocorrelations of the residuals, we estimate equations (6a) and (6b) using ordinary least-squares regression and adjust the t-values by Rogers standard errors clustered at the firm level. The regression model is as follows:

()

()

where the dependent variable is A-M (or A-H) share premium. Post_period_t is a dummy variable that is set to one for the period from 31 March 2010 to 30 June 2010, when margin trading and short selling are permitted for the 89 pilot stocks and zero otherwise. Following Chan et al. (2010), we run the regression with two different interaction dummy variables to represent declining markets. Post_period*Dummy_MKT^A is an interaction dummy variable set to one on days the market declines during the postperiod and zero otherwise. Post_period*Dummy_RET^A is an interaction dummy variable set to one, on days the pilot programme stock declines during the postperiod and zero otherwise. These dummy variables are particularly appropriate given the results in Section 5.2. Days when the market is declining are likely to be those days when there is heightened concern about the risk of trading against short sellers. We follow Chan et al. (2010) and include spread and volatility control variables.

The main result of the regression analysis is that the A-M (A-H) premium is negatively related to the postperiod dummy and the interaction term as shown in Model (1)A, (1)B, and (1)C of Panels A and B. This is consistent with our hypothesis 1A that the premium becomes smaller when short selling is allowed in China. Further, this effect of smaller premiums is more pronounced on days when the Chinese stock markets decline. The evidence also suggests that the effect of short sellers dominates the Chinese market as compared to margin purchasers during the period examined.

The results of some control variables are inconsistent with expectations. For example, in Model (1)E of Panel A, the evidence indicates that the A-M premium is negatively associated with M-share volatility, suggesting that an increase in M-share volatility is accompanied by an increase in M-share prices and, thus, a decline in the A-M premium. Furthermore, in Model (1)E of Panel B, only one estimation model with control variables has significant and consistent results with what we expected. That is, the A-H premium is positively associated with H-share volatility, suggesting that an increase in H-share volatility is accompanied by a decrease in H-share prices and, thus, the A-H premium increases.

5.3.2 Trading value ratio

We test hypothesis 2 that A-M (A-H) value ratio is higher for those A-shares allowed for margin trading and short selling than for those ineligible in China, that is, matched M-shares and shortable H-shares in Hong Kong.17 We estimate regression models similar to Equations (6a) and (6b), except that the dependent variable is Value_Ratio_i,t.

The results are presented in Panels A and B of Table 4. The key variable of interest is the postperiod and the interaction term. Similar to our univariate results, we find evidence contrary to our hypothesis 2; the coefficient on the postperiod dummy and the interaction term is significantly negative in Model (1)A and (1)B of Panels A and B. This means that A-share trading value declined relative to M-share (H-share) following the introduction of the pilot programme. Similarly, when the A-share market goes down, for the group of affected A-share stocks, there is a decline of A-share trading value relative to M-share (H-share) trading value. Our results are inconsistent with the literature that documents liquidity increases when short-sale constraints and margin requirements are reduced. However, the evidence is consistent with the model of Ausubel (1990) as uninformed investors reduce their exposure to the increased asymmetric information risk by reducing trading activity or simply avoiding the 89 pilot stocks.

Unlike the regression results in Panels A and B of Table 3, where Premium_i,t is used as the dependent variable, the coefficients associated with the control variables in Panel A and B of Table 4 are highly significant. We find that the Value_Ratio_i,t is positively related to M-share (H-share) bid-ask spread and positively (negatively) related to A-share (M-share) volatility.

Table 3. Premium regressions

	SSE A-share index			Individual A-share
	(1)A	(1)B	(1)C	(1)D	(1)E	(2)B	(2)C	(2)D	(2)E
Panel A: China match
Post_period	−0.0610** (−2.25)		−0.0584** (−2.22)	−0.0067 (−0.27)	−0.0564** (−2.17)		−0.0528** (−1.96)	−0.0050 (−0.20)	−0.0501* (−1.88)
Post_period*Down_dummy		−0.0461** (−2.29)	−0.0043** (−2.04)	0.0053 (1.14)	−0.0060 (−1.14)	−0.0512** (−2.36)	−0.0152 (−1.12)	0.0026 (0.27)	−0.0184 (−1.33)
Relative bid-ask spread (A-share)				−0.2794*** (−5.38)				−0.2794*** (−5.38)
Relative bid-ask spread (M-share)					0.0651 (1.40)				0.0655 (1.41)
Relative price volatility (A-share)				−0.0196 (−0.53)				−0.0195 (−0.53)
Relative price volatility (M-share)					−0.1410*** (−3.17)				−0.1411*** (3.18)
Intercept	−0.0441 (−1.06)	−0.0608 (−1.40)	−0.0441 (−1.06)	−2.1461*** (−5.15)	−0.0555 (−0.17)	−0.0608 (−1.39)	−0.0441 (−1.06)	−2.1458*** (−5.15)	−0.0536 (−0.16)
Observations	10 890	10 890	10 890	9851	10 249	10 890	10 890	9851	10 249
R-Squared	0.0044	0.0021	0.0044	0.1698	0.0369	0.0025	0.0045	0.1698	0.0370
Panel B: Hong Kong match
Post_period	−0.0885*** (−5.96)		−0.0840*** (−5.70)	−0.0840*** (−4.35)	−0.0862*** (−5.83)		−0.0940*** (−5.03)	−0.0928*** (−4.48)	−0.0941*** (−5.15)
Post_period*Down_dummy		−0.0678*** (−6.48)	−0.0071*** (−5.77)	−0.0063* (−1.92)	−0.0048 (−1.40)	−0.0550*** (−4.42)	0.0096 (0.60)	0.0082 (0.52)	0.0084 (0.59)
Relative bid-ask spread (A-share)				0.0217 (0.41)				0.0215 (0.41)
Relative bid-ask spread (H-share)					0.0793 (1.59)				0.0792 (1.58)
Relative price volatility (A-share)				0.1001** (2.08)				0.1002** (2.09)
Relative price volatility (H-share)					0.0706* (1.77)				0.0708* (1.78)
Intercept	0.1975*** (4.04)	0.1742*** (3.53)	0.1975*** (4.04)	0.7106 (1.61)	0.9422** (2.42)	0.1681*** (3.37)	0.1975*** (4.04)	0.7099 (1.60)	0.9423** (2.42)
Observations	2931	2931	2931	2828	2926	2931	2931	2828	2926
R-Squared	0.0295	0.0153	0.0296	0.0623	0.0860	0.0095	0.0296	0.0624	0.0861

• Panel A (B) reports the results of fixed-effects panel regressions for the 89 (26) pilot and M (H) matched stocks. The dependent variable is premium, which is calculated for each stock as (pilot stock price – matched stock price)/pilot stock price. The sample period is from 1 October 2009 to 30 June 2010. Post_period is a dummy variable that is equal to one for the period from 31 March 2010 to 30 June 2010 and zero otherwise. Post_period*Down_dummy is an interaction dummy variable set to one, when the return of A-share market (MKTA) in Models 1 and individual pilot A-share (RETA) in Models 2 is less than 0 during the postperiod and zero otherwise.
• 
  
• The t-value is adjusted by the Rogers standard error clustered by firm. *, ** and *** indicate significant at 10%, 5% and 1% level, respectively.

Table 4. Value ratio regressions

	SSE A-share index			Individual A-share
	(1)A	(1)B	(1)C	(1)D	(1)E	(2)B	(2)C	(2)D	(2)E
Panel A: China match
Post_period	−0.0265 (−2.33)		−0.0252** (−2.25)	−0.0233** (−1.99)	−0.0238** (−2.16)		−0.0207* (−1.85)	−0.0193* (−1.67)	−0.0180* (−1.68)
Post_period*Down_dummy		−0.0203** (−2.36)	−0.0022 (−0.77)	−0.0031 (−0.93)	−0.0066** (−2.02)	−0.0247*** (−2.76)	−0.0102** (−2.13)	−0.0103* (−1.93)	−0.0172*** (−3.65)
Relative bid-ask spread (A-share)				−0.0012 (−0.07)				−0.0010 (−0.06)
Relative bid-ask spread (M-share)					0.0364** (2.24)				0.0367** (2.25)
Relative price volatility (A-share)				0.0230** (1.97)				0.0228** (1.96)
Relative price volatility (M-share)					−0.0470*** (−4.70)				−0.0472*** (−4.71)
Intercept	0.7296*** (57.18)	0.7225*** (54.56)	0.7296*** (57.18)	0.7999*** (6.14)	0.8298*** (6.99)	0.7233*** (54.65)	0.7296*** (57.18)	0.8005*** (6.14)	0.8310*** (7.02)
Observations	10 298	10 298	10 298	9647	9905	10 298	10 298	9647	9905
R-Squared	0.0003	0.0030	0.0060	0.0097	0.0461	0.0042	0.0065	0.0101	0.0472
Panel B: Hong Kong match
Post_period	−0.1095*** (−5.43)		−0.1035*** (−5.09)	−0.0966*** (−4.79)	−0.1053*** (−5.17)		−0.0898*** (−4.38)	−0.0874*** (−4.48)	−0.0908*** (−4.55)
Post_period*Down_dummy		−0.0844*** (−5.52)	−0.0093** (−2.37)	−0.0089** (−2.56)	−0.0088* (−1.87)	−0.0963*** (−5.28)	−0.0335** (−2.25)	−0.0255* (−1.81)	−0.0342** (−2.51)
Relative bid-ask spread (A-share)				−0.0253 (−0.54)				−0.0251 (−0.53)
Relative bid-ask spread (H-share)					0.0780** (2.52)				0.0780** (2.52)
Relative price volatility (A-share)				0.1523*** (4.37)				0.1517*** (4.34)
Relative price volatility (H-share)					0.0534 (1.40)				0.0536 (1.40)
Intercept	0.6377*** (16.19)	0.6092*** (15.68)	0.6377*** (16.19)	1.0010*** (2.98)	1.3124*** (5.02)	0.6106*** (15.27)	0.6377*** (16.19)	1.0008*** (2.98)	1.3133*** (5.04)
Observations	2878	2878	2878	2828	2878	2878	2878	2828	2878
R-Squared	0.0551	0.0287	0.0553	0.1653	0.1126	0.0358	0.0577	0.1666	0.1151

• Panel A (B) reports the results of fixed-effects panel regressions for the 89 (26) pilot and M (H) matched stocks. The dependent variable is Value_Ratio, which is computed for each stock as pilot stock value/(pilot stock value + matched stock value). The sample period is from 1 October 2009 to 30 June 2010. Post_period is a dummy variable that is equal to one for the period from 31 March 2010 to 30 June 2010 and zero otherwise. Post_period*Down_dummy is an interaction dummy variable set to one, when the return of A-share market (MKTA) in Models 1 and individual pilot A-share (RETA) in Models 2 is less than 0 during the postperiod and zero otherwise.
• 
• 
• The t-value is adjusted by the Rogers standard error clustered by firm. *, ** and *** indicate significant at 10%, 5% and 1% level, respectively.

We also estimate the regression model using individual stock returns rather than market returns to form a dummy variable for declining returns. These results, which are reported in the last four columns of Panels A and B in Table 4, are qualitatively similar to those reported for the A-share index market return. Again, the evidence is inconsistent with hypothesis 2, with the coefficient on the postperiod dummy and the interaction term being significantly negative in Model (2)B, (2)C, (2)D and (2)E in Panels A and B of Table 4, respectively. This suggests that when the A-share price of an individual stock goes down, there is a decline of A-share trading value relative to the M-share (H-share) trading value. Finally, the coefficients of the relative volatility and relative spread in Model (2)D and (2)E are similar to those in Model (1)D and (1)E in Table 4.

5.3.3 Relative bid-ask spread differential

We test hypothesis 3 that the A-M (A-H) relative spread differential Spread_Diff_i,t is lower for those A-shares allowed for margin trading and short selling than for those ineligible in China, that is, matched M-shares and shortable H-shares in Hong Kong. We estimate regression models similar to Equations (6a) and (6b), except that the dependent variable here is Spread_Diff_i,t.

The results presented in Panels A and B of Table 5 are inconsistent with our hypothesis 3 of a decline in the A-M (A-H) spread differential following the change in regulation. The A-H results (Panel B) show a clear pattern of an increase in relative spreads following the change in regulation. This is evident in Models (1)A, (1)C, (1)D, (1)E, (2)D and (2)E. The postperiod dummy variable is not consistently statistically significant in the M matching results (Panel A), but when it is statistically significant, it is positive (see Models (1)C, (1)E and (2)E). However, the interaction dummy variable representing the incremental impact of down periods in the postperiod is sometime positive in Panel B, which is difficult to explain. Overall, we conclude there is evidence that spreads increase rather than decrease following the change in regulation.

Table 5. Panel regressions of difference in relative bid-ask spread

	SSE A-shares index					Individual A-share
	(1)A	(1)B	(1)C	(1)D	(1)E	(2)B	(2)C	(2)D	(2)E
Panel A: China match
Post_period	0.0164 (0.68)		0.0422* (1.64)	−0.0211 (−0.59)	0.0895*** (2.82)		0.0152 (0.57)	−0.0373 (−1.04)	0.0621* (1.92)
Post_period*Down_dummy		−0.0128 (−0.70)	−0.0428*** (−3.20)	−0.0444*** (−3.07)	−0.0357** (−2.54)	0.0126 (0.66)	0.0021 (0.12)	−0.0185 (−1.06)	0.0110 (0.58)
log (Price) A-share				−0.3443*** (−5.48)				−0.3442*** (−5.46)
log (Price) M-share					0.0810 (1.07)				0.0817 (1.08)
log (Volume) A-share				0.0060 (0.15)				0.0063 (0.16)
log (Volume) M-share					0.1055** (2.45)				0.1061** (2.46)
Relative price volatility (A-share)				0.0264 (0.78)				0.0256 (0.76)
Relative price volatility (M-share)					0.0539 (0.68)				−0.0535 (0.67)
Intercept	−0.1087** (−2.41)	−0.0964* (−2.16)	−0.1087** (−2.41)	0.8846 (1.47)	−1.1210* (−1.78)	−0.1040** (−2.30)	−0.1087** (−2.41)	0.8785 (1.46)	−1.1294* (−1.79)
Observations	9290	9290	9290	9290	9290	9290	9290	9290	9290
R-Squared	0.0001	0.0001	0.0006	0.1318	0.0239	0.0001	0.0001	0.1314	0.0236
Panel B: Hong Kong Match
Post_period	0.1382*** (3.07)		0.1287** (2.32)	0.1514*** (3.17)	0.1358*** (2.90)		0.1048 (1.52)	0.1443** (2.74)	0.1238** (2.29)
Post_period*Down_dummy		0.1080*** (3.70)	0.0149 (0.53)	0.0280 (0.97)	0.0038 (0.14)	0.1301*** (3.18)	0.0570 (0.85)	0.0436 (0.90)	0.0245 (0.45)
log (Price) A-share				−0.5141*** (−5.40)				−0.5130*** (−5.41)
log (Price) H-share					−0.3199*** (−3.60)				−0.3197*** (−3.59)
log (Volume) A-share				0.2223*** (3.44)				0.2233*** (3.44)
log (Volume) H-share					0.2404*** (4.82)				0.2404*** (4.83)
Relative price volatility (A-share)				−0.2082*** (−3.32)				−0.2076*** (−3.33)
Relative price volatility (H-share)					−0.3095*** (−6.74)				−0.3095*** (−6.75)
Intercept	−0.8722*** (−6.92)	−0.8367*** (−6.63)	−0.8722*** (−6.92)	−2.6842*** (−2.83)	−3.7128*** (−5.29)	−0.8405*** (−6.55)	−0.8722*** (−6.92)	−2.6958*** (−2.84)	−3.7132*** (−5.29)
Observations	2829	2829	2829	2828	2829	2829	2829	2828	2829
R-Squared	0.0064	0.0034	0.0064	0.4211	0.3954	0.0047	0.0069	0.4213	0.3955

• Panel A (B) reports the results of fixed-effects panel regressions for the 89 (26) pilot and M (H) matched stocks. The dependent variable is Spread_Diff, which is calculated for each stock as log[pilot stock (Ask – Bid)/((Ask + Bid)/2)] – log[matched stock (Ask – Bid)/((Ask + Bid)/2)]. The sample period is from 1 October 2009 to 30 June 2010. Post_period is a dummy variable that is equal to one for the period from 31 March 2010 to 30 June 2010 and zero otherwise. Post_period*Down_dummy is an interaction dummy variable set to one, when the return of A-share market (MKTA) in Models 1 and individual pilot A-share (RETA) in Models 2 is less than 0 during the postperiod and zero otherwise.
• 
• 
• The t-value is adjusted by the Rogers standard error clustered by firm. *, ** and *** indicate significant at 10%, 5% and 1% level, respectively.

Although the results are inconsistent with the recent literature on short-sale bans during the global financial crisis, our evidence is consistent with our trading activity results. An increase in the ability of informed traders to extract higher returns from margin trading and short selling appears to lead to greater asymmetric information risk which is consistent with wider bid-ask spreads in the 89 pilot stocks. The decrease in liquidity could also be expected to flow directly into wider bid-ask spreads.

5.3.4 Relative volatility differential

We test hypothesis 4 that there is no clear trend in the A-M (A-H) relative volatility differential Volatility_Diff_i,t after the regulation is introduced. In the hypothesis development section, we do not form a view on the direction of the relative volatility differential due to the mixed empirical findings in different settings. However, our univariate analysis presented in Table 1 shows a general decline in relative volatility, albeit only significantly so for the A-M match. In addition, the Appendix III results using 6-month pre- and postperiods show that pilot firm volatility significantly declines relative to both matches.

We estimate regression models similar to Equations (6a) and (6b), except that the dependent variable here is Volatility_Diff_i,t. The results are presented in Panels A and B of Table 6. The key variables of interest are the postperiod dummy and the downmarket interaction term. We find a decrease in the volatility of A-shares relative to their M-share counterparts in the postperiod, yet an increase in A-share volatility relative to their H-peers in the postregulation period (in downmarkets). For the A-M volatility differential, we find that the coefficient on both the postperiod and the downmarket interaction term is significantly negative in Model (1)A, (1)B and (2)B of Panel A. This means that when the A-share market (or individual A-share return) goes down, there is a decline of A-share high/low volatility relative to M-share in the postperiod.

Table 6. Panel regressions of difference in relative volatility

	SSE A-shares index					Individual A-share
	(1)A	(1)B	(1)C	(1)D	(1)E	(2)B	(2)C	(2)D	(2)E
Panel A: China match
Post_period	−0.0553** (−2.49)		−0.0586** (−2.49)	−0.0286 (−1.15)	−0.0926*** (−3.74)		−0.0502* (−1.89)	−0.0236 (−0.86)	−0.0843*** (−3.15)
Post_period*Down_dummy		−0.0367** (−2.13)	0.0054 (0.45)	0.0093 (0.78)	−0.0016 (−0.13)	−0.0442*** (−2.99)	0.0091 (−0.60)	0.0009 (0.06)	−0.0165 (−1.06)
log (Price) A-share				0.1145** (2.40)				0.1144** (2.39)
log (Price) M-share					−0.0596 (−1.11)				−0.0600 (−1.11)
log (Volume) A-share				0.0244 (1.12)				0.0243 (1.11)
log (Volume) M-share					−0.0785*** (−3.26)				−0.0787*** (−3.27)
Intercept	−0.0721*** (−3.01)	−0.0887*** (−3.74)	−0.0721*** (−3.01)	−0.6455** (−2.30)	0.8160** (2.47)	−0.0872*** (−3.50)	−0.0721*** (−3.01)	−0.6437** (−2.28)	0.8188** (2.48)
Observations	10 283	10 283	10 283	10 283	10 283	10 283	10 283	10 283	10 283
R-Squared	0.0030	0.0011	0.0030	0.0217	0.0229	0.0015	0.0030	0.0216	0.0231
Panel B: Hong Kong match
Post_period	−0.0272 (−0.68)		−0.0622 (−1.50)	0.0601 (1.32)	−0.0609 (−1.49)		−0.0083 (−0.21)	0.1026** (2.45)	−0.0077 (−0.20)
Post_period*Down_dummy		0.0099 (0.29)	0.0550** (2.23)	0.0704*** (3.14)	0.0553** (2.24)	−0.0381 (1.17)	−0.0324 (−1.46)	0.0028 (0.14)	−0.0308 (−1.42)
log (Price) A-share				0.2132*** (4.38)				0.2123*** (4.36)
log (Price) H-share					0.0344 (0.83)				0.0340 (0.82)
log (Volume) A-share				0.2092*** (5.84)				0.2082*** (5.43)
log (Volume) H-share					0.0062 (0.25)				0.0061 (0.24)
Intercept	0.0316 (0.90)	0.0145 (0.44)	0.0316 (0.90)	−2.7585*** (−5.40)	−0.1109 (−0.36)	0.0291 (0.88)	0.0316 (0.90)	−2.7456*** (−5.35)	−0.1083 (−0.35)
Observations	2876	2876	2876	2876	2876	2876	2876	2876	2876
R-Squared	0.0007	0.0001	0.0021	0.0983	0.0047	0.0012	0.0012	0.0961	0.0037

• Panel A (B) reports the results of fixed-effects panel regressions for the 89 (26) pilot and M (H) matched stocks. The dependent variable is Volatility_Diff, which is calculated for each stock as log[pilot stock (High – Low)/((High + Low)/2)] - log[matched stock (High – Low)/((High + Low)/2)]. The sample period is from 1 October 2009 to 30 June 2010. Post_period is a dummy variable that is equal to one for the period from 31 March 2010 to 30 June 2010 and zero otherwise. Post_period*Down_dummy is an interaction dummy variable set to one, when the return of A-share market (MKTA) in Models 1 and individual pilot A-share (RETA) in Models 2 is less than 0 during the postperiod and zero otherwise.
• 
• 
• The t-value is adjusted by the Rogers standard error clustered by firm. *, ** and *** indicate significant at 10%, 5% and 1% level, respectively.

However, the A-H volatility differential becomes significantly positive when the market is down during the postperiod in Model (1)C. Similarly, after inclusion of both the A- and H-share log(price), and log(volume) in Model (1)D and (1)E, the A-H volatility differential remains strongly positive for the interaction term. Hence, our volatility results based on the 3-month pre- and postperiod windows are mixed. For robustness, in Appendix IV, we re-run the volatility regression models using the 6-month pre- and postwindows. The results are much more consistent. For both the A-M and A-H volatility differential, the postperiod and interaction term are significantly negative in Model (1)A, (1)B and 2(B). When both the postperiod and downmarket interaction term are included in Model 1(C) and 2(C), the postperiod dummy is significantly negative in all models, while the interaction term is also significantly negative for Model 1(C) for A-M and Model 2(C) for both matches. Further, after the inclusion of both A- and H-share log(price), and log(volume), the A-H volatility differential for the interaction term is no longer significantly positive for any model.

5.3.5. Results summary

Our results suggest it is the risk rather than the reality of short-sales activity that leads to the prices of pilot programme stocks decreasing on average, relative to their peers, following the regulation change. Liquidity decreases in pilot programme stocks (relative to matched firms), which is also consistent with uninformed investors reducing trading activity due to the heightened risk of being adversely affected by informed traders. This explanation is also in line with the increase in pilot programme stock spreads (compared to peer firms). On balance, the evidence points to a decline in volatility following the regulation. As mentioned earlier, our key results are robust to matching pilot programme stocks with their Hong Kong-listed H-shares and similar firms listed in China. As a final robustness check, we rule out the possibility that there was a systematic change in the relation between Chinese-listed A-shares and their H-share counterparts by looking at A and H stock pairs that were not affected by the regulation change. These results are available from the authors on request.

6. Conclusion

We examine the effect of margin trading and short sales on relative prices, trading value, bid-ask spreads and price volatility. Our results contribute to the literature in several ways. We have an event that includes both margin trading and short selling, which allows us to see the relative effect of each. Further, the introduction of short selling occurred at a time when market regulators across the globe were increasing restrictions around the short-selling activities. Cash-constrained investors with a positive view of stocks could purchase stocks by borrowing against their house and prior to the commencement of pilot programme. Hence, it is unsurprising that our results show that the short selling effect dominates the margin trading effect. After the launch of the pilot programme by Chinese regulators, the prices of A-shares eligible for short selling and margin trading decrease more than those of noneligible M-shares (or matched H-shares). As a result, the A-M (A-H) premium becomes smaller in the postperiod. Thus, the premium results are consistent with the theoretical models of Miller (1977) and Chen et al. (2002) that constraining short sellers causes overvaluation. However, we find actual short-sales activity is very low. In contrast to the theoretical short-selling models, it appears it is investor selling due to the risk of trading against informed investors (short sellers) rather than the actions of short sellers that drive prices lower in China. This result is broadly consistent with the Ausubel (1990) theory.

In contrast to the Chinese regulator's desire for the short-selling and margin trading programme to improve stock liquidity, we find that pilot firms' liquidity decline is significantly larger relative to the matched samples. Pilot firms' average daily trading value drops significantly during the postperiod and is significantly lower compared with both matched samples. In addition, spreads widen for pilot firms' A-shares relative to their matched H-shares after the launch of pilot programme. The results are in contrast with our hypotheses and the broad short-selling and margin trading literature. However, this evidence supports Ausubel's (1990) prediction that uninformed traders will reduce or avoid trading in situations when they expect an increased likelihood of transacting with informed traders. Further, the ability of informed traders to exploit superior private information appears to lead to an increase in bid-ask spreads in the presence of margin trading and short selling for pilot securities. Assuming the deterioration in liquidity is due to heightened risk of trading against an informed investor, then policy-makers could enact regulation to improve investor confidence. For example, policies enforcing insider trading regulation and encouraging continuous disclosure may improve investor confidence.

Finally, while our volatility results are less consistent than those for premium, value and spread, most of the evidence suggests volatility declines. We find that after the pilot programme launch, volatility decreases when firms are matched with M-shares across all periods, while volatility declines versus H-share pairs over 6-month pre- and postperiods.