Short-selling regulations and market liquidity: Evidence from Europe

Transcription

1 Short-selling regulations and market liquidity: Evidence from Europe PhD Candidate: Giannoula Karamichailidou Department of Accounting and Finance, Business School The University of Auckland Supervision: Associate Professor Russell Poskitt and Associate Professor Alastair Marsden Trinity College Dublin, September 6-8,

2 ABSTRACT In the autumn of 2008, regulators around the world introduced short-selling rules to halt stock price declines of major banks and to restore confidence in the banking system. One consequence of the adoption of these short-selling rules was a significant increase in the bid/ask spreads of stocks affected by these regulations (Beber and Pagano, 2011).This paper examines the impact of short-selling regulations on market liquidity in a European context for the period January, 2008 to December, Bid/ask spreads are decomposed into their informational and non-informational components as measured by price impact and realized spread, respectively. We use this decomposition to investigate the channels through which bid/ask spreads increased during this period. We also examine depth to provide a more complete picture of the effects of the 2008 short-selling regulations on liquidity of European bank stocks. Overall, we find that both price impact and realized spreads increase after the implementation of short-selling regulations for all the countries in the sample. However, the results are mixed regarding depth. Liquidity unambiguously deteriorates for British, Swiss, and French bank stocks where higher bid/ask spreads are accompanied by lower depth. Moreover, liquidity deteriorates for Dutch, Spanish, Portuguese, and Danish bank stocks but only in terms of higher bid/ask spreads. Median depth for these bank stocks does not change statistically significantly after the introduction of short-selling rules. Lastly, the impact on liquidity is ambiguous for Belgian, Irish, and German bank stocks as higher bid/ask spreads are associated with greater depth. 2

3 1. INTRODUCTION Empirical evidence shows that the introduction of recent short-selling rules led to a significant increase in the bid/ask spreads (see for example, Clifton and Snape, 2008, and Beber and Pagano, 2011). Existing literature, to my knowledge, does not develop a theoretical framework of how short-selling regulations affect each component of the bid/ask spread or depth. Beber and Pagano (2011) provide some guidance on the issue in terms of the bid/ask spread; however, they do not decompose the spread to test their argument as they use daily data. This paper looks at the impact of 2008 short-selling regulations on market liquidity as measured by bid/ask spreads, their components, and depth of European bank stocks. It is important to understand the channels through which financial regulators policies increased transaction costs and the impact of their policies on market depth in a broader geographical area in order to implement more effective and efficient measures in the future. There is very little evidence on how short-selling regulations affect the informational and non-informational bid/ask spread components of bank stocks as well as their depth in a European context. Boehmer et al. (2008) and Marsh and Payne (2010) are the exception. Boehmer et al. (2008) estimate the price impact that represents the informational component of the bid/ask spread after the introduction of the short-selling ban on September 18, 2008, in the US. Marsh and Payne (2010) look at the impact of short-selling ban introduced on September 19, 2008 and lifted on January 16, 2009 on depth of the UK equity markets. I contribute to the literature by estimating the bid/ask spread components and depth for 10 European countries before and after the introduction of the 2008 short-selling rules. The rest of the paper is organized as follows. Section 2 is a review of prior studies examining the impact of recent short-selling rules on market liquidity. Section 3 develops the hypotheses while section 4 analyses the methodology. Section 5 describes the sample and data. Section 6 presents and discusses the empirical results and finally, section 7 summarizes the main findings and concludes. 3

4 2. LITERATURE REVIEW 2.1.US On July 15, 2008, the Securities Exchange Commission in the US introduced an Emergency Order (EO). The order required anyone engaging in short-selling in 19 specified financial stocks that should borrow the shares before the sale took place and deliver them at the settlement. Basically, this new rule prohibited naked short-selling and was lifted on August 15, One of the first papers that looks at the impact of 2008 short-selling regulations on market liquidity as measured by quoted and percentage quoted bid/ask spread is that of Bris (2008). He finds that the quoted spread is greater for the stocks affected by the rule compared to that of the US control group stocks in both pre-eo and post-eo periods. The difference is statistically significant for both periods but is greater in the post-eo period. In the pre-eo period, the average bid/ask spread of the non-us financial stocks is greater than that of the 19 affected stocks. In the post-eo period, the magnitude of the spread does not significantly change for the foreign financial stocks but the value of the average spread of the 19 financial stocks reaches the level of the spread of the non-us financial stocks. The main conclusion is that the EO seems to have deteriorated market liquidity as measured by the quoted absolute and percentage spread of the 19 financial stocks subject to this EO. Boulton and Braga-Aves (2009) also examine the impact of the EO on the pricing efficiency and liquidity of the affected financial stocks. Each financial stock subject to short-selling constraints has a unique match, a financial stock not affected by the Emergency Order. They find that quoted spreads for restricted stocks increase by 33.6% as opposed to 14.2% for the matched stocks during the EO period compared to their levels in the pre-eo period. Their multivariate analysis shows that quoted spreads of restricted stocks increase approximately 19% more than those of non-restricted firms from the pre-eo period to the post-eo period. The authors argue that liquidity deteriorates more for the restricted stocks during the EO period and that the greater transaction costs during that period could be partly attributed to the 2008 short-selling rules. Boehmer, Jones and Zhang (2008) examine how the short-selling ban, introduced by the EO affected market quality as measured by spreads and price impacts. They consider two groups of stocks: treatment group including 146 stocks on the ban list and a control group that contains 1,066 stocks. The authors find that median effective spreads increase to 32 basis points for the non-banned stocks, while the median effective spreads for the banned stocks increase to 77 basis points. Price impact is also found to be significantly greater in the ban 4

5 period for the banned stocks. The results imply that the short-selling ban significantly increases the bid/ask spreads as well as the price impact of the banned stocks after its implementation. 2.2.UK Clifton and Snape (2008) investigate the impact of the 2008 short-selling ban introduced in the UK on market liquidity. They analyse two samples. Stocks in the banned sample are also constituents of the FTSE 100. The non-banned stocks comprise the remaining FTSE 100 stocks. The sample period is divided into two pre-ban periods of 30 and 60 days preceding the introduction of the short-selling ban and a post-ban period of 30 trading days after the implementation of the ban. Their findings show that the average spread in the pre-ban period is similar for both samples. Nevertheless, after the introduction of the short-selling ban, the spread in banned stocks increases by 140% compared to the 56% increase in spreads in the non-banned stocks. Overall, spreads of stocks subject to the short-selling ban rise by 150% more than these of the stocks of the control sample. Hansson and Fors (2009) also examine the effects of the short-selling ban on the market quality of banned stocks in the UK. The treatment group comprises 23 banned stocks, while the control group comprises 321 stocks constituents of FTSE 350. The authors find that bid/ask spreads increase for both groups however the increase is greater for the banned stocks. In the post-ban period, bid/ask spreads drop in both groups. The drop is found significantly larger in banned stocks. Another study that examines the impact of short-selling rules on the quality of the UK markets is that of Marsh and Payne (2010). Each banned stock, 23 in total, is matched with 10 non-banned stocks used as a control sample with the most similar average market capitalizations. The average market capitalization of each stock is computed over the first half of The authors find that before the introduction of the short-selling ban the bid/ask spreads for both financial stocks and control group stocks is similar. During the ban period, bid/ask spreads rise to 35 basis points for the non-banned stocks, while those for banned stocks rise to 52 basis points. In the post-ban period, bid/ask spreads for the 17 banned stocks of small firms fall compared to their ban-period levels. However, banned stocks of 6 large firms face even greater bid/ask spreads in the post-ban period compared to their bid/ask spreads in the ban period. The authors also look at the impact of short-selling regulations on the limit order book liquidity provision. They find a collapse of depth in the order book at all 5

6 prices for the banned stocks. Both sides of the book appear to have suffered from liquidity drains. Depth deteriorates during the ban for control group stocks as well but the change in the book depth for banned stocks is much greater Multi-country study Beber and Pagano (2011) examine the effects of short-selling regulations on price discovery, liquidity, and the level of stock prices in 30 countries from January 2008 to June Their findings show that while bid/ask spreads increase when countries in the sample ban short-selling, the time pattern is also related to the financial crisis. Bid/ask spreads start increasing two weeks before countries ban short-selling. More specifically, for countries that banned short-selling of financial stocks the bid/ask spread is nearly double that of nonfinancial stocks. The authors also perform regression analysis to examine the impact of shortselling regulations on bid/ask spreads controlling for different forms of regulations and stock characteristics. They find that the naked and covered short-selling ban is associated with a greater increase in the bid/ask spread compared to the increase in the bid/ask spread under only naked short-selling ban. Disclosure requirements on the other hand appear to lower the bid/ask spread. Finally, the increase in bid/ask spread is greater for stocks with a low market value, high volatility, and no listed options. 3. HYPOTHESES DEVELOPMENT 3.1.Bid/ask spread Bid/ask spreads measure the costs of executing a trade. Liquidity providers, either designated market makers and/or traders that place limit orders, need to be compensated for the costs involved. According to theory (see Stoll, 1989) the bid/ask spread can be divided into three components. These are order processing costs (OPC), inventory holding costs (IHC), and adverse selection costs (ASC). Order processing costs arise from matching the buying and selling orders and the insurance for completing transactions by liquidity providers (Tinic, 1972). Inventory holding costs refer to liquidity providers holding of a portfolio of stocks that is not fully diversified in order to provide immediacy to investors (Stoll, 1978). Lastly, liquidity suppliers need to be compensated for the risk of trading with informed traders, which represents the adverse selection cost (Glosten and Milgrom, 1985; Voetmann, 2008)). The first two components comprise the non-informational part of the bid/ask spread and their impact on stock prices are temporary. The third component is the informational component of the bid/ask spread and its impact on the price is permanent. 6

7 There is scant theoretical and empirical literature investigating the impact of short-selling regulations on the components of the bid/ask spread during the recent financial crisis. Beber and Pagano (2011) provide some insights on how the 2008 short-selling bans introduced around the world could potentially affect the informational component or adverse selection cost and the non-informational component or the inventory holding cost. They make the point that in the absence of any exemption, market makers will find it more difficult to hedge their inventory position and this will increase inventory holding costs, widening spreads. Even if market makers are exempt from the short-selling ban, the reduced competition they will face from other liquidity suppliers will tend to widen spreads. Furthermore, prices will tend to be less informative in an environment where short-selling is prohibited and this will tend to increase the risk to market makers and other liquidity suppliers, again putting upward pressure on spreads. Diamond and Verrecchia (1987) show that short selling prohibitions will hinder the rapid adjustment of prices to new information, particularly bad news. Beber and Pagano (2011) observe that the delayed resolution of uncertainty will put upward pressure on adverse selection costs. However, they also observe that if the short sellers are informed traders, then the introduction of short-selling bans will tend to reduce trading activity by informed traders, putting downward pressure on adverse selection costs. Thus, the effects of 2008 short-selling regulations on the components of the bid/ask spread is an empirical issue. Based on the insights provided by Beber and Pagano (2011) I hypothesize that: H1: The non-informational component of the spread (the realized spread) of bank stocks subject to 2008 short-selling rules should increase during the ban. H2: The informational component of the spread (the price impact) of bank stocks subject to short-selling rules depends on which effect dominates: H2A: It increases if the effect of delayed resolution of uncertainty is stronger H2B: It decreases if the effect of reduced trading activity by informed traders is stronger. 3.2.Depth Existing theoretical research predicts that specialist depth quotes have a negative relationship with both adverse selection cost of the spread (see for example, Mann and Ramanlal, 1996; and Dupont, 2000) and the inventory-holding cost of the spread (see for example, Chordia, 7

8 Roll, and Subrahmanyam, 2000). Market makers and other liquidity providers have an incentive to offer less shares for trading at each quoted bid or ask price in the presence of greater information asymmetries and/or when managing their inventories is more expensive to minimize their costs. Moreover, Corwin (1999) finds that quoted depth is positively related to average trading frequency, market value of equity, and negatively related to standard deviation of return and share price. Prior studies (for example Clifton and Snape, 2008) find that the trading activity of stocks affected by the short-selling regulations decreases after the introduction of these regulations while volatility increases. Thus, if depth decreases when volatility increases and trading activity decreases I hypothesize that: H3: Depth decreases for bank stocks subject to short-selling rules after the implementation of these rules. 4. METHODOLOGY 4.1.Univariate analysis The quoted spread measures the cost of completing a buy and a sell, if trades are executed at the quoted prices. I use half-spreads to standardize on the friction associated with one trade because the spread is the cost of two trades (see Stoll, 2000). Moreover, effective bid/ask spread is decomposed into an informational (adverse selection component) and the noninformational component (inventory holding and order processing components) as presented in Bessembinder and Kaufman (1997). Below are the formulas that calculate the bid/ask spread and depth measures. (1) [ ] (2) where Ask it and Bid it are the posted ask price and bid price for stock i at time t and M it = is a proxy for the stock s true value (e.g., quote midpoint). The effective half-spread comprises a better measure of trading costs when transactions take place inside the quoted spread. It is computed as (3) [ ] (4) 8

9 where P it is the transaction price for security i at time t; D it is a trade indicator variable that equals 1 for buyer-initiated orders, i.e., when the transaction price is greater than the quote midpoint; and (-1) for seller-initiated orders, i.e., when the transaction price is less than the quote midpoint; and M it is an observable proxy for the true value of a stock i at time t (e.g., quote midpoint). Trading costs are divided into informational and non-informational components of trading costs, depending on price behavior following a transaction. The adverse selection cost that liquidity supplier incurs is estimated as (5) [ ] (6) where M it+n represents the stock s true value n periods after the transaction i.e., quote midpoint at time t+n. I use n equal to one minute. The realized spread on the other hand measures the non-informational component. It is calculated as (7) [ ] Depth is measured as the average number of shares that can be traded at the best bid and ask quotes (Heflin et al., 2005). (8) Euro depth is calculated as the average of the sum of the number of shares quoted at the ask price plus the number of shares quoted at the bid price, times their respective quoted prices (Heflin et al., 2005). 9

10 4.2.Multivariate analysis I regress the liquidity measures (i.e., percentage quoted, effective, and realized-half spread as well as price impact, depth and euro depth) on price, number of trades, and volatility, fundamental determinants that have been documented in the literature to affect liquidity measures. The regressions are estimated by using (pooled) OLS and fixed effects panel regression at the stock level on daily data. All variables undergo logarithmic transformation to reduce skewness and heteroskedasticity problems that might occur in the regression analysis. More specifically, the estimated model takes the following form: Where is the daily bid/ask spread measure ie., percentage quoted half spread (PQHS), percentage effective half spread (PEHS), percentage price impact (PPI), percentage realized spread (PRS), Depth, and Euro depth, respectively for stock i on day t. average daily transaction price. is the is the average daily number of trades. is measured as the logarithmic difference between the highest and lowest price on day t. Market_value is the market capitalization of the stocks under examination and it controls for the well documented relationship between the firm size and liquidity (see Boulton and Braga-Alves, 2010). is a dummy variable that takes the value of one when the introduction of new short-selling rules takes place and zero otherwise. For countries with a post-ban period available (i.e., the Netherlands, Switzerland, and the UK), there is a zero otherwise. dummy variable that takes the value of one after short-selling rules are lifted and 5. SAMPLE AND DATA I use intraday quote and trade data provided by Thomson Reuters Tick History (TRTH) database of SIRCA. My sample includes 90 bank stocks from 10 European countries. 1 The sample period is from mid-january 2008 to mid-february 2009 and it is divided into three sub-periods. Sub-period 1 (pre-ban period) is defined as the period before the introduction of new short-selling regulations. Sub-period 2 (ban period) is the period when short-selling regulations are introduced and remain in force. Finally, sub-period 3 (post-ban period) is defined as the period where short-selling regulations are lifted (see Table 1). 1 These countries are: Belgium, Denmark, France, Germany, Ireland, the Netherlands, Portugal, Spain, Switzerland, and the UK. 10

11 [Please insert Table 1 here] Raw data obtained from SIRCA undergo some filtering. I winsorize the data at 1% and 99% to remove outliers from the analysis. Daily data are then derived from the intraday datasets. Market value and number of shares outstanding used in the analysis are obtained from DATASTREAM. Three countries in my sample, i.e., Denmark, Switzerland, and the UK use a different currency than euro. I use the exchange rates provided by COMPUSTAT GLOBAL to convert the values expressed in domestic currencies into euro. SAS software is used for the whole analysis. 6. RESULTS 6.1.Descriptive statistics Table 2 reports the mean values of the daily transaction price, total daily volume (i.e., the average of daily total number of shares traded), daily number of trades, and volatility, which is measured as the logarithmic difference between the daily high and low prices for the three sub-periods: the pre-ban period, the ban period, and the post-ban period, respectively. All variables are aggregated across bank stocks traded in each country that were subject to shortselling rules introduced in As can be seen in Table 2, the average value of the transaction price significantly decreases for all the countries in the ban period. Transaction prices of Danish, Swiss, and British banks are expressed in euros. Qualitatively, price changes in Euros for these countries compared to the associated changes in domestic currencies remain the same. However, the UK is an exception. More specifically, the average transaction price of British bank stocks decreases when using the euro and increases when using the British pound in the post-ban period due to greater euro appreciation relative to the British pound during that period. Panel B of Table 3 shows that the prices increase slightly for the Netherlands but decrease in the case of Switzerland and the UK during the post-ban period. The mean of daily volatility reported in Table 2 Panel A is much greater during the ban period for all the countries. However, the increase in volatility for Portugal from to and Spain from to is negligible. In Table 2 Panel B, volatility drops for all the countries with post-ban period data available. The results are mixed for trading activity measures. For example, as shown in Table 2, during the ban period the mean value of the total daily volume increases for Belgian, Danish, Irish, Dutch, and Spanish bank stocks while it decreases for the rest of the countries. The number of 11

12 daily trades increases for Denmark and slightly for Ireland and the Netherlands while it decreases for the other countries. Again, in the case of Portugal and Spain there is negligible change in the number of trades between the pre-ban period and the ban period. More precisely, the average daily number of trades for Portugal decreases from 211 in the pre-ban period to 191 in the ban period. For Spain the related numbers of trades are 1587 and 1566, respectively. During the post-ban period the number of trades increases even more for the Netherlands compared to its ban period level. Swiss stocks experience even further drops in average daily number of shares in the post-ban period compared to their level in the ban period. Finally, the average daily number of shares increases for British stocks compared to their previous level in the ban period. [Please insert Table 2 here] Table 3 presents the medians of daily values of the transaction price, total daily volume defined as the average of daily total number of shares traded over the period under examination, daily number of trades, and volatility measured as the logarithmic difference between the daily high and low prices at a country level. The results are similar qualitatively to the results described in Table 2. The only difference is that the median value of the Volume variable for Spain decreases while the mean value increases during the ban. Also, the median value of the number of trades decreases for the Netherlands during the ban while the mean value increases. [Please insert Table 3 here] Overall, prices decrease, volatility increases, and trading activity decreases for most of the countries in the sample during the ban period. 6.2.Bid/ask spread measures in percentage terms Table 4 presents the means of the bid/ask spread measures in percentage terms. I use half spreads to standardize on the cost associated with one trade since the spread is the cost of two trades (Stoll, 2000). The sample period spans from mid-january 2008 to mid-december 2009 and it is divided into three sub-periods; i.e., the pre-ban period, the ban period, and the postban period, as described in section 5. The table is split in two panels. Panel A presents the results for countries that have two subperiods: the pre-ban period and the ban-period. These countries are: Belgium, Denmark, France, Germany, Ireland, Portugal, and Spain. Panel B contains results for countries with all 12

13 the sub-periods available. These countries are: the Netherlands, Switzerland, and the UK. Additionally, variable Diff is the difference between the values of the bid/ask spread measures between the pre-ban period and the ban period, and the ban period and the post-ban period (wherever available), respectively. I employ the t-test to test the statistical significance of the differences. P-values are given in parentheses. PQHS is the percentage quoted half-spread while PEHS is the percentage effective halfspread. According to Table 4, both the PQHS and PEHS increase for all countries during the new short-selling regime. In the post-ban period, as shown in Table 4 Panel B, the percentage bid/ask spread measures decrease for all the countries. [Please insert Table 4 here] Table 5 shows the results for the medians of the quoted and effective half-spreads in percentage terms. The statistical significance of the results is tested by using the Wilcoxon test with p-values given in parentheses. Both medians of percentage quoted and effective half-spreads increase for all countries. Regarding the post-ban period results, as shown in Table 5, Panel B, I find that there is a decrease in percentage bid/ask spread measures for all the countries. [Please insert Table 5 here] Results from earlier studies on the impact of 2008 short-selling regulations show that there was an increase in bid/ask spreads of the stocks subject to these regulations ((Boehmer, Jones et al., 2008; Clifton and Snape, 2008; Autore, Billingsley et al., 2009; Beber and Pagano, 2011; Boulton and Braga-Aves, 2010; Kolasinski, Reed et al., 2009). I also find that the PQHS and PEHS increase for all countries during the new short-selling regime. Moreover, the general trend in the behavior of the percentage bid/ask spread is an increase after the introduction of the new short-selling rules and a subsequent drop in the period when these rules are lifted for all the countries. These results support and enhance the findings of previous studies. I further decompose the bid/ask spread into its informational and non-informational components, respectively. Next section provides a detailed description of the univariate results of the components of the bid/ask spread measures in percentage terms at a country level. 13

14 6.3.Components of the bid/ask spread measures in percentage terms Table 6 presents the means of the components of the effective bid/ask spread as measured by the price impact and the realized spread in percentage terms. PPI refers to the percentage price impact, while PRS is the percentage realized spread. The PPI increases for all countries while PRS increases for all countries but Switzerland in the ban period. All the results are statistically significant. As can be seen in Table 6 Panel B the PPI decreases for all three countries. The PRS increases for Switzerland but decreases for the Netherlands and the UK. Only the decrease for the Netherlands is not statistically significant. [Please insert Table 6 here] Table 7 shows the medians of the components of the bid/ask spread as measured by price impact and realized spread. The statistical significance of the results is tested by using the Wilcoxon test providing p-values in the parentheses. The PPI increases for all ten countries during the ban. All the differences are statistically significant as can be seen in the same Table. The medians of the PRS increase for all countries but Switzerland. The differences are found statistically significant for all countries. For countries that have a post-ban period available, the results show that PPI decreases for all countries after the ban period. In the post-ban period the PRS decreases for the Netherlands and the UK but increases for Switzerland. Again the decrease for the Netherlands is not statistically significant. [Please insert Table 7 here] Overall, it is found that both price impact and realized spread after standardized by the quote midpoint increase during the ban with a few exceptions. The results support my hypotheses. As described in section 3, I would expect the non-informational component to increase after the introduction of the ban. The results for the informational component show that during the ban the effect of the delayed resolution of uncertainty, which exerts an upward pressure on the adverse selection component of the spread, is stronger than the effect of reduced trading activity by informed short sellers. In the post-ban period, I would expect both the price impact and the realized spread to decrease as short-selling regulations relax. The results support the hypothesis regarding the informational component but they are mixed regarding the non-informational component. Switzerland exhibits an inverse pattern. PRS of Swiss bank stocks decreases during the ban and increases in the post-ban period. 14

15 6.4.Depth Liquidity has two main dimensions (see for example Lee et al., 1993) Depth is one of them. Looking at both bid/ask spreads and depth measures provide a more detailed picture of any potential changes in liquidity during the sample period under examination. Table 8 shows the descriptive statistics of mean values of bid and ask prices as well as the number of shares available for trading at these prices-i.e., bid and ask sizes. Additionally, Table 8 reports the means of Depth and Euro depth variables and the statistical significance of the test for equality of their means. T-test is employed and the p-values of the t-statistic are given in the parentheses. Depth is the average number of shares that can be traded at the best bid and ask quotes. Euro depth is calculated as the average of the sum of the number of shares quoted at the ask price plus the number of shares quoted at the bid price, times their respective quoted prices (Heflin et al., 2005). As can be seen in Table 8, the bid and ask prices exhibit the same behavioral pattern as the transaction prices in Table 2. There is a significant decline during the ban period for all the countries. Depth decreases for seven out of ten countries. It increases for Belgium, Ireland, and Germany during the ban period. All the changes are statistically significant. In the postban period, the depth increases again for all these countries-i.e., the Netherlands, Switzerland, and the UK. I would expect reduced depth after the implementation of the short-selling ban as some traders would stay out of the market and thus there would be less liquidity providers and trading in general. Euro depth decreases during the ban as expected for all the countries because prices of the stocks under examination dropped significantly during that period. On the contrary, the Euro depth variable increases for all these countries in the post-ban period. [Please insert Table 8 here] Table 9 shows the medians of bid and ask prices, bid and ask sizes, as well as Depth and Euro depth values. Results for countries that have two sub-periods, the pre-ban period and the ban period (i.e., Belgium, Denmark, France, Germany, Ireland, Portugal, and Spain) are shown in Panel A. Results for the Netherlands, Switzerland and the UK, which have a postban period available are illustrated in panel B. Table 9 also reports the statistical significance of the test for equality of medians of Depth and Euro depth. The Wilcoxon test is used. P- values of the are given in the parentheses. Depth increases statistically significantly for Belgium, Ireland, and Germany while it decreases for France, Switzerland, and the UK. The increases are not statistically significant for Denmark, Portugal, and Spain 15

16 and the decrease is not statistically significant for the Netherlands. Euro depth decreases during the ban as expected for all the countries because prices of the stocks under examination drop significantly during that period. On the contrary, the Euro depth variable increases for all the countries in the post-ban period. [Please insert Table 9 here] To summarize, there is an unambiguous deterioration in liquidity during the ban period for French, British, and Swiss bank stocks. In these countries the spread increases while the depth decreases after the introduction of the short-selling regulations. The change in liquidity is ambiguous for Belgium, Ireland, and Germany where the increase in spread is also accompanied by an increase in depth. Liquidity appears to worsen for Denmark, Portugal, Spain, and the Netherlands in terms of wider spread and nearly unchanged depth when looking at medians. Overall, liquidity deteriorates for most of the countries after the implementation of short-selling rules. 6.5.Percentage bid/ask spread, price impact, and depth There are at least two dimensions to liquidity. A price dimension or bid/ask spread dynamics and a quantity dimension or depth dynamics. In order to evaluate any change in liquidity looking at both bid/ask spread and depth would be appropriate. An increase in bid/ask spread does not mean necessarily a deteriorated liquidity if it is accompanied by an increase in depth. According to existing literature (e.g., Lee et al., 1993) an unambiguous decrease in liquidity takes place when bid/ask spread increases and depth decreases. Alternatively, there is more liquidity when narrower bid/ask spread is accompanied by greater depth. In the present sub-section, an analysis is provided including the changes in both bid/ask spread and depth. I describe the results for the percentage effective half-spread (percentage quoted half spread could be used alternatively as it exhibits the same behavioral patterns with the percentage effective half-spread) and the two depth measures. Furthermore, according to the literature (Heflin and Shaw, 2000) market depth has a negative relationship with measures of asymmetric information. Price impact is used as a proxy for the informational component of the bid/ask spread. Greater price impact is expected to be accompanied by lower depth. All mean values of percentage effective half-spread (percentage price impact) increase during the ban. In order for liquidity to unambiguously decrease depth should decrease during the ban. When looking at the mean values of Depth as measured by the average number of 16

17 shares available for trade at best bid and ask prices, this is the case for Denmark, France, Portugal, Spain, the Netherlands, the UK, and Switzerland. The change in liquidity is ambiguous for Belgian, Irish, and German stocks where although the bid/ask spread (price impact) increases depth increases as well. For countries that have post-ban period data available (i.e., the Netherlands, Switzerland, and the UK) I find that the bid/ask spread (price impact) in percentage terms decreases while depth increases after the short-selling regulations are lifted. The results imply improved liquidity in the post-ban period. Scatter plot1 illustrates the change in percentage effective half-spread from the pre-ban period to the ban period versus the percentage change in depth from the pre-ban period to the ban period using mean values. On the vertical axis is the difference in percentage effective half-spread as given in Table 4. On the horizontal axis is the change in depth expressed in percentages. There are two groups of countries. In group one, which is depicted by the blue marker, both measures are found statistically significant. In group two, only the percentage change in the effective half-spread is statistically significant. [Please insert Scatter plot 1 here] When looking at medians, the UK, Switzerland, and France experience an increase in the bid/ask spread (price impact) and a decrease in depth during the ban period. Bid/ask spread (price impact) increases for Denmark, Portugal, Spain, and the Netherlands. This increase is associated with unchanged or statistically insignificant change in depth. For Belgium, Germany, and Ireland the bid/ask spread (price impact) increases with depth. Euro depth decreases with an increase in the percentage effective half-spread (price impact) for all the countries in the ban period. In the post-ban period, liquidity improves for all countries in terms of lower percentage bid/ask spreads. Depth increases statistically significantly for Switzerland and does not change statistically significantly for the Netherlands and the UK. Euro depth variable increases for all the countries. Scatter plot2 illustrates the change in percentage effective half-spread from the pre-ban period to the ban period versus the percentage change in depth from the pre-ban period to the ban period using median values. On the vertical axis is the difference in percentage effective half-spread as given in Table 5. On the horizontal axis is the change in depth expressed in percentages. There are two groups of countries. In group one, which is depicted by the blue 17

18 marker, both measures are found statistically significant. In group two, depicted in red color, only the percentage change in the effective half-spread is statistically significant. [Please insert Scatter plot 2 here] 6.6.Multivariate analysis Table 10 presents the results from a multivariate regression using stock-level fixed effects analysis. Quoted spread, effective half-spread, price impact and realized spread in percentage terms are regressed on price, number of trades, and volatility. Depth and Euro depth variables are also regressed on the same variables. Moreover, I use a dummy variable that equals one when short-selling regulations are in force and zero otherwise. For countries with post-ban period available, I use a post-ban dummy variable that equals one when short-selling regulations have lapsed and zero otherwise. The regressions are estimated by using OLS on daily observations. Fixed effects and robust standard errors clustered at the stock level are also used. Currency values for Denmark, Switzerland, and the UK are converted into euros. There are 90 bank stocks from 10 European countries pooled together. All the variables but volatility are logarithmically transformed. Volatility is already a logarithmic difference between high and low daily prices thus logarithmic transformation is redundant. Theoretically, one would expect a negative relationship between price and transaction costs i.e., spread. All the dependent variables (i.e., percentage quoted, effective, and realized spreads, price impact, depth and euro depth) are found to have a negative relationship with price. Trading activity is expected to have a negative relationship with bid/ask spread measures and positive relationship with depth measures. I find that this is the case. The coefficients of number of trades for PQHS, PEHS, PPI, and PRS have a negative sign while the coefficients of number of trades for Depth and Euro Depth have a positive sign. All the results are statistically significant. Market value is also expected to have a negative relationship with transaction costs and positive relationship with depth measures. This is the case for all the variables. All the results are statistically significant. Bid/ask spread measures are anticipated to have a positive relationship with volatility while depth measures negative. My results support the theoretical prediction and all the estimates are statistically significant. Based on the hypotheses developed in section 3, I would expect to find an increase in percentage quoted-half, effective-half, and realized-half spreads. For price impact, which proxies for the informational component of the bid/ask spread the result would depend on which effect is stronger as again specified in section 3. More precisely, I would expect an 18

19 increase if the effect of delayed resolution of uncertainty is stronger and a decrease if the effect of reduced trading activity by informed short sellers is stronger. The results support the first argument. The coefficient of the ban dummy variable is positive indicating an increase in the informational component of the spread during the ban. The ban dummy variable for depth measures appears with a negative sign which supports what I would expect to find i.e., lower depth during the ban. In the post ban period, I would expect to find a decrease in bid/ask spread measures, which I do. On the contrary, the depth measures seem to further decrease after the ban period which contradicts what I would expect to find. The sign of the estimates of the post-ban dummy variable for the depth measures is negative and the results are statistically significant. [Please insert Table 10 here] 7. CONCLUSIONS In this paper, I look at the impact of various short-selling regulations introduced in autumn 2008 on market liquidity of European bank stocks subject to these short-selling regulations. Prior studies examine the impact of 2008 short-selling rules on market liquidity focusing mostly on bid/ask spreads. The contribution of the present study is twofold. Firstly, I decompose the spread to study the behaviour of the components of the bid/ask spreads as measured by price impact and realized spread before, during, and after the introduction of short-selling rules. Secondly, bid/ask spread components analysis is accompanied by depth analysis which provides a more complete picture of what happened to liquidity in European markets after the implementation of new short-selling regulations. The main findings can be summarized as follows. Both informational and non-informational components of the bid/ask spread in percentage terms increase during the ban period and subsequently decrease in the post-ban period. In terms of price dimension of liquidity, percentage bid/ask spreads increase after the introduction of short-selling rules for all the bank stocks. In terms of quantity dimension of liquidity, median depth as measured by the average number of shares that can be traded at the best bid and ask quotes decreases on average for British, French, and Swiss stocks while it increases for Irish, Belgian, and German stocks during the ban period. Median Depth measure does not change statistically significantly for Dutch, Portuguese, Spanish, and Danish bank stocks. Consequently, liquidity unambiguously decreases in case of the UK, Switzerland, and France as well as in case of the Netherlands, Portugal, Spain, and Denmark however only in terms of higher percentage 19

20 bid/ask spreads. Impact on liquidity in Ireland, Belgium, and Germany is ambiguous as the increase in bid/ask spreads is associated with an increase in depth. Furthermore, multivariate analysis confirms my hypotheses with only exception the negative coefficient of the depth variables in the post-ban period. The main challenge of the research undertaken is the confounding effects of an unfolding global financial crisis during which and as a response to which various short-selling rules were introduced. Incorporation of analysis of stocks that were not affected by the new shortselling rules and comparison of the results with those of the bank stocks subject to shortselling regulations would partly ameliorate the confounding effects of the financial crisis which affected all stocks. 20

21 8. REFERENCES Autore, D., R. Billingsley, et al. (2009). "Short sale constraints, dispersion of opinion, and market quality: Evidence from the short sale ban on US financial stocks." Working Paper. Beber, A. and M. Pagano (2011). "Short-selling bans around the world: Evidence from crisis." Working Paper. Bessembinder, H. and H. M. Kaufman (1997). "A Comparison of Trade Execution Costs for NYSE and NASDAQ-Listed Stocks." Journal of Financial & Quantitative Analysis 32(3): Bessembinder, H. and K. Venkataraman (2009). "Bid/ask spreads: Measuring trade execution costs in financial markets." Working Paper. Boehmer, E., C. Jones, et al. (2008). "Which shorts are informed?" Journal of Finance LXIII(2): Boehmer, E. and J. Wu (2009). "Short selling and the informational efficiency of prices." Working Paper. Boulton, T. J. and M. V. Braga-Alves (2010). "The skinny on the 2008 naked short-sale restrictions." Journal of Financial Markets 13(4): Bris, A. (2008). "Short sellnig activity in financial stocks and the SEC July 15th emergency order." Working Paper. Chordia, T., R. Roll, et al. (2000). "Commonality in liquidity." Journal of Financial Economics 56(1): Clifton, M. and M. Snape (2008). "The effect of short-selling restrictions on liquidity: Evidence from the London Stock Exchange." Working Paper. Corwin, S. A. (1999). "Differences in Trading Behavior across NYSE Specialist Firms." Journal of Finance 54(2): Dupont, D. (2000). "Market Making, Prices, and Quantity Limits." The Review of Financial Studies 13(4): Glosten, L. and P. Milgrom (1985). "Bid, ask, and transaction prices in a specialist market with heterogeneously informed traders." Journal of Financial Economics 14: Hansson, F. and E. R. Fors (2009). "Get shorty? Market impact of the U.K. short selling ban." Working Paper. Heflin, F. and K. W. Shaw (2001). "ADVERSE SELECTION, INVENTORY-HOLDING COSTS, AND DEPTH." Journal of Financial Research 24(1): 65. Heflin, F. L., K. W. Shaw, et al. (2005). "Disclosure Policy and Market Liquidity: Impact of Depth Quotes and Order Sizes." Politique d'information et liquidité du marché: incidence des quantités cotées et de la taille des ordres. 22(4): Kolasinski, A., A. Reed, et al. (2009). "Prohibitions versus constraints: The 2008 short sales regulations." Working Paper. Lee, C. M. C., B. Mucklow, et al. (1993). "Spreads, Depths, and the Impact of Earnings Information: An Intraday Analysis." The Review of Financial Studies 6(2):

22 Mann, S. V. and P. Ramanlal (1996). "THE DEALERS' PRICE/SIZE QUOTE AND MARKET LIQUIDITY." Journal of Financial Research 19(2): 242. Marsh, I. W. and R. Payne (2010). "ning short sales and market quality: The UK's experience." Working Paper. Stoll, H. (1978). "The supply of dealer services in securities markets." Journal of Finance 33: Stoll, H. (1989). "Inffering the components of the bid/ask spread: Theory and empirical tests." Journal of Finance 46( ). Stoll, H. (2000). "Friction." Journal of Finance 55: Tinic, S. (1972). "The economics of liquidity services." Quarterly Journal of Economics 86: Voetmann, T. (2008). "Changes in the bid-ask components around earnings annoucements: Evidence from the Copenhagen stock exchange." Working Paper. 22

23 Table1 describes the type of short-selling regulations introduced, the duration of each sample sub-period, and the number of financial stocks affected by short-selling regulations for each country. Number of Sample sub-periods Type of shortselling ban affected ban financial stocks Country Belgium naked 4 15/01/ /09/ /09/ /12/2009 N/A France naked 11 15/01/ /09/ /09/ /12/2009 N/A Germany naked 10 15/01/ /09/ /09/ /12/2009 N/A Portugal naked 8 15/01/ /09/ /09/ /12/2009 N/A Spain disclosure requirements of net short positions 13 15/01/ /09/ /09/ /12/2009 N/A Denmark naked and covered 4 15/01/ /10/ /10/ /12/2009 N/A Ireland naked and covered 3 15/01/ /09/ /09/ /12/2009 N/A Netherlands naked and covered 7 15/01/ /10/ /10/ /05/ /06/ /12/2009 Switzerland naked and covered 5 15/01/ /09/ /09/ /01/ /01/ /12/2009 UK naked and covered 25 15/01/ /09/ /09/ /01/ /01/ /12/2009 Total 90 Table 2 presents the mean values of the daily transaction price, total daily volume-i.e., the average of total daily number of shares traded, daily number of trades, and volatility, which is measured as the logarithmic difference between the daily high and low prices at a country level for the three sub-periods: the pre-ban period, the ban period, and the post-ban period, respectively. The sub-periods for each country are defined in Table1. Panel A shows the figures for countries with two subperiods available. These countries are Belgium, Denmark, France, Germany, Ireland, Portugal, and Spain. Panel B reports the results for countries with three sub-periods available i.e., the Netherlands, Switzerland, and the UK. Price for Denmark, Switzerland, and the UK are expressed in euros. Table 2: Descriptive Statistics Panel A: Countries with two sub-periods (means) Country Sub-periods Price Volume Number of trades Volatility Belgium ,400,227 2, ,294,996 2, Denmark ,195, ,436, France ,034,071 4, ,908,344 3, Germany ,852,667 5, ,621,901 4, Ireland ,062, ,372, Portugal , , Spain ,608,691 1, ,492,565 1, Panel B: Countries with three sub-periods (means) Netherlands ,386,727 2, ,402,564 2, ,434,535 2, Switzerland ,372,115 4, ,675,814 3, ,377,490 2, UK ,584,974 2, ,844,873 1, ,836,158 2,