Short selling in OTC stocks: Informative or manipulative?

Short selling in OTC stocks: Informative or manipulative? Archana Jain Assistant Professor Saunders College of Business Rochester Institute of Technology Rochester, NY 14623 Voice: 901-652-9340 Email: ajain@saunders.rit.edu Chinmay Jain* Assistant Professor Faculty of Business and IT University of Ontario Institute of Technology Oshawa, Ontario L1H 7K4 Voice: 901-652-9319 Email: chinmay.jain@uoit.ca July 2017 * Corresponding author 1

Short selling in OTC stocks: Informative or manipulative? Abstract We examine short selling activity in OTC stocks using short volume data for a period from 2009 to 2016. We find that on average short volume constitutes about 24 percent of daily trading volume in OTC stocks on non-zero trading volume days. A portfolio with short position in stocks with non-zero short volume and long position in stocks with zero short volume has significant positive future returns over a period of 1 month. However, a portfolio with short position in stocks with highest quintile of non-zero short volume and long position in lowest quintile of nonzero short volume has positive return only for 1-day and 5-day holding periods, and it turns negative for 1-month holding period. Following Diether, Lee, and Werner (2009), we test an alternate hypothesis that short sellers are opportunistic risk-bearers and provide liquidity in times of volatility. We find that short sellers are able to predict future negative returns for 1-day and 5- day period after controlling for volatility and turnover of stocks, but high short selling predicts a positive return for a 1-month period. These results suggest a price-manipulating behavior of short sellers in OTC stocks that may need further scrutiny. Keywords: Short selling; Over-the-counter stocks; Price manipulation; JEL Classifications: G10, G12 2

1. Introduction Several studies have examined the asset pricing dynamics in the over-the-counter (OTC) market. OTC markets are less liquid, disclose less information, and have lower institutional ownership compared to listed stocks (Ang, Shtauber, Tetlock (2013)). Based on their study, Ang, Shtauber, and Tetlock (2013) provide evidence that OTC stocks have higher short sale constraints and higher dispersion of opinion, making them overpriced and resulting in negative returns, on average. Eraker and Ready (2015) document that OTC stocks are generally small companies that have a high probability of failure and a small probability of success. The stocks in their sample have large negative average returns. This also makes OTC stocks an ideal target for short sellers as they profit from a fall in stock prices and with the failure of a company their profits would be much more pronounced. Eraker and Ready (2015) also suggest that the low average return of OTC stocks reflects investors preferences for positively skewed stocks. Engelberg, Reed, and Ringgenberg (2013) find that a substantial portion of short sellers trading advantage comes from their ability to analyze public information. OTC markets are less transparent and it makes it harder for short sellers to trade based on their ability to process public information. If short sellers trade in OTC markets, they may do so based on their private information or they may do so to take advantage of falling stock prices by pushing them further down temporarily. Thus, it is interesting to study the short-term and long-term profitability of their trades in OTC stocks. Ang, Shtauber, and Tetlock (2013) studies short selling in OTC stocks for a sample of 50 stocks and they find that Fidelity allows retail investors to short sell only 1 of those stocks (and only 8 out of 50 similar sized listed-stocks). They also look at the short interest data for those 50 stocks and based on a smaller float as a % of shares outstanding, conclude that investors have difficulty in short selling OTC stocks. They do not study the actual short selling volume data and thus, they provide only a snapshot of short selling in OTC markets. 3

We aim to study short selling on a larger scale and to find the extent of short selling in OTC stocks and its impact on the performance of OTC stocks. OTC markets are less transparent, less liquid, and are not as regulated as exchange-traded stocks. Thus, acquiring information about these stocks should result in potentially higher returns for traders. Also, as these markets are less regulated, it makes it easier for short sellers to manipulate stock prices in the downward direction. Another stream of research focuses on the relationship of short selling activity with returns based on superior information of short sellers. These studies find that heavily shorted stocks underperform lightly shorted stocks (Boehmer, Jones, and Zhang, 2008) and short sellers are correctly able to predict stocks that will have future abnormal negative returns (Diether, Lee, and Werner, 2009). Some studies also find that aggressive short selling can amplify the magnitude of price declines (Shkilko, Van Ness, and Van Ness, 2013). All of these studies have examined stocks that are traded on exchanges, are liquid, and have more transparent financial disclosures. We are interested in analyzing OTC stocks, which have different characteristics. These stocks are less liquid and disclose less financial information to the public compared to the exchange-traded stocks. This makes OTC stocks more vulnerable to be targets of price manipulation based on misleading information (Aggarwal and Wu (2006) and Nelson, Price, and Rountree (2013)). Since short sellers have been blamed for causing bear raids, and OTC stocks are more susceptible to manipulation, it is naturally interesting to study the impact of short selling on the OTC stocks. In this paper, we contribute to the literature by studying the short selling activity in the OTC stocks. We find that the short volume constitutes 23.68 percent of trading in OTC stocks on 4

non-zero trading volume days, and short sales constraints do not seem binding in these stocks that are part of our sample. 1 Second, we test the predictability of future returns based on short volume in OTC stocks. We find that a portfolio with short position in stocks with non-zero short volume and long position in stocks with zero short volume has significant positive future returns for 1-day, 5-day, and 20-day holding periods. We also double sort the stocks using both size and short volume and find similar results for all size categories. However, a portfolio with short position in stocks with highest quintile of non-zero short volume and long position in stocks with lowest quintile of nonzero short volume has significant positive future returns only for 1-day and 5-day holding periods. These returns become negative for a 1-month (20-day) holding period, specifically for low and medium sized firms. Third, we study the determinants of short volume in OTC stocks. Short selling in OTC stocks is contrarian in nature; OTC stocks have higher short volume after a period of positive returns and on days of positive returns. Stocks with higher past and current volatility and higher market capitalization also have higher short volume. Lastly, we test an alternate opportunistic risk-bearing hypothesis that short sellers provide liquidity in times of volatility. We find that volatile stocks have negative future returns, but short selling remains a significant predictor for 1-day and 5-day negative returns after controlling for volatility and turnover, but high short selling does not predict negative returns for a 1-month period. 1 Short selling volume in FINRA daily RegSHO data might be overstated because a market maker with a customer order to sell may sell short against their quote (which gets reported to the tape as a media transaction) and then buy from the seller on their books in a non-media transaction. Short selling as part of the market making process need to be split out from the daily numbers as mentioned by the CEO (Cromwell Coulson) of OTC Markets Group. However, our return predictability results should still hold as we don t believe that the short selling will be overstated more in some stocks and less in others stocks. 5

2. Literature Review and hypothesis development 2.1 Informative role of short sellers Short selling allows a trader (the short seller) to borrow a stock and sell it without actually owning it. After a short seller takes a short position in a stock, at a later date, the short seller has to close the position by purchasing the shares. Short sellers make a profit or loss based on the difference between their selling and buying prices. Because short sellers have to buy the shares after the sell, they expect the price to go down in the future. Using data for the year 2005, Diether, Lee, and Werner (2009) find that short sales represented 31 percent of share volume for NASDAQ-listed stocks and 24 percent of share volume for NYSE-listed stocks, which shows the pervasiveness of short selling activity. These numbers are even higher for recent periods. For example, Jain, Jain, and McInish (2012) find that the short selling constitutes about 40% of trading volume. Many studies argue that short sellers are informed and that they act based on their private information. Miller (1977) hypothesizes that in a market without short selling, the demand for a security will come from the most optimistic investors; this causes the stocks to become overvalued and the stocks experience subsequent negative abnormal returns to correct this overvaluation. Diamond and Verrecchia (1987) argue that short sellers are traders who are motivated by information rather than liquidity needs because they cannot use the proceeds of the sale. Bris and Goetzmann (2007) find that in countries where short sales are more prevalent, prices incorporate negative information at a faster rate. Boehmer, Jones, and Zhang (2008) study short sales using proprietary NYSE order data from 2000 to 2004 find that heavily shorted stocks underperform lightly shorted stocks, indicating that short sellers are well informed. They also find that short sellers are important contributors to efficient stock prices. Diether, Lee, and 6

Werner (2009) also find that short selling predicts negative abnormal future returns. They formulate a trading strategy consisting of a short position in stocks with high short-selling activity and a long position in stocks with low short-selling activity and obtain a significant positive abnormal return of about 1.4 percent per month. Thus, short selling is informative and appears to play a positive role in financial markets. Engelberg, Reed, and Ringgenberg (2013) find that a substantial portion of short sellers trading advantage comes from their ability to analyze publicly available information. They do not find any evidence that short sellers anticipate news. They find that short selling increases following news events. Thus, they find that short sellers trade on publicly available information. Even though most papers seem to argue in favor of short selling activity, there are some arguments against the role of short sellers in financial markets. Brunnermeier and Pedersen (2005) argue that large institutional investors sell because of liquidity needs. Short sellers who are actively watching the market may then exacerbate the price declines by short selling, suggesting that short sellers follow manipulative and predatory trading strategies, leading to less informative prices. Blau and Wade (2012) find that short-selling patterns surrounding both analyst downgrades and upgrades are symmetric suggesting that short selling is not informative. Informative short selling would have led to higher short selling before the downgrades and lower short selling before the upgrades. Henry and Koski (2010) find that short selling around the issue date of seasoned equity offerings is manipulative and not informative. Regulators also banned short selling during the financial crisis of 2008 and implemented a new regulation restricting short selling (Rule 201) in 2011. The controversy over short selling surfaced as far back as the 16 th century when a group of Dutch businessmen shorted shares of the East India Company in a so-called bear raid. Bear raids were very common in the 1900s, when groups of traders would 7

target vulnerable stocks held by long investors in margin accounts. That results in margin calls forcing the long investors to sell their shares to cover their margin calls, which leads to further price declines. The short sellers would then buy shares at lower prices to cover their positions. More recently, short sellers were blamed by regulators for causing an artificial price decline in stocks during the financial crisis of 2008. Soon after removing the uptick rule in 2007, the SEC took an about turn in its stance on short selling by implementing a short selling ban on 797 financial stocks in 2008. In 2011, the SEC approved Rule 201, which places restrictions on short selling in stocks that have a 10 percent intraday decline. Jain, Jain, and McInish (2012) find that this rule does not serve its intended purpose as short selling was already lower on the days of a large decline in stock price. Thus, short sellers do not seem to be exacerbating stock price declines. Nevertheless, regulators are still considering further attempts at curbing short selling. 2.2 OTC stocks Relatively few studies have shed light on trading behavior in OTC stocks. Ang, Shtauber, and Tetlock (2013) test theories of cross-sectional return premiums in OTC markets. They find that OTC premium for illiquid stocks is quite high, the OTC premiums for size, value, and volatility are similar to listed stocks, and the OTC premium for momentum is three times lower than listed stocks. They find that premiums for illiquidity, size, value, and volatility are largest among those OTC stocks that are held almost exclusively by retail investors and those that do not disclose financial information. They suggest that these returns premiums can be explained by theories of differences in investors opinions and short sales constraints. They also find that the majority of investors in the OTC firms are retail investors and suggest that as stock lending is mostly done by institutions, short selling of OTC stocks is difficult. White (2016) uses a proprietary database of transaction-level OTC data and confirms that average return in OTC 8

investment is severely negative. He also finds that older, retired, low-income, and less educated investors experience poorer outcomes in OTC stock markets. Luft, Levine, and Larson (2001) find that a portfolio composed of OTC-BB stocks yields a lower return and higher risk than a similar portfolio composed of stocks listed on stock exchanges. They also find that the returns of OTC stocks do not co-move with the returns of stocks listed on exchanges. Luft and Levine (2004) sort OTC stocks by size and liquidity and find that focusing on larger firms is a better investment strategy than focusing on most liquid stocks. Several studies have examined that OTC stocks are less liquid compared to exchangelisted stocks and spreads increase for stocks after they delist. Sanger and Peterson (1990), Harris, Panchapagesan, and Werner (2008), and Macey, O Hara, and Pompilio (2008) find that spreads increase for firms that move from an exchange to OTC markets. These studies suggest that the shift in trading from exchanges to OTC markets makes stocks less liquid as these firms have to disclose less information now resulting in higher information asymmetry. Though Bushee and Leuz (2005) document that after June 2000, regulations on OTC markets have increased substantially. OTC markets organize firms into three markets; OTCQX, OTCQB and Pink. To qualify for OTCQX, companies must meet high financial standards, follow best practice corporate governance, demonstrate compliance with U.S. securities laws, and have a professional third-party sponsor introduction. The OTCQB market is for entrepreneurial and development stage U.S. and international companies that are unable to qualify for OTCQX. The Pink market offers trading in a wide spectrum of equity securities through any broker. This market is for all types of companies that are there by reasons of default, distress or design. 9

Eraker and Christie (2015) examine the return on OTC stocks over a period from 2000 to 2008 and find that these returns are extremely negative on average. They also find that the distribution of returns on OTC stocks has a positive skewness, which might be a possible explanation for negative premium on these stocks. They also find that delisted stocks perform much better than the native stocks that were never listed on stock exchanges. In another paper by Nofsinger and Varma (2014), they find that OTC trades do not have much information content. Another set of papers examine manipulation in stock markets and identify OTC markets are easier to manipulate. Aggarwal and Wu (2006) study 142 cases of stock market manipulation from 1990 to 2001 and find that most manipulation cases happen in OTC Bulletin Board and the pink sheets that are small and illiquid. Böhme and Holz (2006) study stock spams that advertise stocks traded on the OTC market. They find that spam message campaigns are associated with an increase in trading activity along with positive cumulative abnormal return shortly after the publication of the messages. In a similar study, Frieder and Zittrain (2007) find that stocks experience a significantly positive return on days prior to heavy touting by spam. The trading volume increase following touting. Returns in the days that follow touting messages are negative. This suggests that the spammers tout in order to increase the trading volume so that they can unload the position they took before touting. These studies show evidence that market manipulation can affect OTC stocks. 3. Data We obtain the daily short volume data for OTC stock from the FINRA website for the period from August 2009 to January 2016. This dataset also includes total trading volume for the stocks. If a stock has no short volume for a day, then the stock will not appear in the FINRA 10

database for that day. Therefore, we manually assign a value of zero as short volume to those missing stock-days. We only insert these zero records between the first and the last appearance of a stock in the sample period, to avoid the error of using periods before a new listing or after a de-listing. Also, we do not insert zeros for days when there is no trading volume in the Datastream data. We remove ADRs from our sample. We merge this short volume data with Datastream data and trim the variables at p1 and p99 values. The daily sample of OTC stocks comprises of 2,585,734stock-days and 7,119 unique stocks. We compute scaled short volume i,d as shares sold short i.d divided by total trading volume for this sample, where subscript i and d stands for stock and day, respectively. We download the price, number of shares, price high, price low, return index, market capitalization, trading volume, and industry sector variables from Datastream International. We calculate the following variables from this dataset. Return i,d-1,d-250 is calculated as (return index i,d- 1 return index i,d-251 )/( return index i,d-251 ). Return i,d-1,d-5 is calculated as (return index i,d-1 return index i,d-6 )/( return index i,d-6 ). Return i,d is calculated as (return index i,d return index i,d-1 )/( return index i,d-1 ). Return i,d-d+1 is calculated as (return index i,d+1 (return index i,d )/( return index i,d ). Return i,d-d+5 is calculated as (return index i,d+5 return index i,d )/( return index i,d ). Return i,d-d+20 is calculated as (return index i,d+20 return index i,d )/( return index i,d ). We calculate volatility i,d as (price high i,d -price low i,d )/price low i,d. Volatility i,d-1-d-5 is calculated as the average of last 5 days volatility. Volatility i,d-1-d-250 is calculated as the average of last 250 days volatility. Turnover i,d is calculated as trading volume for the day divided by shares outstanding. Turnover i,d-1-d-5 is average of last 5 days turnover. In Table 1, Panels A and B we provide summary statistics of our sample. In Panel A, we find that the average number of shares sold short and scaled short volume, for non zero trading 11

volume day, in our sample is 141,177 and 23.68 percent, respectively. In columns 3 to 10, we present the values for 5 th percentile, 10 th percentile, 25 th percentile, median, 75 th percentile, 90 th percentile, 95 th percentile, and the standard deviation of these variables. The mean daily return for the stocks in our sample is 0.27 percent. The mean daily volatility is 7.71 percent, and the mean market value is 1.656 million dollars, while the median of market value is 11 million dollars. In Panel B, we provide the short selling numbers separately for different industry classifications using Datastream. We find that the Insurance industry has highest scaled short volume. [Insert Table 1 here] Next, we perform correlation analysis between short selling activity, past returns, future returns, and other control variables and report the results in Table 2. We find that scaled short volume is positively related with contemporaneous return (Return i,d ) and lagged 5-day return (Return i,d-1-d-5 ). These findings are consistent with Diether, Lee, and Werner (2009) and Jain, Jain, and McInish (2012), who find higher short selling for stocks with higher past returns. 1-day future return (Return i,d-d+1 ), 5-day future return (Return i,d-d+5 ), and 1-month future return (Return i,dd+20) are all negatively correlated with short volume indicating informed short selling in OTC stocks consistent with Diether, Lee, and Werner (2009). Short volume is positively related with contemporaneous volatility (Volatility i,d ) and lagged 5-day volatility (Volatility i,d-1-d-5 ). These findings are consistent with Diether, Lee, and Werner (2009), who find higher short selling for higher return volatility. Short volume is positively related with market value indicating higher short volume for larger firms. Short volume is also positively related with turnover indicating higher short volume on days of higher overall trading volume. 12

[Insert Table 2 here] 4. Results and discussion 4.1 Short selling activity and future returns In Table 3, we report return predictability of short sellers using short volume data. In Panel A, we report the equal weighted future returns of portfolios with zero short volume and portfolios with non-zero short volume in columns 2 and 3, respectively. The average daily number of OTC stocks with zero short selling is 723 and the average daily number of OTC stocks with non-zero short selling is 615. These numbers indicate that short selling in OTC stocks is not as constrained as found by past literature (Ang, Shtauber, Tetlock (2013)). Next, we compute the future return of a portfolio that is long in stocks with zero short volume and short in stocks with non-zero short volume, and we report the return of this long-short strategy in column 4. We report the corresponding t-statistics of our long-short strategy in column 5. We find that our long-short strategy has return of 0.51 percent, 1.64 percent, and 3.60 percent over the next 1- day, 5-day, and 1-month period, respectively. These returns are statistically significant at 0.01 percent level. These results indicate that short selling is OTC market is informed. In Table 3, Panel B, we focus only on stock-days with non-zero short volume. We create quintiles of stocks by short volume and compute the equal weighted future returns of portfolio composed of stocks in each quintile. We report these quintile portfolio returns in columns 2 to 6. Average scaled short volume in the lowest quintile portfolio is 7.09 percent and the average scaled short volume in the highest quintile portfolio is 99.12 percent. Next, we compute the future return of a portfolio that is long in stocks with lowest short volume and short in stocks with highest short volume, and we report the return of this long-short strategy in column 7. We report the corresponding t-statistics of our long-short strategy in column 8. We find that our 13

long-short strategy has return of 0.62 percent, 0.81 percent, and -0.65 percent over the next 1-day, 5-day, and 1-month period, respectively. These returns are statistically significant at 0.01 percent level. The negative returns over the next 1-month period indicate the short selling in OTC stocks may be abusive and manipulating the stock prices in downward direction. [Insert Table 3 here] In Table 4, we report the value weighted return similar to the equal weighted returns presented in Table 3. In Panel A, we find positive returns for portfolio that is long in stocks with zero short volume and short in stocks with non-zero short volume. In Panel B, we find positive returns for portfolio that is long in stocks with lowest short volume and short in stocks with highest short volume. Using the value-weighted method, we do not see any reversal in the returns. [Insert Table 4 here] To better understand the return predictability in OTC stocks, we also perform a double sort analysis for future return, where we create terciles of market value and then within each tercile we create two groups of zero short volume and non-zero short volume. We report these results in Table 5, Panel A. We compute the equal weighted future return of a portfolio that is long in stocks with zero short volume and short in stocks with non-zero short volume. We report the return of this long-short strategy for low market value firms and the corresponding t-statistics in columns 2 and 3, respectively. We report the similar values for medium market value firms (high market value firms) in columns 4 and 5 (in columns 6 and 7). We find that our long-short strategy has Return i,d-d+1, Return i,d-d+5, and Return i,d-d+20 of 0.63 percent, 2.68 percent, and 6.18 percent, respectively, for low market value firms. These returns are statistically significant at 0.01 percent level. We find similar results for medium market value and high market value firms, 14

though the magnitude of returns is highest for low market value firms. This again suggests informed short selling in OTC market. In Table 5 Panel B, we look at only the stocks with non-zero short volume. We create quintiles of short volume for each tercile of market value. We compute the equal weighted future return of a portfolio that is long in stocks with low short volume and short in stocks with high short volume. We report the return of this long-short strategy for all three terciles of market value firms and the corresponding t-statistics. We find that our long-short strategy has positive 1- day and 1-week future returns (Return i,d-d+1 and Return i,d-d+5 ) for all three terciles of market value similar to that in Panel A. However, we find that 1-month return (Return i,d-d+20 ) is negative and significant for low and medium market value terciles and it is insignificant for the high market value terciles. These results indicate that short selling, specifically in smaller and medium size OTC stocks may be abusive and manipulating the stock prices in downward direction. [Insert Table 5 here] In Table 6, we report the value weighted return similar to the equal weighted returns presented in Table 5. In Panel A, we find that future value weighted returns for portfolio that is long in stocks with zero short volume and short in stocks with non-zero short volume are positive and significant for all three terciles of market value. In Panel B, we find that 1-day and 1-week future returns (Return i,d-d+1 and Return i,d-d+5 ) for portfolio that is long in stocks with low short volume and short in stocks with high short volume are positive and significant for all three terciles of market value. However, we find that 1-month return (Return i,d-d+20 ) is negative and significant for low and medium market value terciles. Here again, we find that extreme level of short selling, specifically in smaller and medium size OTC stocks, might be manipulating the stock prices in downward direction. 15

[Insert Table 6 here] Overall, looking at the results in Table 3 to Table 6, we see informed short selling in OTC stocks based on zero and non-zero short selling. However, we see some return reversals when we further examine short selling further by creating quintiles of non-zero short selling. When we use double sort methodology, we find that this return reversal exists only in smaller and medium size firms but not in larger firms. These results indicate that short sellers might be taking advantage of the opaque environment of smaller OTC firms rather than trading primarily based on information. 4.2 Determinants of short volume Next, we perform a regression analysis to find determinants of short volume and report the results in Table 7. We report the results of following regression equation in Model 1: h, = +, +, +, +, +, +,, (1) where α0 α5 are parameters to be estimated and ε is a random error term. Subscripts i and d denote firm and day, respectively. In Model 2, we add the industry fixed effects and in regression equation 1. We find that the coefficients on contemporaneous return (Return i,d ) and lagged 5-day return (Return i,d-1-d-5 ) are positive and significant. These results are consistent with the findings of Diether, Lee, and, Werner (2009). Thus, short sellers act as contrarian traders in stocks with positive returns. The coefficients on contemporaneous volatility (Volatility i,d ) and lagged 5-day volatility (Volatility i,d-1-d-5 ) are positive, suggesting that stocks with higher past volatility seem attractive to short sellers consistent with the findings of Diether, Lee, and Werner 16

(2009). The coefficient of market value i,d is positive, suggesting that short sellers are more active in larger OTC stocks. [Insert Table 7 here] 4.3 Are Short Sellers Informed Traders or Opportunistic Risk-Bearers? Our results in Tables 4 through 6 provide some evidence that short sellers are able to predict future returns in OTC stocks. Following Diether, Lee, and Werner (2009), we test if they are able to do so because they are informed or because they provide liquidity in periods of heightened uncertainty. The heightened uncertainty may be due to informed trading or due to differences in opinion. Once the increased uncertainty subdues, the prices return to their normal level. We report the results of regressing future returns on short selling and past returns in Table 8. We only use the firm-days with non-zero short selling. To allow for non-linear effects of short selling and past returns, we first sort all stocks on day t into scaled short volume quintiles, and define a dummy variable high (low) to be one for all stock in the highest (lowest) quintile of scaled short volume. Similarly, we sort all stocks on day t into recent return r 1, 5 quintiles, and define a dummy variable winner (loser) to be one for all stocks in the highest (lowest) quintile of past returns. The results in Table 8 show that losers outperform winners and the magnitude is 0.26 percent per day as per model 1 (1.07 percent for 5-day period as per model 3 and 0.33 percent per month as per model 5). High scaled short volume is a significant predictor of negative future 1-day return and 5-day return in models 1 and 3. Specifically, stocks in the highest quintile of short-selling activity experience significant negative future returns by about 0.44 percent for 1-day period and 0.56 percent for 5-day period, respectively. By contrast, the lowest quintile of short-selling activity predicts positive future returns of 0.15 percent for 1-day 17

period and 0.25 percent for a 5-day period, respectively. The difference in predicted future returns for the high minus the low quintiles is highly significant. However, similar to our univariate results, we find that highest quintile of short selling does not result in negative future return over next 1-month period. In fact, the difference between high and low is positive and significant in model 5. Finally, we add controls for the opportunistic risk-bearing hypothesis in model 2, 4, and 6 of Table 8. We again find that higher short selling predicts negative returns only for 1-day and 5-day period, and is insignificant in predicting returns for 1-month period. We include r 5, 1 to allow for weekly return reversals. There is clear evidence of weekly reversals. Scaled short volume remains a significant predictor of future 1-day and 5-day negative returns even after controlling for weekly patterns in returns over next one week. We find that the coefficient of volatility is negative and significant in models 4 and 6, implying that higher volatility predicts future negative returns. Even after controlling for the volatility as suggested by the alternative hypothesis, scaled short volume still remains a significant predictor of future negative returns for 1-day and 5-day periods, but not for a 1-month period. High turnover in the previous week also predicts negative future returns for 1-day and 5-day periods. These results are consistent with our findings from Tables 5 and 6, short selling predicts negative returns only for 1-day and 5-day period, but fails to predict return for 1-month period. [Insert Table 8 here] 5. Conclusion We examine short selling activity in OTC stocks using short volume data for a period from 2009 to 2016. We find that on average short selling constitutes about 24 percent of daily 18

trading volume in OTC stocks on non-zero trading volume days. Short sellers are most active in Insurance industry. Similar to exchange traded stocks, a portfolio with long position in stocks with zero short volume and short position in stocks with non-zero short volume has significant positive returns. This strategy with value-weighted portfolio results in a profit of 0.29 percent in 1-day period, 0.52 percent in 5-day period, and 0.54 percent in 1-month period. The value-weighted returns of long-short strategy, using long in firms with zero short volume and short in firms with non-zero short volume, is also positive and significant for all three terciles of market value when we use a double sort strategy. However, we see return reversal when we look at the different magnitudes of non-zero short selling. The value-weighted returns for 1 day and 5 day holding periods are still positive, but one-month returns are negative, for low and medium market value terciles, when we use a long short strategy based on long in firms with lowest quintile of non-zero short volume and short in firms with highest quintile of non-zero short volume. These results indicate that short sellers might be taking advantage of the opaque environment of smaller OTC firms rather than trading primarily based on information. We also find that OTC stocks have higher short volume after a period of positive returns and on days of positive returns. Short volume is higher for stocks with higher past volatility and higher contemporaneous volatility. Short volume is also higher for stocks with higher market value. Following Diether, Lee, and Werner (2009), we test an alternate opportunistic riskbearing hypothesis and find that short sellers increase their activity to provide liquidity in times of heightened uncertainty. Volatility predicts negative future returns as predicted by the opportunistic risk-bearing hypothesis. Nonetheless, scaled short volume still remains a 19

significant predictor for future returns for 1-day and 5-day periods even after controlling for volatility and turnover, but fails to predict return for a 1-month period. 20

References Aggarwal, R. K., and G. Wu, 2006, Stock Market Manipulation: Theory and Evidence, Journal of Business 79, 1915 1953. Ang, A., Assaf A. Shtauber, and Paul C. Tetlock, 2013. Asset pricing in the dark: The cross section of OTC stocks, Review of Financial Studies (12): 2985-3028. Blau, B. M., and C. Wade. 2012. Informed or speculative: short selling analyst recommendations. Journal of Banking & Finance 36: 14 25. Boehmer, E., C. M., Jones, and X. Zhang. 2008. Which shorts are informed? Journal of Finance 63: 491-527. Böhme, R., and T. Holz, 2006, The Effect of Stock Spam on Financial Markets, working paper, Technische Universität Dresden. Bris, A., W.N. Goetzmann, and N. Zhu, 2007. Efficiency and the bear: Short sales and markets around the world, Journal of Finance 62, 1029 1079. Brunnermeier, M. K., and L. H. Pedersen. 2005. Predatory trading. Journal of Finance 60: 1825 1863. Bushee, B. J., and C. Leuz, 2005, Economic Consequences of SEC Disclosure Regulation: Evidence from the OTC Bulletin Board, Journal of Accounting and Economics 39, 233 264. Diamond, D., and R. Verrecchia. 1987. Constraints on Short-Selling and Asset Price Adjustment to Private Information. Journal of Financial Economics 18:277 311. Diether, K. B., K. Lee, and I.M. Werner, 2009. Short-sale strategies and return predictability, Review of Financial Studies 22, 575-607. Engelberg, J. E., A.V. Reed, and M.C. Ringgenberg, 2012. How are shorts informed? Journal of Financial Economics105(2), 260-278. Eraker, B., and M. J. Ready, 2015, Do Investors Overpay for Stocks with Lottery-Like Payoffs? An Examination of the Returns on OTC Stocks, Journal of Financial Economics 115, 486-504. Frieder, L. and J. Zittrain, 2007, Spam Works: Evidence from Stock Touts and Corresponding Market Activity, working paper, Purdue University. 21

Harris, J. H., Panchapagesan, V., Werner, I. M., 2009. Off but not gone: A study of nasdaq delistings, Working paper.. Henry, T. R., and J. L. Koski. 2010. Short selling around seasoned equity offerings. Review of Financial Studies 23: 4389 4418. Jain, C., P. K. Jain, and T. H. McInish. 2012. Short selling: The impact of SEC rule 201 of 2010. Financial Review 47: 37 64. Luft, C., L. M. Levine, and S. Larson, 2001, Over the Counter Bulletin Board Exchange: Market Structure, Risk, and Return, Journal of Alternative Investments Fall, 33 42. Luft, C., and L. M. Levine, 2004, Over the Counter Bulletin Board Exchange: The Impact of Liquidity and Size to Return, Volatility, and Bid/Ask Spread, Journal of Alternative Investments Winter, 95 106. Macey, J. R., M. O Hara, and D. Pompilio, 2008, Down and Out in the Stock Market: The Law and Economics of the Delisting Process, Journal of Law and Economics 51, 683 713. Miller, E. 1977. Risk, Uncertainty, and Divergence of Opinion. Journal of Finance 32:1151 68. Nelson, K., Price, R., and B. Rountree, 2013, Are individual investors influenced by the optimism and credibility of stock spam recommendations? Journal of Business Finance and Accounting 40, 1155 1183. Nofsinger, J., Varma, A., 2014. Pound wise and penny foolish? OTC stock investor demographics and portfolios, Review of Behavioral Finance 6 (1), 2-25. Shkilko, A., B. Van Ness, and R. Van Ness, 2012. Short selling and intraday price pressures, Financial Management 41, 345-370. White, J. T., 2016. Outcomes of investing in OTC stocks, U.S. Securities & Exchange Commission DERA white paper. 22

Appendix: Variable definitions and data source Variable Definitions: Shares sold short i,d = total number of shares sold short for firm i on day d; Scaled short volume i,d = short volume i,d divided by total volume i,d ; Total short interest i,ft = number of outstanding shorted shares for firm i on fortnight ft; Scaled short interest i,d = total short interest i,ft divided by shares outstanding; Return i,d-1,d-250 = (return index i,d-1 return index i,d-251 )/ return index i,d-251 ; Return i,d-1,d-5 = (return index i,d-1 return index i,d-6 )/ return index i,d-6 ; Return i,d = (return index i,d return index i,d-1 )/return index i,d-1 ; Return i,d-d+1 = (return index i,d+1 (return index i,d )/ return index i,d ; Return i,d-d+5 = (return index i,d+5 return index i,d )/ return index i,d ; Return i,d-d+20 = (return index i,d+20 return index i,d )/ return index i,d ; Volatility i,d-1-d-250 = calculated as the average of last 250 days volatility; Volatility i,d-1-d-5 = average of last 5 days volatility; Volatility i,d = (price high i,d -price low i,d )/price low i,d ; and Market value i,d = number of shares outstanding times the stock price. Data source: We use two datasets for short selling activity. First, we obtain the daily short volume data for OTC stock from the FINRA website for the period from August 2009 to January 2016. This dataset also includes total trading volume for the stocks. Second, we download the short interest data from the shortsqueeze.com for the period from August 2009 to January 2016. This dataset also includes shares outstanding. We only keep the observations with listing exchange as Bulletin Board (BB) or National Bulletin Board (NBB). We delete ADRs from both datasets. We also download daily closing prices, high price, low price, return index, and market capitalization from Datastream International. 23

Table 1 Short volume in OTC stocks In this table, we provide descriptive statistics of short volume and the control variables used in our paper. In Panel A, we report the descriptive statistics of our overall sample. In Panel B, we report the descriptive statistics of our sample by industry classification. Please see the Appendix for variable definitions and sample coverage. Panel A Mean P5 P10 P25 Median P75 P90 P95 Standard deviation Shares sold short i,d 141,177 0.2 0.4 3 69 8,318 101,245 506,345 725,739 Scaled short volume i,d 23.68% 0.01% 0.01% 0.04% 1.24% 40.33% 90.83% 100.00% 33.72% Return i,d 0.27% -18.21% -10.34% -2.48% -0.02% 1.82% 9.87% 19.88% 12.35% Volatility i,d 7.71% 0.00% 0.00% 0.00% 1.80% 11.11% 24.75% 34.22% 11.72% Market value (in millions) i,d 1,656 0.1 0.3 2 11 103 2,488 8,584 7,009 Turnover 0.93% 0.00% 0.00% 0.00% 0.02% 0.11% 0.58% 2.00% 5.89% Avg. number of stocks d 1,571 Panel B Shares sold short Scaled short volume Avg. number of stocks Industrial 128,371 23.94% 1,116 Utility 137,322 24.55% 49 Transportation 59,224 25.90% 14 Banking 2,847 23.44% 68 Insurance 3,096 27.35% 12 Other Financial 40,591 21.25% 46-250,770 22.61% 286 24

Table 2 Correlation matrix In this table, we provide correlation matrix of short selling variables and the control variables used in our paper. Please see the Appendix for variable definitions and sample coverage. Variable Scaled short volume i,d Return i,d Return i,d-1-d-5 Return i,d-d+1 Return i,d-d+5 Return i,d-d+20 Volatility i,d Volatility i,d-1-d-5 Market value i,d Return i,d 0.0716*** Return i,d-1-d-5 0.0217*** -0.1618*** Return i,d-d+1-0.0332*** -0.1596*** -0.0132*** Return i,d-d+5-0.0396*** -0.1715*** -0.0283*** 0.3185*** Return i,d-d+20-0.0400*** -0.1200*** -0.0226*** 0.1751*** 0.4004*** Volatility i,d 0.0325*** 0.0425*** 0.0334*** -0.0086*** -0.0235*** -0.0676*** Volatility i,d-1-d-5 0.0338*** -0.0010 0.0305*** -0.0058*** -0.0286*** -0.0997*** 0.5858*** Market value i,d 0.0455*** -0.0033*** -0.0028*** -0.0045*** -0.0052*** 0.0123*** -0.1312*** -0.1917*** Turnover i,d 0.0594*** 0.0057*** 0.0021*** -0.0088*** -0.0157*** -0.0331*** 0.1515*** 0.1575*** -0.0363*** ***Significant at 0.01 level, **Significant at 0.05 level, *Significant at 0.10 level 25

Table 3 Equal weighted returns of portfolios based on short volume In Panel A of this table, we provide the equal weighted holding period returns of portfolio with zero short volume in columns 2 and equal weighted holding period return of portfolio with non-zero short volume in columns 3. In column 4, we report the equal weighted holding period returns of a portfolio that is long in stocks with zero short volume and short in stocks with nonzero short volume. We report the corresponding t-statistics of our long-short strategy in column 5. Our holding periods are 1 day, one week, and one month. We also provide the average number of daily stock for each portfolio. In Panel B of this table, we only look at the stocks that have short volume greater than zero. We create quintiles of short volume and report the equal weighted holding period returns of these quintiles in columns 2 to 6. In column 7, we report the equal weighted holding period returns of a portfolio that is long in stocks with low short volume and short in stocks with high short volume. We report the corresponding t-statistics of our longshort strategy in column 8. Our holding periods are 1 day, one week, and one month. We also provide the average scaled short volume for each portfolio. Please see the Appendix for variable definitions and sample coverage. Panel A: Future returns for zero vs non-zero short volume Variable Zero short volume Non-zero short volume Avg. number of stocks d 723 615 (Zero short volume Non-zero short volume) t-stat Return i,d-d+1 0.39% -0.12% 0.51%*** (38.95) Return i,d-d+5 1.16% -0.48% 1.64%*** (59.54) Return i,d-d+20 0.58% -3.02% 3.60%*** (77.19) Panel B: Future returns for low vs high short volume Lowest quintile Highest quintile (Low - High) Variable 0 1 2 3 4 0-4 Scaled short volume i,d 7.09% 23.36% 42.07% 70.40% 99.12% t-stat Return i,d-d+1 0.10% -0.01% -0.10% -0.07% -0.52% 0.62%*** (22.82) Return i,d-d+5-0.14% -0.52% -0.58% -0.22% -0.95% 0.81%*** (14.66) Return i,d-d+20-2.64% -3.85% -3.98% -2.60% -1.99% -0.65%*** (-6.63) ***Significant at 0.01 level, **Significant at 0.05 level, *Significant at 0.10 level 26

Table 4 Value weighted returns of portfolios based on short volume In Panel A of this table, we provide the value weighted holding period returns of portfolio with zero short volume in columns 2 and value weighted holding period return of portfolio with non-zero short volume in columns 3. In column 4, we report the value weighted holding period returns of a portfolio that is long in stocks with zero short volume and short in stocks with nonzero short volume. We report the corresponding t-statistics of our long-short strategy in column 5. Our holding periods are 1 day, one week, and one month. We also provide the average number of daily stock for each portfolio. In Panel B of this table, we only look at the stocks that have short volume greater than zero. We create quintiles of short volume and report the value weighted holding period returns of these quintiles in columns 2 to 6. In column 7, we report the value weighted holding period returns of a portfolio that is long in stocks with low short volume and short in stocks with high short volume. We report the corresponding t-statistics of our longshort strategy in column 8. Our holding periods are 1 day, one week, and one month. We also provide the average scaled short volume for each portfolio. Please see the Appendix for variable definitions and sample coverage. Panel A: Future returns for zero vs non-zero short volume Variable Zero short volume Non-zero short volume Avg. number of stocks d 703 601 (Zero short volume Non-zero short volume) t-stat Return i,d-d+1 0.14% -0.15% 0.29%*** (19.79) Return i,d-d+5 0.31% -0.21% 0.52%*** (18.02) Return i,d-d+20 0.64% 0.10% 0.54%*** (11.33) Panel B: Future returns for low vs high short volume Lowest quintile Highest quintile (Low - High) Variable 0 1 2 3 4 0-4 Scaled short volume i,d 6.80% 23.20% 42.55% 72.78% 99.44% t-stat Return i,d-d+1 0.03% 0.02% -0.08% -0.15% -0.32% 0.35%*** (12.2) Return i,d-d+5 0.07% 0.03% -0.13% -0.12% -0.54% 0.61%*** (10.3) Return i,d-d+20 0.25% 0.08% -0.15% 0.13% -0.12% 0.37%*** (3.54) ***Significant at 0.01 level, **Significant at 0.05 level, *Significant at 0.10 level 27

Table 5 Equal weighted returns of the double sort portfolios based on market value and short volume In this table, we provide the equal weighted holding period returns of double sort methodology. In Panel A, we report the equal weighted holding period returns of a portfolio that is long in stocks with zero short volume and short in stocks with non-zero short volume for three tercile groups based on market value in columns 2, 4, and 6. We report the corresponding t- statistics of our long-short strategy in columns 3, 5, and 7. Our holding periods are 1 day, one week, and one month. In Panel B of this table, we only look at the stocks that have short volume greater than zero. We report the equal weighted holding period returns of a portfolio that is long in stocks with low short volume and short in stocks with high short volume for three tercile groups based on market value in columns 2, 4, and 6. We report the corresponding t-statistics of our long-short strategy in columns 3, 5, and 7. Please see the Appendix for variable definitions and sample coverage. Panel A: Future returns for zero vs non-zero short volume by market value Market value (Low) Medium Market value (High) (Zero short volume nonzero short (Zero short volume nonzero short (Zero short volume nonzero short Variable volume) t-stat volume) t-stat volume) t-stat Return i,d-d+1 0.63%*** (18.38) 0.58%*** (26.44) 0.31%*** (25.58) Return i,d-d+5 2.68%*** (35.69) 1.43%*** (34.08) 0.79%*** (31.34) Return i,d-d+20 6.18%*** (53.78) 3.22%*** (44.72) 2.00%*** (42.4) Panel B: Future returns for low vs high short volume by market value Market value (Low) Medium Market value (High) Variable (Low High) t-stat (Low High) t-stat (Low High) t-stat Return i,d-d+1 0.79%*** (9.62) 0.69%*** (14.35) 0.37%*** (14.07) Return i,d-d+5 0.92%*** (5.27) 0.73%*** (8.34) 0.61%*** (10.87) Return i,d-d+20-1.36%*** (-5.17) -0.34%** (-2.35) -0.14% (-1.38) ***Significant at 0.01 level, **Significant at 0.05 level, *Significant at 0.10 level 28

Table 6 Value weighted returns of the double sort portfolios based on market value and short volume In this table, we provide the value weighted holding period returns of double sort methodology. In Panel A, we report the value weighted holding period returns of a portfolio that is long in stocks with zero short volume and short in stocks with non-zero short volume for three tercile groups based on market value in columns 2, 4, and 6. We report the corresponding t- statistics of our long-short strategy in columns 3, 5, and 7. Our holding periods are 1 day, one week, and one month. In Panel B of this table, we only look at the stocks that have short volume greater than zero. We report the value weighted holding period returns of a portfolio that is long in stocks with low short volume and short in stocks with high short volume for three tercile groups based on market value in columns 2, 4, and 6. We report the corresponding t-statistics of our long-short strategy in columns 3, 5, and 7. Please see the Appendix for variable definitions and sample coverage. Panel A: Future returns for zero vs non-zero short volume by market value Market value (Low) Medium Market value (High) (Zero short volume nonzero short (Zero short volume nonzero short (Zero short volume nonzero short Variable volume) t-stat volume) t-stat volume) t-stat Return i,d-d+1 0.76%*** (17.94) 0.50%*** (22.69) 0.29%*** (19.67) Return i,d-d+5 2.55%*** (30.49) 1.26%*** (29.75) 0.51%*** (17.85) Return i,d-d+20 5.27%*** (40.85) 2.96%*** (38.96) 0.53%*** (11.06) Panel B: Future returns for low vs high short volume by market value Market value (Low) Medium Market value (High) Variable (Low High) t-stat (Low High) t-stat (Low High) t-stat Return i,d-d+1 0.87%*** (8.74) 0.64%*** (13.65) 0.36%*** (12.29) Return i,d-d+5 0.87%*** (4.5) 0.67%*** (7.78) 0.67%*** (11.5) Return i,d-d+20-0.81%*** (-2.63) -0.43%*** (-2.93) 0.48%*** (4.73) ***Significant at 0.01 level, **Significant at 0.05 level, *Significant at 0.10 level 29