The relationship between short interest and stock returns in the Canadian market

Journal of Banking & Finance 29 (2005) 1729 1749 www.elsevier.com/locate/jbf The relationship between short interest and stock returns in the Canadian market Lucy F. Ackert a,b, *, George Athanassakos c,1 a Department of Economics and Finance, Michael J. Coles College of Business, Kennesaw State University, 1000 Chastain Road, Kennesaw, GA 30144, USA b Research Department, Federal Reserve Bank of Atlanta, 1000 Peachtree Street NE, Atlanta, GA 30309, USA c Ben Graham Chair in Value Investing, Richard Ivey School of Business, The University of Western Ontario, 1151 Richmond Street North, London, Ontario, Canada N6A 3K7 Received 19 August 2003; accepted 28 June 2004 Available online 23 September 2004 Abstract This paper provides new insight into the relationship between short sales and stock market returns using a sample of stocks sold short in Canada. Short interest is defined in relation to trading volume. The results strongly support the assertion that short sales and excess returns are contemporaneously negatively correlated in Canada. The paper further finds that excess returns are more negative for small firms because the supply of shortable shares is constrained for these firms. Excess returns are less negative for stocks with associated options and convertible bonds. Importantly, the evidence is consistent with the proposition that informed traders short sell Canadian interlisted stocks in Canada, rather than the US, to exploit lower execution costs. Together the results suggest that less restrictive regulation of short sales will improve the efficiency of markets. Ó 2004 Elsevier B.V. All rights reserved. * Corresponding author. Present address: Department of Economics and Finance, Michael J. Coles College of Business, Kennesaw State University, 1000 Chastain Road, Kennesaw, GA 30144, USA. Tel.: +1 770 423 6111; fax: +1 770 499 3209. E-mail addresses: lucy_ackert@coles2.kennesaw.edu (L.F. Ackert), gathanassakos@ivey.uwo.ca (G Athanassakos). 1 Tel.: +1 519 661 4096. 0378-4266/$ - see front matter Ó 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.jbankfin.2004.06.034

1730 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 JEL classification: G12; G14; G18 Keywords: Short interest; Informed traders 1. Introduction Selling borrowed stock (e.g., short selling) is a direct method to take advantage of declines in stock prices resulting from adverse information. Due to fear that shorting in a down market will precipitate a collapse in stock prices, the practice of selling stock short is constrained by securities law. These constraints affect the information content of stock transactions, as well as the speed of adjustment of stock prices to information. This paper provides new insight into the relationship between short sales and contemporaneous excess returns when markets are subject to short sales restrictions. Miller (1977) argues that stocks with severe short sales constraints are over-valued because only the most optimistic investors set the price. Moreover, because shorting is costly, relatively uninformed or liquidity traders will be less likely to sell stock short and only informed traders with strong negative information will be willing to bear the cost of short selling. According to Diamond and Verrecchia (1987), rational market participants incorporate this information into their trading decisions. A high level of short sales may serve as a signal that negative information is excluded from stock prices so that stocks are overvalued. Prices will adjust downward when this information becomes public. The more costly or constrained short selling becomes, the more likely it is that only those who expect to reap the greatest benefit (those with the largest negative information about a stock) will short sell. Brent et al. (1990) offer another perspective on short selling. They argue that some short selling results from arbitrage, hedging, and tax related activities and the behavior of these short sellersõ does not reflect negative information. We investigate the relationship between short interest, defined as the ratio of the number of shares short to trading volume, and contemporaneous excess returns for Canadian stocks. Unlike previous studies, we focus on the contemporaneous relationship and find strong evidence that short sales and excess returns are negatively correlated. Using semimonthly data for 1991 1994 and 1998 1999, we find that high short interest is contemporaneously associated with negative abnormal returns. 2 We are not claiming that negative excess returns result from high short interest, but rather that high short interest is an indicator that the stock is overvalued. Obviously, a short seller holds a short position because he believes the stock is overvalued. The behavior of short sellers is a signal to other investors of stock overvaluation. We also find that excess returns are positively related to firm size for our sample of shorted stocks. We further examine whether exchange-traded options and convertible bonds impact the relationship between short interest and excess returns for stocks sold 2 Our data set does not span the 1990s because the earlier data were collected by hand. As discussed subsequently, the later data are available electronically.

L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 1731 short. With derivatives trading, informed traders trade in other markets where transactions costs are lower so that the negative relationship between short interest and excess returns is moderated. In examining the effect of derivatives trading, we also recognize that a great deal of short selling activity in stock with associated derivatives is likely to be related to arbitrage and hedging activities and is not motivated by information. In addition to examining short sales in Canada, our data set affords a unique opportunity. We investigate how constraints on short sales affect stock returns by examining the performance of shorted Canadian stocks that are interlisted in the United States. As discussed subsequently, there are important differences in the rules governing short sales in Canada and the United States. The results are consistent with the notion that informed traders short sell in Canada because shorting stock is easier and the execution costs are lower compared to the United States. This paper is organized as follows. In Section 2, we review and compare short selling procedures in Canada and the United States. In Section 3, we briefly review the prior literature. In Section 4, we provide empirical predictions. In Section 5, we discuss the data selection process and present summary statistics on short interest in Canada. In Section 6, we present the empirical results. Finally, we provide concluding remarks in Section 7. 2. Short selling in the United States and Canada Aggregation and reporting of short interest positions are similar in the two counties, with the exception that the information is collected and reported on a semimonthly basis in Canada. In the US, the New York, American, and NASDAQ stock exchanges require brokers to file short positions as of the 15th of each month. Upon aggregation, information is publicly released within the next three or four days and is subsequently carried in widely read periodicals, including the New York Times and Wall Street Journal. The Toronto Stock Exchange requires brokers to file short sale reports on all short positions taken over a 15-day period within two business days after the 15th and month end. One day later the exchange makes these reports available to members and subscribers and the information is reported with a lag of approximately one week in national outlets such as the National Post/Financial Post and the Globe and Mail/Report on Business. Short selling is viewed with suspicion and the practice is strictly regulated in the United States and Canada. According to a recent Wall Street Journal article, short sellers are reviled for profiting from other investorsõ misery and accused of spreading innuendo and false-hoods to torpedo targets (Gasparion and McGough, 2000). One aspect of the regulation surrounding the short sale of stock results in increases in the execution cost of a sale. In the US, the Securities and Exchange Commission requires that short sales be made on a plus tick or a zero plus tick, meaning that a stock can be short sold only at a price above the last transaction price or at a price equal to the last transaction price if the most recent price movement was upward. This uptick rule did not apply to NASDAQ stocks until June 1994. Prior to

1732 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 June 1994 short sales could be made at a zero tick, meaning that a stock can be short sold only at a price equal to or above the price for the most recent trade on the exchange. In Canada, the Toronto Stock Exchange (TSX) requires short sales be made on a zero tick. In certain conditions, a stock can be sold short at prices below the last trade. 3 Examples include when a convertible security on the stock is owned and tendered, options, rights, or warrants are owned and exercised, or the trade is a program trade. Thus, execution costs are lower in Canada. Another important aspect of selling stock short is the requirement regarding the borrowing of stock to be short sold. Relative to the US, short selling in Canada is easier because short sellers can short without borrowing the stock first. In contrast, most short sellers in the US must make an affirmative determination before selling shares they do not currently own, i.e., they must borrow the stock before it can be sold (Kedrosky, 2003). 4 Thus, it is easier to short sell in Canada, particularly when identification of the shares borrowed is difficult. With no borrowing requirement, investors can more quickly respond to news in Canada. 5 Finally, there are some differences in the holding costs for short sellers in the US and Canada. In the US, the proceeds from the short sale are not available to the short seller, though large short sellers most often receive interest, called a rebate, on the proceeds (DÕAvolio, 2002). On the other hand, short sellers may pay a premium or negative rebate to borrow stock that is in short supply. Most small, individual investors receive no rebate (Asquith and Meulbroek, 1996). Short sellers are also required to deposit 50% of the market value of the stock as margin. In Canada, for stocks with a price of $2.00 or more, the additional margin required is also 50% (Canadian Securities Institute, 1996). 6 However, for stocks with associated options, the additional margin is only 30%. In summary, in the US and Canada short sales are constrained and discouraged by high execution and holding costs. In comparing the rules governing short sales across the two countries, we observe that it is easier to short sell stock in Canada. In Canada, execution costs are lower because the uptick rule does not apply and stock can be sold short at a price equal to the most recent execution price. Furthermore, shares do not have to be borrowed before they are sold short. Finally, margin requirements are similar in the two countries, with the exception that stocks with associated options have lower margins in Canada. 3 See rule number 4-301 of The Rules of The Toronto Stock Exchange. 4 Evans et al. (2003) point out that options market makers in the US can short sell without first identifying deliverable shares. 5 However, naked shorts can lead to short sellers paying a buy-in premium of 10 20% when they deliver the (hard to borrow) stock on settlement date. 6 Minimum credit balances in Canada for stocks with prices under $2 are higher (approximately 100% of market value). Although this suggests that low price stocks have higher margin requirements in Canada, further investigation indicates that the effective margin requirements are similar across Canada and the United States. For example, E*TRADE requires a margin of 100% of market value for stocks with prices less than $5.00. Other US brokers had similar requirements, though some expressly forbid short sales for low prices stocks (e.g., PreferredTrade.com).

L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 1733 3. Short interest and excess returns Restricting short sales impedes the adjustment of prices to information. According to Diamond and Verrecchia (1987), because short selling is costly, only those informed investors with large negative information who anticipate substantial profits will be willing to incur the cost of shorting. Short sellers are short because they predict a price decline. An unexpected increase in short sales is bad news for a stock because it reveals short sellersõ private information. Reducing the constraints imposed on short sellers and the cost of short selling will increase the speed of price adjustment to information, as well as the information reflected in stock prices. Several early studies examined the relationship between short interest and excess returns but a strong relationship was not consistently identified (Figlewski, 1981; Brent et al., 1990; Woolridge and Dickinson, 1994). One explanation for the inconsistent results is that these studies did not analyze the whole universe of firms with short interest or did not sample firms based on the level of short interest. Some used random samples of firms, and others used data reported in the media that included only a subset of short sold stocks. If the universe of firms with short interest is not analyzed, a study may omit firms with large or diverse levels of short interest and lack the ability to detect a significant relationship. More recent research suggests that short sellers are able to successfully identify poorly performing stocks (Asquith and Meulbroek, 1996; Dechow et al., 2001; Desai et al., 2002; Arnold et al., forthcoming). These studies define short interest as the ratio of shares short to the total shares outstanding and focus on stocks that are heavily shorted. Generally these are stocks with short interest of greater than 2.5%, which is about the top decline. Using intraday data, Aitken et al. (1998) confirm that short sales are instantaneously bad news for Australian stocks. In Australia, short sales are transparent immediately after execution. They further note that some previous research fails to find conclusive evidence in support of Diamond and VerrecchiaÕs hypothesis because of delays in reporting short interest of up to one month. We also focus on firms that attract the interest of short sellers to examine the relationship between the level of short interest and stock performance. However, there are several key differences between our approach and those taken previously. First, we focus on the level of short interest. Some earlier analysis has considered changes in short interest, which is more closely aligned to Diamond and VerrechiaÕs prediction that announcements of unexpected increases in short interest are bad news. While one may argue that changes in short interest move markets, the level of short interest reflects an active decision by traders to hold a short position open. The short position provides a signal that informed traders do not believe the stockõs price has fully adjusted to the bad news. If a short seller holds a short position and is correct in his belief that the stock is overvalued, the stock will perform poorly as the short position signals. In a recent study, Desai et al. (2002) also focus on the level of short interest but note that inferences based on changes in short interest are similar.

1734 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 Consistent with Desai et al., we find that excess returns and short interest are negatively correlated regardless of how we measure short interest. 7 In addition, unlike previous research, we define short interest as the ratio of the number of shares short to the semimonthly trading volume. We define short interest in relation to volume, rather than total shares outstanding, because volume reflects actual trading. From our perspective, trading volume and short sales reflect tradersõ observed decisions to buy or sell, unlike the number of shares outstanding, which may not be informative particularly in Canada. In Canada, the number of shares outstanding is not reflective of a stockõs actual float. A recent study shows that more than 20% of Canadian companies have dual class shares and a majority of traded companies have dominant shareholders (Pitts, 2002). Also unlike previous studies that investigate the relationship between short interest and subsequent excess returns (Asquith and Meulbroek, 1996; Desai et al., 2002), we focus on the contemporaneous relationship. We regress abnormal return on contemporaneous short interest and several additional explanatory variables, to be defined subsequently. 8 With reporting delays in Canada and the US, some market participants are aware of short trades well before the data are released. In fact, reported data reflects the settlement of a trade with the actual trades taking place three to five business days prior to settlement (Asquith and Meulbroek, 1996). 9 Aitken et al. (1998) conclude that in the Australian market where short sales transactions are transparent at execution, short sales are bad news. According to Fong et al. (2004), brokers or other managers may leak information on trades to investment managers. Although we cannot provide direct evidence regarding the speed of information dissemination of short sales information in Canada, the evidence presented subsequently is consistent with a strong contemporaneous relationship. 10 Furthermore, Canadian brokers have indicated in private conversations that sharing information regarding short sales is not uncommon. Finally, unlike many previous studies, our sample better reflects the universe of stocks short sold in Canada. The set of high short interest stocks, as defined in the earlier literature, is a limited subset of all stocks that are sold short. To focus on this subset reduces the ability to detect a significant relationship. Our sample selection process is more inclusive, as we describe after outlining our empirical predictions. 7 We re-estimated the regressions reported subsequently with the dependent variable defined as the number of shares short (# shares short) and the change in short interest (D short interest). The results are generally consistent with those reported in the paper for our short interest measure, defined as the number of shares short divided by trading volume. Most importantly, the estimated coefficient on short interest is significantly negative in every case. 8 Our goal is not to examine the announcement affects of the level of short interest, as would be done using an event study methodology. Rather, our goal is to examine whether stocks that are shorted perform poorly, contemporaneously. 9 In June 1995 the settlement delay for trades in Canada and the US was changed to three business days from the five business days prevailing in earlier years. 10 We repeated all analyses reported subsequently using lagged and lead short interest. The relationship between contemporaneous short interest and excess returns is much stronger, as there is weak statistical evidence of a lagged relationship and no relationship at all for a lead relationship.

L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 1735 4. Empirical predictions Earlier empirical studies have presented evidence that short interest is bad news for a stock (Asquith and Meulbroek, 1996; Aitken et al., 1998; Desai et al., 2002). We examine the relationship between short sales and contemporaneous excess returns in Canada to provide further insight into market outcomes with short sales constraints. Based on a sample of semi-monthly Canadian data, we examine whether contemporaneous excess returns are negatively associated with short interest. To this end, we regress abnormal return on contemporaneous short interest and several additional explanatory variables, to be defined subsequently. The extant literature suggests a negative relationship between short interest and excess returns, so that our first null hypothesis is H 1,0 : There is no relationship between the excess returns of shorted stocks and short interest. The prior literature provides directional predictions regarding the effects of additional explanatory variables on excess returns. We include market capitalization as a proxy for the difficulty in obtaining shortable shares. Because firm size proxies for the supply of shortable shares, it also proxies for the cost of short selling (Boehme et al., 2002). 11 According to Boehme, Danielson, and Sorescu, a positive relationship between excess returns of shorted stocks and market value is expected because it is easier to identify shares that can be borrowed for large firms. Also, recall that short sellers may have to pay a penalty to short stock that is difficult to borrow. In addition, naked shorts may eventually lead to short sellers paying a buy-in premium of 10 20% when they will be required to deliver the (hard to borrow) stock on settlement date, thus adding another cost to short selling of smaller, less liquid, stocks. For small firms, an informed trader must have very negative information if he is to trade a stock with short sales constraints. Thus, the contemporaneous excess return is expected to be more negative for small firms. Our second hypothesis, based on the arguments of Boehme et al. (2002), is H 2,0 : There is no relationship between the excess return of shorted stocks and market value. We also investigate whether exchange-traded options or convertible bonds impact the relationship between excess returns and short interest. The literature has recognized that the availability of traded options reduces the cost of establishing a short position when short sales are constrained if options transactions can be effected at lower cost. Figlewski (1981) argues that stocks subject to short sales constraints are overpriced so that high short interest is associated with lower risk-adjusted return. Option trading reduces overpricing because traders with negative information about the stock can short sell indirectly. If negative information is impounded in 11 Market capitalization has been used in the literature to measure other correlated constructs, in addition to the difficulty in obtaining shortable shares. Thus, care should be exercised in interpretation.

1736 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 stock prices through option trading, the negative relationship between short interest and excess returns should be mitigated. In other words, options reduce the informational inefficiency resulting from short sales constraints in the stock market (Figlewski and Webb, 1993). Options create a method by which short sales constrained traders with negative information can act on their information, and, thus, option trading enhances the informational efficiency of the stock market. Investors with negative information could short sell, but if constraints on short selling result in prohibitive costs, they transact in the option market, taking a long put position, for example. Thus, we expect trading of options to mitigate the negative relationship between excess return and short interest. Our third hypothesis, based on the arguments of Figlewski (1981) and Figlewski and Webb (1993), is H 3,0 : The presence of tradable options will not affect the relationship between the excess return of shorted stocks and short interest. Options trading may be motivated by negative information. However, much short selling in stocks with options, and also convertible bonds, is likely to be related to arbitrage and hedging activities, rather than being motivated by information. 12 Some traders may short sell a stock with options or convertibles when they have no negative information in order to effect an arbitrage or hedging transaction. As argued by Brent et al. (1990), traders take short positions for tax and arbitrage reasons, in addition to speculative reasons. They find higher short interest for stocks with options and convertibles, supporting the proposition that arbitrage activities are significant. Because positions taken for tax or arbitrage reasons do not reflect negative information, we expect to find that the effect of short interest on excess returns is smaller for stocks with traded options and convertible bonds. For options, this effect further mitigates the relationship between excess return and short interest, as postulated in our third hypothesis. For convertibles, our fourth hypothesis, based on the arguments of Brent et al. (1990), is H 4,0 : The presence of convertible bonds will not affect the relationship between the excess return of shorted stocks and short interest. In our examination of the relationship between short interest and excess returns, we also investigate whether changing constraints on short sales affects the excess returns of shorted stocks by examining the performance of Canadian stocks that are interlisted in the United States. Both countries have large and well-developed markets in which securities are lent (Bris et al., 2002). However, as discussed previously, there are important differences in the rules governing short sales in Canada and the United States. Although there are considerable restrictions on short sales in both 12 A recent Globe and Mail article provides an example. The arbitragers buy the convertible debentures and simultaneously short sell St. Laurent stock... The bulk of it does not reflect bearish sentiment on St. Laurent prospects... Instead most of short shares are in hands of a small number of hedge funds that are involved in a complicated arbitrage play connected to the companyõs outstanding convertible debentures (Northfield, 1996).

L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 1737 countries, short sales in Canada are both easier and less costly. Consequently, informed traders will take advantage of negative information faster and with lower cost by trading interlisted stocks in Canada. 13 Thus, we expect to find more negative contemporaneous excess returns for shorted Canadian interlisted than non-interlisted stocks. Furthermore, we expect to find that the effect on excess returns of interlisting is more negative for the latter sample years (after June 1994). Until June 1994 shorted stocks in the NASDAQ market were not subject to the uptick rule in the US. After June 1994, execution costs increased in the US for these stocks and, thus, traders could take advantage of negative information concerning interlisted stocks in Canada faster and with lower cost. Accordingly, following June 1994, we can expect a shift in selling activity for Canadian interlisted stocks traded on NASDAQ to Canada along with associated negative returns. Our fifth hypothesis is H 5,0 : Interlisting of Canadian stocks in the United States will have no effect on the excess return of shorted stocks. In the following section we describe our sample of shorted stocks. 5. Descriptive information on short interest in Canada 5.1. Data sources and sample selection Short sales data are directly from the Toronto Stock Exchange (TSX) for the periods January 1991 December 1994 and January 1998 December 1999. Only ordinary common stocks are included in the analysis. Data from January 1991 to December 1994 are obtained in hard copy. More recent data from January 1998 to December 1999 are provided by the TSX electronically. Importantly, during the earlier sample period, our database includes information when the number of shares short exceeds 5000 or the net change in shares short exceeds ±5000. 14 The more recent data include all firms with positive shares shorted or non-zero net change in shares shorted during the semimonthly time period. Short sales data are semi-monthly at the 15th and end of each month with semi-monthly returns calculated by compounding daily returns. Additional stock data, including stock returns, trading volume, shares outstanding, market price per share, and beta are from the Canadian Financial Markets Research Centre (CFMRC). Information on whether a firm is interlisted, trading 13 In the analysis reported subsequently, we do not control for changes in the trading volume of interlisted stocks across Canada and the United States. Increased volume is not a necessary implication of informed trading. While there is an empirical link between volume and prices, the theory concerning relative changes in volume is not so clear. In KyleÕs (1985) model of informed trading volume has no impact on the price adjustment process. 14 The hand-collected information excludes very small short sales to keep the task manageable. Hand collection of the earlier data was highly labor intensive. Ensuring accuracy was of utmost concern, which made the process extremely time consuming. Although our data do not span the 1990s, our sample better represents the universe of short sold stocks because it is not limited to high short interest stocks.

1738 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 volume on the US and Canadian exchanges, and the presence of outstanding options or convertible bonds is from the TSX Index Review. The market value of equity is obtained by multiplying the stock price at the beginning of each semi-monthly period by shares outstanding for the corresponding period. All data are adjusted for stock splits and stock dividends. In total, our sample includes 72,021 observations for 1789 firms. We carefully match semi-monthly short interest data to the excess returns observed over the same semi-monthly time period. Using the CFMRC database, we compute three excess return series. First, we use a market-adjusted returns model and calculate the excess return (ER MM ) as the difference between the return on each common stock and the return on the value-weighted CFMRC universe stock index during each semimonthly period. Second, we calculate excess returns based on the Capital Asset Pricing Model (ER CAPM ). In applying the CAPM, we use CanadaÕs Treasury long-term bond yield to measure the risk-free interest rate and the value weighted CFMRC universe stock index as the market portfolio. 15 The Treasury bond yield and value-weighted index data are from the CFMRC. 16 A third measure of excess returns follows recent work by Lyon et al. (1999). Lyon et al. question the power and specification of traditional modeling and encourage the measurement of abnormal returns based on reference portfolios. Here, the reference portfolio is a portfolio of stocks with size similar to that of the shorted stock in question and, in this way, we estimate size-adjusted relative performance (ER SizeAdj ). 17 Unlike previous studies that use an arbitrary measure to identify heavily shorted stocks, we divide shorted stocks into quartiles based on the magnitude of short interest. We employ the following procedure in forming the quartiles. We rank all stocks in the sample each year by short interest from low to high and then divide the ranked firms into four groups of equal size. We repeat this for each sample year and then group all data into four quartiles from low to high. Membership in a group changes each year because short interest changes from year to year. Inclusion in a quartile depends on a stockõs short interest in relation to that of other stocks. Because the level of short interest changes over time, an arbitrary measure across time for all sample stocks would be inappropriate. 15 The CAPM is estimated using the trailing 60 months of returns data, if available. If a 60-month history is unavailable, the model is estimated with as few as 24 trailing monthly observations. 16 In general, the results reported subsequently are similar if we use an equally weighted CFMRC index, a value weighted CFMRC index, or the TSX 300 value weighted index. 17 Size-adjusted returns are constructed using size reference portfolios. First, we calculate firm size (market value of equity calculated as price per share times shares outstanding) at the beginning of each semi-monthly period for all firms in the CFMRC database. We then rank all CFMRC firms on the basis of firm size at the beginning of each semi-monthly period and form size-quartile portfolios based on these rankings. We obtain equally weighted returns using all stocks in each size-quartile. According to size, sample stocks are placed in the appropriate CFMRC size-based slot. Size adjusted excess returns are calculated by subtracting the equally weighted returns for each size quartile to which our sample stocks belong from the returns of the shorted stocks.

L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 1739 Table 1 Distribution of short interest across sample years Year Number of observations Mean Median 25th percentile 75th percentile 1991 6154 1.548 0.115 0.031 0.421 1992 5930 1.492 0.096 0.027 0.329 1993 9367 1.005 0.079 0.022 0.287 1994 11,686 5.626 0.167 0.038 0.686 1998 19,984 4.871 0.050 0.006 0.273 1999 18,900 6.625 0.034 0.004 0.204 The table reports the level and distribution of short interest for sample firms. Data are available for January 1991 through December 1994 and January 1998 through December 1999. Data are semi-monthly at the 15th and end of each month. Short interest is defined as the ratio of the number of shares short to the semimonthly trading volume. For each sample year, the table reports the number of observations, the mean and median short interest, and the short interest at the 25% and 75% percentiles. 5.2. Summary statistics For our sample period, the CFMRC database includes information for 1806 stocks. Our sample includes data for 1789 firms so that most stocks are sold short at some time. The median market capitalization of CFMRC firms over our sample period is $105,665,000, the median semi-monthly trading volume is 271,695, and the median volume to shares ratio is 0.0115. As can be seen from Table 2, our shorted stocks have higher median market capitalization, volume, and volume to shares ratio than the CFMRC firms despite the fact that most CFMRC stocks have been shorted at some point. This is to be expected, however, as stocks are shorted because they are overvalued and large trading activity takes place as price moves toward the equilibrium. For TSX stocks, the number of shares sold short has steadily increased over time. Table 1 reports the level and distribution of short interest for sample firms. As discussed above, short interest is defined as the ratio of the number of shares short to the semimonthly trading volume. For each sample year, Table 1 reports the number of observations, the mean and median short interest, and the short interest at the 25% and 75% percentiles. Over all sample years, we observe significant short interest, though with wide dispersion in the degree of shorting across sample firms. 18 Table 2 reports summary statistics for our sample, which includes 72,021 observations. The median values for several variables are reported by short interest quartiles where short interest is defined as the ratio of the number of shares short to the semimonthly trading volume. 19 Note that the number of shares sold short and the ratio of the number of shares short to shares outstanding increase monotonically with our measure of short interest. There is no clear pattern in the change in shares sold short across short interest quartiles. Similarly, we do not see a striking relationship 18 Lower median short interest in 1998 1999 is not unexpected because the data includes all short sales, as compared to 1991 1994 where the data includes short sales of 5000 shares or more. 19 We report medians, rather than means, because the distributions are highly skewed.

1740 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 Table 2 Summary statistics by short interest quartiles Variable Short interest quartile v 2 test Q1 (low) Q2 Q3 Q4 (high) ShortInterest 0.0049 0.0315 0.1435 0.9547 45,562.0 ** # Shares short 4500 12,000 42,169 128,700 14,571.0 ** Change in # shares short 100 110 900 74 873.6 ** # Shares short/outstanding 0.0001 0.0005 0.0016 0.0060 20,941.0 ** Volume 685,231 389,374 301,731 104,994 142.4 ** Volume/shares outstanding 0.0218 0.0146 0.0111 0.0056 5089.6 ** Market value of equity (000Õs) 130,623 147,957 181,416 151,044 128.1 ** Beta 1.032 1.016 0.991 0.970 65.1 ** ER MM 0.0045 0.0053 0.0085 0.0111 64.6 ** ER CAPM 0.0038 0.0043 0.0071 0.0100 53.9 ** ER SizeAdj 0.0080 0.0086 0.0116 0.0146 47.8 ** The table reports summary statistics for sample firms. Data are available for January 1991 through December 1994 and January 1998 through December 1999. Data are semi-monthly at the 15th and end of each month. The median values for several variables are reported by short interest quartiles where short interest is defined as the ratio of the number of shares short to the semimonthly trading volume. Short interest data are from the Toronto Stock Exchange. Trading volume, shares outstanding, market value of equity (in thousands), and beta are from the CFMRC. The table also reports contemporaneous excess returns using a market-adjusted returns model (ER MM ), the Capital Asset Pricing Model (ER CAPM ), and size-adjusted returns (ER SizeAdj ). The final column reports a v 2 test of the null hypothesis of no differences across quartiles. *, ** indicates significance at the 5%, 1% level. between firm size as measured by the market value of equity and short interest across the quartiles. Finally, we observe monotonic declines in trading volume, the ratio of trading volume to shares outstanding, and beta as short interest increases. 20 Importantly, contemporaneous semi-monthly excess returns decline as short interest increases, whether measured using a market-adjusted returns model (ER MM ), the Capital Asset Pricing Model (ER CAPM ), or size-adjusted excess returns (ER SizeAdj ). Brown Mood non-parametric v 2 tests, reported in the final column of Table 2, indicate that all median values are significantly different across quartiles at p < 0.0001. 21 Table 3 reports additional summary information for sample firms. In Table 3, statistics are reported for the 46,756 observations with no missing data as this panel is the basis for the regression analysis reported subsequently. The median values for several variables are reported for firms based on the characteristics of their traded 20 The decline in beta across short interest quartiles may seem counterintuitive. Note, however, that thin trading is more prevalent in the high short interest stocks (see volume/shares outstanding in Table 2). Dimson (1979) shows that estimates of beta are biased downward with thin trading, and the bias is larger with thinner trading. Although estimates of beta can be adjusted for thin trading, our approach is conservative. That we find support for our hypothesis in presence of thin trading is even stronger support for our findings. 21 Because the distributions are highly skewed, we use the Brown Mood test, an approximate v 2 test that does not rely on normality.

L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 1741 Table 3 Summary statistics by security characteristics Variable Options ConvBonds Interlist No Yes No Yes No Yes # Observations 42,735 4021 44,831 1925 35,591 11,165 ShortInterest 0.0928 0.0981 * 0.0880 0.7504 ** 0.0907 0.1009 ** # Shares short 20,000 204,020 ** 21,511 493,300 ** 18,100 68,603 ** Change in # shares short 0 0 0 0 ** 0 0 # Shares short/outstanding 0.0009 0.0024 ** 0.0010 0.0120 ** 0.0009 0.0015 ** Volume 271,187 2,460,221 ** 314,276 638,208 ** 248,800 748,556 ** Volume/shares outstanding 0.0111 0.0213 ** 0.0119 0.0134 ** 0.0107 0.0158 ** Market value of equity (000Õs) 128,314 1,912,178 ** 146,442 349,026 ** 113,504 489,860 ** Beta 0.986 1.182 ** 1.014 0.969 ** 0.965 1.188 ** ER MM 0.0082 0.0047 * 0.0078 0.0140 ** 0.0078 0.0086 ER CAPM 0.0071 0.0046 ** 0.0066 0.0133 ** 0.0065 0.0077 ER SizeAdj 0.0095 0.0072 ** 0.0091 0.0154 ** 0.0086 0.0106 The table reports summary statistics for sample firms. Data are available for January 1991 through December 1994 and January 1998 through December 1999. Data are semi-monthly at the 15th and end of each month. First the table reports the number of observations for each category. The median values for several variables are reported for firms based on the characteristics of their traded securities, including whether options (Options) are traded on the stock, whether a firm has convertible bonds (ConvBonds) outstanding, and whether the stock is interlisted on a US stock exchange (Interlist). Short interest is defined as the ratio of the number of shares short to the semimonthly trading volume. Short interest data are from the Toronto Stock Exchange. Trading volume, shares outstanding, market value of equity (in thousands), and beta are from the CFMRC. The table also reports contemporaneous excess returns using a market-adjusted returns model (ER MM ), the Capital Asset Pricing Model (ER CAPM ), and size-adjusted returns (ER SizeAdj ). We test for a difference in each variable across characteristics using a v 2 test of the null hypothesis of no difference. Significant differences in a variable across Options, Convbonds, and Interlist are indicated by asterisks in the Yes column. *, ** indicates significance at the 5%, 1% level. securities, including whether options (Options) are traded on the stock, a firm has convertible bonds (ConvBonds) outstanding, and the stock is interlisted on a US stock exchange (Interlisted). The table also reports excess returns using a market-adjusted returns model (ER MM ), the Capital Asset Pricing Model (ER CAPM ), and sizeadjusted returns (ER SizeAdj ). We test for a difference in the median value of each variable across characteristics using a v 2 test of the null hypothesis of no difference. Consistent with earlier research, we observe higher short interest for firms with associated options and convertible bonds (Figlewski and Webb, 1993). Though excess returns are less negative for firms with options, excess returns are more negative for firms with convertible bonds. The latter finding is inconsistent with our expectation that trading in convertible bonds moderates excess returns. We observe that stocks with options in our sample are significantly larger than the average stock in our sample. Consistent with our expectation that short selling activity will migrate to the Canadian market, we observe higher short interest for interlisted firms. While, as expected, excess returns are more negative for interlisted than non-interlisted stocks, the difference in excess returns is not statistically significant. Formal statistical tests of our predictions follow.

1742 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 6. Results In order to more formally examine the relationship between contemporaneous excess returns of shorted stocks and short interest, we estimate the following regression: ER i;t ¼ Intercept þ b 1 LSI i;t þ b 2 LMV i;t þ b 3 LSIxOpt i;t þ b 4 LSIxCB i;t þ b 5 Interlist i;t þ e i;t ; ð1þ where ER i,t is the semi-monthly excess return for firm i at time t. The independent variables include the natural log of short interest (LSI i,t ) defined as the ratio of the number of shares short to the semimonthly trading volume, the natural log of the market value of equity (LMV i,t ), LSIxOpt i,t defined as LSI i,t multiplied by a dummy variable that equals 1 if options are traded on the stock and 0 otherwise, LSIxCB i,t defined as LSI i,t multiplied by a dummy variable that equals 1 if a firm has convertible bonds outstanding and 0 otherwise, and Interlist i,t defined as a dummy variable that equals 1 if the stock is interlisted on a US stock exchange and 0 otherwise. 22 The effect of short interest is measured by (b 1 + b 3 ) for stocks with options, (b 1 + b 4 ) for stocks with convertible bonds, (b 1 + b 3 + b 4 ) for stocks with options and convertible bonds and b 1 for stocks with neither. LSIxOpt i,t and LSIxCB i,t are interactive terms whose coefficients capture the joint effect of short interest on excess returns in the presence of options and convertible bonds. 23 Table 4 reports the estimates of regression (1). We estimate the regression separately for each year in the sample, 1991 1994 and 1998 1999. The table reports estimates for excess returns using a market-adjusted returns model (ER MM ) in Panel A, the Capital Asset Pricing Model (ER CAPM ) in Panel B, and size-adjusted returns (ER SizeAdj ) in Panel C. Below each coefficient estimate, we report t-statistics in parentheses. Because the possibility of cross-sectional correlation in the residuals in a given year loomed, we used an approach suggested by Fama and MacBeth (1973) to avoid the possibility of bias in the estimated standard errors. 24 We estimate regression (1) for each semi-monthly period, and then average the semi-monthly estimates across time. Significance levels are based on pooled t-statistics, computed as follows: t j ¼ b j;t p r j = ffiffiffi ; ð2þ T 22 To moderate the effect of outliers and because all our variables are bounded by zero, we use a natural log transformation. 23 Because market value is included as a control for the effect of the supply of shortable shares, it is not modeled as an interaction term. Similarly, we do not model interlisting using an interaction term because the variable is included to examine whether excess returns differ for shorted firms that are interlisted, as compared to non-interlisted shorted firms. 24 Though, pooled standard errors are likely to be underestimated, inferences based on pooled Ordinary Least Squares estimates are similar.

L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 1743 Table 4 Yearly estimates of the relation between excess returns and short positions Independent Year variables 1991 1992 1993 1994 1998 1999 Panel A: Excess returns are measured using a market-adjusted returns model (ER MM ) Intercept 0.0341 0.0411 0.0359 0.0386 0.0901 0.0840 ( 2.57) * ( 2.45) * ( 2.87) ** ( 2.60) * ( 4.63) ** ( 3.80) ** LSI 0.0024 0.0038 0.0044 0.0009 0.0013 0.0039 ( 2.84) ** ( 5.30) ** ( 4.30) ** ( 2.31) * ( 1.81) ( 6.80) ** LMV 0.0016 0.0019 0.0016 0.0018 0.0044 0.0036 (2.50) * (2.46) * (2.50) * (2.68) * (4.93) ** (3.77) ** LSIxOpt 0.0023 0.0042 0.0030 0.0009 0.0019 0.0015 (2.62) * (4.97) ** (3.07) ** (0.66) (1.26) ( 0.83) LSIxCB 0.0110 0.0125 0.0025 0.0011 0.0004 0.0008 (2.85) ** (3.28) ** ( 1.01) (0.80) (0.21) (0.46) Interlist 0.0022 0.0009 0.0017 0.0014 0.0088 0.0088 ( 0.61) ( 0.27) ( 0.45) (0.62) ( 3.39) ** ( 2.78) * Adjusted R 2 0.0073 0.0085 0.0088 0.0029 0.0082 0.0086 Panel B: Excess returns are measured using the Capital Asset Pricing Model (ER CAPM ) Intercept 0.0337 0.0354 0.0474 0.0432 0.0799 0.0825 ( 1.92) ( 1.84) ( 2.70) * ( 2.40) * ( 3.70) ** ( 3.75) ** LSI 0.0022 0.0042 0.0042 0.0010 0.0014 0.0040 ( 2.39) * ( 4.78) ** ( 3.71) ** ( 2.29) * ( 2.58) * ( 6.66) ** LMV 0.0015 0.0015 0.0021 0.0020 0.0038 0.0035 (1.77) (1.69) (2.44) * (2.39) * (3.86) ** (3.55) ** LSIxOpt 0.0019 0.0039 0.0039 0.0012 0.0013 0.0011 (1.83) (4.21) ** (3.56) ** (0.98) (1.01) ( 0.59) LSIxCB 0.0086 0.0145 0.0029 0.0011 0.0006 0.0002 (2.18) * (3.32) ** ( 1.25) (0.87) (0.27) ( 0.06) Interlist 0.0018 0.0001 0.0009 0.0041 0.0084 0.0094 ( 0.47) (0.02) (0.23) (1.68) ( 2.98) ** ( 2.84) ** Adjusted R 2 0.0083 0.0124 0.0112 0.0042 0.0072 0.0091 Panel C: Excess returns are measured using size-adjusted returns (ER SizeAdj ) Intercept 0.0794 0.1500 0.1560 0.0287 0.0207 0.1139 ( 3.67) ** ( 5.91) ** ( 5.19) ** ( 1.82) ( 0.84) ( 5.96) ** LSI 0.0033 0.0038 0.0051 0.0008 0.0009 0.0047 ( 3.42) ** ( 5.18) ** ( 3.95) ** ( 1.61) ( 1.02) ( 7.77) ** LMV 0.0036 0.0070 0.0071 0.0013 0.0010 0.0048 (3.50) ** (5.72) ** (4.98) ** (1.71) (0.97) (5.53) ** LSIxOpt 0.0028 0.0056 0.0042 0.0010 0.0018 0.0010 (2.80) * (5.25) ** (2.85) * (0.59) (0.88) ( 0.51) LSIxCB 0.0148 0.0098 0.0014 0.0015 0.0009 0.0015 (3.90) ** (2.53) * ( 0.54) (1.06) (0.41) ( 0.69) Interlist 0.0002 0.0004 0.0032 0.0004 0.0069 0.0069 ( 0.03) (0.09) ( 0.80) (0.15) ( 2.58) * ( 1.86) Adjusted R 2 0.0111 0.0189 0.0193 0.0017 0.0053 0.0082 (continued on next page)

1744 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 Table 4 (continued) The table reports estimates of the coefficients of (1) using averages of semi-monthly OLS estimates. The estimates are reported for each year in the sample, 1991 1994 and 1998 1999. Data are semi-monthly at the 15th and end of each month. Panel A of the table reports estimates for contemporaneous excess returns using a market-adjusted returns model (ER MM ), Panel B reports estimates based on the Capital Asset Pricing Model (ER CAPM ), and Panel C reports estimates with size-adjusted returns (ER SizeAdj ). The market returns are value weighted. The independent variables include the natural log of short interest (LSI i,t ) defined as the ratio of the number of shares short to the semimonthly trading volume, the natural log of the market value of equity (LMV i,t ), LSIxOpt i,t defined as LSI i,t multiplied by a dummy variable that equals 1 if options are traded on the stock and 0 otherwise, LSIxCB i,t defined as LSI i,t multiplied by a dummy variable that equals 1 if a firm has convertible bonds outstanding and 0 otherwise, and Interlist i,t defined as a dummy variable that equals 1 if the stock is interlisted on a US stock exchange and 0 otherwise. The t-statistic is reported in parentheses. *, ** indicates significance at the 5%, 1% level. where the numerator is the average of the semi-monthly coefficient estimates for a particular independent variable (j), r is the standard deviation of the coefficient estimates of a particular variable, and T is the number of semi-monthly sample periods. Table 4 shows that across all sample years and all excess returns measures, the results support our prediction that short interest and excess returns are negatively correlated. For all sample years and excess return measures, the estimated coefficient of short interest is negative with statistical significance at p < 0.05 in most years (15 of 18 estimates). Also as predicted, the market value coefficient is consistently positive and significant at p < 0.05 in most cases (14 of 18 estimates). The supply of shortable shares is constrained for firms with lower market values so that excess returns are less negative for larger firms. The yearly estimated coefficients for options, convertible bonds, and interlisted stocks are not consistently significant in every sample year. However, the estimates have the predicted sign when significant. Our results are consistent with the notion that the NASDAQ change in the uptick rule had an important effect on trading behavior of NASDAQ interlisted Canadian stocks. Recall that prior to June 1994 stocks that traded in the NASDAQ market were not subject to the uptick rule. After June 1994, NASDAQ stocks became subject to the uptick rule, resulting in higher execution costs in the US. The yearly average semi-monthly estimates of the effect of interlisting on the excess returns of shorted stocks indicate a significant difference in the later sample years. In 1998 and 1999 the effect of interlisting is significantly negative across excess returns measures (5 of 6 estimates at the 5% significance level and 6 of 6 at the 10% significance level), whereas in earlier sample years we do not observe a statistically significant influence. We conducted additional analysis to further examine the effect of interlisting. First, within 1994, we compared the excess returns of interlisted firms before and after June 1994. The excess returns for interlisted stocks for July through December 1994 were more negative than those for January through June 1994, with significance at the 1% or 5% level for two of three measures. In addition short interest was significantly higher for the interlisted stocks during the later six-month period at the 1% level. After June 1994, execution costs increased in the US for NASDAQ stocks so traders could take advantage of negative information concerning interlisted stocks in Canada faster and with lower cost. Second, we compared the US trading volume of

L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 1745 our sample stocks interlisted on NASDAQ in 1994 and 1999. In May 1994 (May 1999) 37.2% (55.9%) of our interlisted stocks were traded on NASDAQ indicating that a large part of our (interlisted) sample consists of stocks interlisted on NAS- DAQ. In May 1994, the average traded value in the US of our NASDAQ interlisted stocks as a percentage of total traded value was 37.5%, whereas in May 1999 it was only 17.5%. The average trading volume for Canadian stocks interlisted on NAS- DAQ is significantly lower in the US in May 1999 than in May 1994 at p < 0.01. These results are consistent with the summary results reported in Table 2 and our hypothesis that traders moved transactions to Canada after the change in the uptick rule because investors with negative information found it faster and less costly to trade in Canada. Table 5 reports the coefficients from a cross sectional, times series regression of excess semimonthly returns for our sample of shorted stocks. We report estimated coefficients and t-statistics for the cross sectional, time series data using standard pooled Ordinary Least Squares (OLS) estimation (Greene, 1993, p. 464). Tests for serial correlation and heteroskedasticity did not indicate that an alternative to OLS was necessary. 25 However, as mentioned above, the possibility of cross-sectional correlation in the residuals in a given year loomed. To avoid the possibility of bias in the estimated standard errors, we follow Fama and MacBeth (1973) and estimate regression (1) for each semi-monthly period, and then average the semimonthly estimates across time. We see strong support for our predictions. For both estimation methods (time series, cross sectional OLS or an average of semi-monthly OLS estimates) and all measures of excess return, the relationship between short interest and the excess returns of shorted stocks is negative. Short sold stocks in our sample experience more negative contemporaneous excess returns as short interest increases. In contrast to earlier studies that examine the whole stock universe (Reinganum, 1981), but consistent with our expectations, we find that firm size is positively related to excess returns for our sample of short sold stocks. In general, the excess returns for firms with options and convertible bonds are less negative. Though the effect of convertible bonds is not always significant with the pooled OLS method, the average semi-monthly estimates are all significant at the 1% level. 26 For our sample of stocks that were sold short, excess returns are more negative for interlisted firms as compared to non-interlisted, though the interlisted variable is not always significantly different from zero. Further consideration of the coefficient estimates provides important insight into the effect of trading in options and convertible bonds on the relationship between short interest and excess returns. Recall that LSIxOpt i,t and LSIxCB i,t are interaction terms whose coefficients capture the joint effect on excess returns of short interest in 25 We employed the SPEC and DW options in the REG SAS procedure to test for heteroskedasticity and serial correlation. 26 We re-estimated the regressions replacing the dummy variable for convertible bonds with a dummy variable that equals one for firms with convertible bonds or convertible preferred stock, and zero otherwise, and the results are similar to those reported in the paper.

1746 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 Table 5 The relation between excess returns and short positions Independent variables Dependent variable ER MM ER CAPM ER SizeAdj Pooled OLS Average semi-monthly Pooled OLS Average semi-monthly Pooled OLS Average semi-monthly Intercept 0.0626 0.0540 0.0611 0.0537 0.0801 0.0929 ( 15.49) ** ( 7.67) ** ( 13.82) ** ( 6.75) ** ( 16.87) ** ( 8.75) ** LSI 0.0022 0.0028 0.0023 0.0028 0.0026 0.0031 ( 11.68) ** ( 8.79) ** ( 10.82) ** ( 8.39) ** ( 11.64) ** ( 8.19) ** LMV 0.0029 0.0025 0.0028 0.0024 0.0036 0.0042 (13.79) ** (7.69) ** (12.24) ** (6.48) ** (14.52) ** (8.50) ** LSIxOpt 0.0014 0.0018 0.0015 0.0018 0.0019 0.0024 (2.57) * (3.34) ** (2.62) ** (3.54) ** (2.95) ** (3.63) ** LSIxCB 0.0012 0.0039 0.0008 0.0036 0.0033 0.0045 (1.32) (3.28) ** (0.81) (2.80) ** (3.16) ** (3.57) ** Interlist 0.0047 0.0035 0.0038 0.0024 0.0042 0.0027 ( 4.59) ** ( 2.66) ** ( 3.54) ** ( 1.73) ( 3.47) ** ( 1.78) Adjusted R 2 0.0047 0.0074 0.0047 0.0087 0.0068 0.0109 61.96 ** 49.62 ** 64.88 ** The table reports the coefficients from a regression of excess semi-monthly returns. Data are available for January 1991 through December 1994 and January 1998 through December 1999. Data are semi-monthly at the 15th and end of each month. The table reports estimates for contemporaneous excess returns using a market-adjusted returns model (ER MM ), the Capital Asset Pricing Model (ER CAPM ), and size-adjusted returns (ER SizeAdj ). The market returns are value weighted. The table reports cross sectional, times series OLS estimates of the coefficients, in addition to estimates using averages of semi-monthly OLS estimates. The independent variables include the natural log of short interest (LSI i,t ) defined as the ratio of the number of shares short to the semimonthly trading volume, the natural log of market value of equity (LMV i,t ), LSIxOpt i,t defined as LSI i,t multiplied by a dummy variable that equals 1 if options are traded on the stock and 0 otherwise, LSIxCB i,t defined as LSI i,t multiplied by a dummy variable that equals 1 if a firm has convertible bonds outstanding and 0 otherwise, and Interlist i,t defined as a dummy variable that equals 1 if the stock is interlisted on a US stock exchange and 0 otherwise. The t-statistic is reported in parentheses. *, ** indicates significance at the 5%, 1% level. the presence of options and convertible bonds. The effect of short interest is measured by (b 1 + b 3 ) for stocks with options, (b 1 + b 4 ) for stocks with convertible bonds, (b 1 + b 3 + b 4 ) for stocks with options and convertible bonds and b 1 for stocks with neither. Notice that while the effect of options alone does not eliminate the negative relationship between short interest and excess returns, when we look at firms that have convertible bonds or options and convertible bonds, the negative effect of short interest is always eliminated. For example, from Table 5 and using excess returns from a market-adjusted returns model (ER MM ), we see that the average semimonthly estimate of the negative effect of short interest ( 0.0028) is weakened, but not eliminated, by options (0.0018). However, when we look at firms that have convertible bonds (0.0039) or options and convertible bonds (e.g., 0.0018 + 0.0039), the negative effect of short interest is eliminated not only in this case, but also when using the other measures of excess returns.

L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 1747 Because excess returns are affected by numerous variables, the explanatory power of the regressions reported in this paper is low, as measured by the adjusted R 2. 27 However, the results provide strong support for our empirical predictions. The predicted relationships find statistical support in our estimates. 28 7. Discussion and concluding remarks This paper reports the results of an examination of the relationship between excess returns and short interest in Canada. The results strongly indicate that stocks that are sold short perform poorly, contemporaneously. Importantly, this finding does not lead to the conclusion that the Canadian market functions inefficiently. Rather, it points to the significant impact of short sales constraints on pricing. Short sales constraints limit the ability of traders to use information. For our sample of shorted stocks, firm size is positively related to excess returns. The supply of shortable shares is less constrained for large firms. We find that shorted stocks with options and convertibles have less negative excess returns. We find that interlisted stocks have more negative excess returns for our sample of shorted stocks. Our results are consistent with the hypothesis that investors shift trade based on negative information from the US market to the Canadian market because the execution cost of short selling is lower in Canada. The impact is important in later sample years because shorted stocks on the NASDAQ became subject to the uptick rule. Nevertheless, further investigation of how the behavior of short sellers in the NASDAQ market was affected by the change in the uptick rule in June 1994 is warranted. Together, our results indicate that regulators might well reconsider strict regulation of the practice of selling short. The findings reported in this paper are consistent with theory that argues that information dissemination is swifter in markets with less stringent regulation on short sales activities. Moreover, lowering the cost of short selling by relaxing constraints on this activity will allow stock prices to reflect a fuller set of information and result in more efficient market pricing. Acknowledgement The views expressed here are those of the authors and not necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve System. Financial support provided by the Federal Reserve Bank of Atlanta and Wilfrid Laurier University is 27 To ensure that a few extreme outliers do not drive the results, and thus lead to the low adjusted R 2, we repeated all analyses excluding the largest 10% outliers. The results are similar for the pooled and average semi-monthly regressions to those reported in Table 5 and also for the average semi-monthly, yearly regressions reported in Table 4. 28 See Figlewski and Webb (1993) who discuss the lack of explanatory power in their regressions. Even with low adjusted R 2, estimated coefficients are generally significant and have the expected sign.

1748 L.F. Ackert, G. Athanassakos / Journal of Banking & Finance 29 (2005) 1729 1749 gratefully acknowledged. The authors thank Despina Athanassakos, Robert Fedrock, and Rob Kalbfleisch for research assistance and three anonymous referees, Neil Brisley, Bryan Church, Scott Hein, Brian Smith, and seminar participants at the Helsinki School of Economics and University of Western Ontario for helpful comments. References Aitken, M.J., Frino, A., M.S., McCorry, P.L., Swan, 1998. Short sales are almost instantaneously bad news: Evidence from the Australian Stock Exchange. Journal of Finance 53 (6), 2205 2223. Arnold, T., Butler, A.W., Crack, T.F., Zhang, Y., forthcoming. The information content of short interest: A natural experiment. Journal of Business. Asquith, P., Meulbroek, L., 1996. An empirical investigation of short interest, Working paper, Harvard Business School, Harvard University. Boehme, R.D., Danielson, B.R., Sorescu, S.M., 2002. Short-sale constraints and overvaluation, Working paper, DePaul University. Brent, A., Morse, D., Stice, E.K., 1990. Short interest: Explanations and tests. Journal of Financial and Quantitative Analysis 25 (2), 273 289. Bris, A., Goetzmann, W.N., Zhu, N., 2002. Efficiency and the bear: Short sales and markets around the world, Working paper, Yale School of Management. Canadian Securities Institute, 1996. Short selling of equities, The Canadian Securities Course Textbook, Chapter 6, Part 3, 6-33 to 6-37. Dechow, P.M., Hutton, A.P., Meulbroek, L., Sloan, R.G., 2001. Short-sellers, fundamental analysis, and stock returns. Journal of Financial Economics 62, 77 106. Desai, H., Ramesh, K., Thiagarajan, S.R., Balachandran, B.V., 2002. An investigation of the informational role of short interest in the Nasdaq market. Journal of Finance 57 (5), 2263 2287. DÕAvolio, G., 2002. The market for borrowing stock. Journal of Financial Economics 66, 271 306. Diamond, D.W., Verrecchia, R.E., 1987. Constraints on short-selling and asset price adjustment to private information. Journal of Financial Economics 18 (2), 277 311. Dimson, E., 1979. Risk measurement when shares are subject of infrequent trading. Journal of Financial Economics 7 (2), 147 226. Evans, R.B., Geczy, C.C., Musto, D.K., Reed, A.V., 2003. Failure is an option: Impediments to short selling and options prices, Working paper, The Wharton School, University of Pennsylvania. Fama, E.F., MacBeth, J.D., 1973. Risk, Return, and Equilibrium, Empirical Tests. Journal of Political Economy 81 (3), 607 636. Figlewski, S., 1981. The informational effects of restrictions on short sales: Some empirical evidence. Journal of Financial and Quantitative Analysis 16 (4), 463 476. Figlewski, S., Webb, G.P., 1993. Options, short sales, and market completeness. Journal of Finance 48 (2), 761 777. Fong, K., Gallagher, D.R., Gardner, P., Swan, P.L., 2004. A Closer Examination of Investment Manager Herding Behavior, Working paper, The University of New South Wales. Gasparion, C., McGough, R., 2000. A short-seller trades blows with a target who just wonõt fold. Wall Street Journal (June 26), A1:6 A6:1. Greene, W.H., 1993. Econometric Analysis, second ed. Macmillan Publishing Company, New York. Kedrosky, P., 2003. Selling short-selling short, National Post (Financial Post section, March 25), FP15. Kyle, A.S., 1985. Continuous auctions and insider trading. Econometrica 53 (6), 1315 1335. Lyon, J.D., Barber, B.M., Tsai, C.-L., 1999. Improved methods for tests of long-run abnormal stock returns. Journal of Finance 54 (1), 165 201. Miller, E.M., 1977. Risk, uncertainty, and divergence of opinion. Journal of Finance 32 (4), 1151 1168. Northfield, S., 1996. St. Laurent strong despite short sellers, Globe and Mail, January 23, B19. Pitts, G., 2002. Share system called open to abuse, Globe and Mail/Report on Business, November 4, B5.