How short selling might contribute to price discovery

(1)

How short selling might contribute to price

discovery

Abstract: Following a research by Boehmer and Wu (2013), I look at how the presence of

short selling in markets might contribute to the price discovery process. I assume efficient markets and the idea that short sellers might be informed, sophisticated traders (Christophe et al., 2004) and test whether short selling contributes to price discovery around negative earnings announcements. This negative announcement should be in line with expectations when short interest is higher for this stock (Christophe et al., 2004) and prices might change before the announcement is made (Boehmer & Wu, 2013). In case of a positive announcement, no price discovery is expected as a result of short selling since short sellers have negative expectations and high short interest in combination with a positive announcement would need a high price change once the positive announcement is made (Hong et al., 2012). Findings here are that only for a 11 day window, 5 days leading up to the event and 5 days after, the influence of short interest on abnormal returns for positive announcements is higher than for negative announcements. So only here we could conclude there might be a relation between short selling and price discovery, but no strong evidence is found.

Key words: short selling, efficient markets, price discovery, earnings announcements

Student name: Sarah van Kempen

University: University of Amsterdam

Student id: 5624703

Date: 21 February 2014

Bachelor: Finance and Organization 12 EC

Specialization: Finance

(2)

2 Introduction

There is much to do around short selling; while some researchers believe that short selling could be a good thing and helps identify temporarily overpriced stock and have them return to their true underlying value, short sellers are also seen, at least by regulators, as the bad guys causing financial distress (Diether, Lee & Werner, 2009). The short sell bans of 2008 show how short selling seems to worry regulators around the world, with researchers questioning whether these bans have had the desired effect (Bris, Goetzmann & Zhu, 2007). Goal of the 2008 ban on short selling was to stop prices of all financial stock from falling, because extreme short selling would be the main force behind this drop in prices (Battalio, Mehran & Schultz, 2012). However, there is research that claims that short selling contributes to more efficient prices (Boehmer & Wu, 2013; Bris et al., 2007) with stock prices adjusting faster to new public information when short selling is allowed (Aitken, Frino, McCorry & Swan, 1998; Boehmer & Wu, 2013). Also, Boehmer and Wu (2013) show no support for the idea that high levels of short interest push prices down beyond proportion.

Short selling is the main subject of this thesis. However, as a lot of recent research focuses on how short sellers make decisions on which stock to sell short and how restricting short selling influences markets, I will be looking at what high levels of short selling mean for market returns in case short selling is allowed. Idea here is that increases in short selling might precede negative news or returns (Christophe, Ferry & Angel, 2004; Diether et al., 2009), if allowing short selling contributes to price discovery (Boehmer & Wu, 2013) in efficient markets, then –based on research by Boehmer and Wu, 2013- I expect prices of stock with high short interest to adjust prior to negative news announcements resulting in low cumulative abnormal return, since the higher levels of short selling indicate negative market sentiment. However, when a positive announcement takes place while short interest has gone up leading up to this event, short selling does not contribute to price discovery and high returns will follow, because here the announcement is not what was expected by the market (Hong, Kubik and Fishman, 2012). There will be a large increase in prices because short sellers will have to close their positions quickly to protect them against great losses, this is what Hong et al. (2012) find that happens after unexpected positive earnings announcements on high short interest stock. My research question is: ‘Does short selling contribute to price discovery in case of negative earnings announcements?’. I will be looking at a time period concerning three years which will be the period 2007-2009, and I will use NASDAQ and NYSE stock.

(3)

3

in case of the unexpected positive announcement, for the unexpected negative earnings announcement no significant effect on cumulative abnormal returns is found. Although there is a negative correlation between short interest and the size of the forecast error for the negative announcement group, suggesting high short interest to be correlated with lower forecast errors here but no statement can be made about the causal relationship because I can not say whether the high short interest causes the smaller forecast error or the other way around. Further, I found no evidence suggesting that short interest contributes to price discovery in case of negative announcements. However I did find, although only for the [-5, +5] window, that the regression coefficient of the short interest variable for unexpected positive earnings announcements is higher than the regression coefficient of the short interest variable for unexpected negative earnings announcements, so a higher return is needed when there is high short interest and positive earnings.

This thesis is build up as follows; first I will discuss existing literature on short selling, price discovery and efficient markets, then the possible relationship between short selling and earnings announcements and this will lead to my hypothesis; in the next section I will describe my research design, explain my dependent, independent and control variables and also the regressions I use to test my hypothesis; after the research design I describe the results which lead to the conclusion and discussion.

Literature review

When selling a stock short, you sell a stock you do not own. In order to do this you will have to borrow the stock from someone who does own it and eventually you will have to buy back the borrowed stock and return it to whom it was borrowed from (Dechow, Hutton, Meulbroek and Sloan, 2001). Investors who choose to make a short sale expect a stock price to decline, a short seller must buy back the shares and pay possible dividends paid by the firm to whoever he lend the shares from, therefore he only makes a profit when price decline is higher than the sum he needs to repay. (Brent, Morse & Stice, 1990). Danger in selling a stock short is that your potential loss can be very high if prices rise instead of decline (Shiller, 2003). A short sale can either be covered or naked. It is said to be covered when the investor borrows the stock he wants to sell short before the sale is made, when the investor does not borrow this stock while making the short sale this sale is said to be naked (Beber & Pagano, 2013). In this thesis, I look at short selling levels in general, no distinction will be made between naked or covered short sales it is about short interest which is the amount of stock sold short.

(4)

4

influences markets and stock prices (Diether et al. 2009; Boehmer & Wu,2013). Stock prices are determined by the expected future cash flows they will generate, discounted for the holding period (Sadka, 2007). There is a lot of research on stock pricing and the behavior of these stocks. Allowing short selling could contribute to price discovery, which is the process of how fast new information incorporates into stock prices (Boehmer & Wu, 2013). In their research on how short selling might contribute to efficiency and price discovery, Boehmer and Wu (2013) find that when short sales increase, stock prices will follow a random walk and there is less post earnings announcement drift (PEAD) after negative earnings news, which they claim shows that short selling contributes to price discovery and causes prices to adjust before the announcements. Post earnings announcement drift is used to describe the process that when earnings are higher than expected, cumulative abnormal returns in the following weeks will be higher as well, while with lower earnings than expected cumulative abnormal returns would be lower in the following weeks (Livnat & Mendenhall, 2006). The random walk that Boehmer and Wu (2013) mention is part of stock market efficiency, in an efficient market stock prices follow a random walk which means that the price movements are unpredictable and prices will change as soon as new information is made public (Bodie, Kane & Marcus, 2011). This idea of efficient markets is also known as Fama’s efficient market hypothesis, which he claims means that stock prices reflect all information available at that time and prices will adjust almost immediately after new information has been made public (Fama, 1991). In a research conducted in 2004 during a period of three months, Christophe et al., (2004) claim that if short interest goes up and the rest of the market would be aware of this increase in short interest, this could lead to smoother price movements because with this transparency prices can now incorporate news about this short interest faster and so efficiency would go up. As of 2007, companies listed on among others, the NASDAQ and NYSE, must report their short interest positions at least on a bi-monthly base which means on the 15th and last business day of each month (FINRA, 2007). Following the idea by Christophe et al., (2004) which I described above, this obligation to report short interest could aid the price discovery process, even more so since existing literature tends to describe the short selling investor as an informed and sophisticated trader who might be able to identify firms with poor earnings, negative news or fraudulent reports before the rest of the market can (Christophe et al., 2004; Desai, Krishnamurthy and Venkataraman, 2006) or claims that they are simply better at analysing publicly available information than other market participants (Engelberg, Reed and Ringgenberg, 2012).

(5)

5

& Wu, 2013), in which direction would this influence returns? Investors go short on stock of which they believe prices will go down, short sellers represent a group of investors who have pessimistic beliefs on the underlying stock and excluding them from the market would mean prices will only reflect optimistic investors’ beliefs leading to overvalued prices (Miller, 1977). Several studies show short selling goes up in the period leading up to negative or bad earnings announcements (Christophe et al., 2004; Diether et al., 2009). Earnings announcements in this thesis are quarterly earnings announcements, these announcements are planned events, forecasts are made about earnings and therefore earnings can be positive, negative or in line with expectations (Christophe et al., 2004). High short interest before an earnings announcement might indicate negative earnings surprises (Christophe et al., 2004). With efficient markets, short interest available and short sales possibly influencing price discovery, what would this mean for stock prices around announcements? Hong et al. (2012) suggest that stock prices of firms with high levels of short interest are more sensitive to positive earnings announcements than are stock prices of firms with low levels of short interest. Which in terms of price discovery seems likely, if investors are rational and informed then they would go short when they expect prices to go down, if an unexpected positive earnings announcement occurs then this is in conflict with what investors believed and they will need to close their positions to prevent high losses, what Hong et al. (2012) also mention as the forced covering of short positions by short investors. Furthermore, Hong et al. (2012) show that prices of stock with high short interest overshoot their value in the short run after unexpected positive news and that price sensitivity is higher for stocks with a high short interest. So there might be no price discovery if there is high short interest and a positive earnings announcement follows, however negative earnings surprises could be in line with investors’ expectations if there is high short interest since some research suggests high short interest precedes negative earnings (Christophe et al., 2004; Desai et al., 2006; Diether et al., 2009). Based on the literature described above, I form the following hypotheses:

(1) High short interest in combination with a positive earnings announcement will have high cumulative abnormal returns since this was not expected. Short sellers will quickly close their positions, pushing prices up.

(2) High short interest in combination with a negative earnings announcement will have lower cumulative abnormal returns since in a market where short selling is allowed,

(6)

6

this is expected to contribute to price discovery leading prices to adjust before the negative announcement is made.

Research Design

In order to answer my research question: ‘Does short selling contribute to price discovery in case of negative earnings announcements?’ I formed two hypotheses. Because I look at how short selling might influence abnormal returns I will first have to compute expected returns, for this I will use the market model:

( )

My first hypothesis is based on the research findings by Hong et al. (2012), high short interest in combination with a positive earnings announcement will have high cumulative abnormal returns since this positive news was not expected. Short sellers will quickly close their positions, pushing prices up (Hong et al., 2012). For this hypothesis, the event is a positive earnings surprise for a high short interest firm. For the second hypothesis, the event is a negative earnings announcement.

The estimation window will be one year before the event window, which includes 252

trading days. The estimation window ends the day before the event window starts so, for a 41-day event window for example, it looks like this

For most stock the first event date will take place in march 2007, so expected returns will be calculated based on the period march 2006 until march 2007.

Short interest is defined as:

(7)

7

Short interest (SI) data is gathered using the Compustat database, this database reports short sales for NASDAQ and NYSE individual companies twice a month. Merging this data with the shares outstanding information I download from CRSP helps me calculate the short interest I need for my SI variable. In contrast to what Hong et al. (2012) did, I will use the percentage values of the short interest variable and not create a dummy to measure whether short interest is high or low. I’m interested in the magnitude of this Beta on abnormal returns.

Earnings surprise, which can be positive or negative or of course in line with expectations, is defined as

following research by Hong et al. (2012) and Boehmer and Wu (2013). Forecasted earnings will be computed using I/B/E/S since most literature uses this as well (Hong et al., 2012; Boehmer & Wu, 2013). It is based on actual earnings per share and forecasted earnings per share. Using STATA to compute the earnings surprise I did not take the previous price into account and so the earnings surprise is in my case an absolute value, this is done by Cheong and Thomas (2011) as well, however they also made scales to divide magnitude of the surprises into, which I do not. The F_ERROR variable in my dataset is the magnitude of the difference between what was forecasted and the actual value, it is the forecast error. I divide my sample into two groups where the earnings surprise, which can only be positive or negative in my sample since I filter out all earnings announcements which are in line with expectations, is used as a proxy for news. I have two datasets, one for which there is a positive earnings surprise and one for which there is a negative earnings surprise.

I include the following regressions

(1) if (2) if

(8)

8

SI is the short interest variable, BM is book to market ratio and SIZE is a size dummy. BM and SIZE are control variables. BM is included because of a possible book to market effect, this effect shows that stocks of firms with high BM values might also have high abnormal returns (Bodie et al., 2011). SIZE is a control variable I added following research by Hong et al. (2012) and is included to measure the effect of size on cumulative abnormal returns and is a measure of market capitalization.

First I will look at whether there are abnormal returns. To test whether short interest (SI) is an indicator of negative news and contributes to price discovery, I will merge the two datasets I used for my two regressions. Then I will create a dummy variable for unexpected earnings announcements (UE). UE is equal to one when there is a positive earnings announcement and UE is equal to 0 in case of a negative earnings announcement. Next, following research by Hong et al. (2012), I create an interaction variable, to see whether the influence of short interest on cumulative abnormal returns is different in case of a negative earnings announcement opposed to a positive earnings announcement. Running the following regression will help me find whether the β for a positive announcement and short selling is different from the β negative announcement and short selling.

is the β when there is a positive unexpected earnings announcement and there is short

interest, is the β in case there is a negative unexpected earnings announcement. The reason I use a one sided hypothesis test is because based on the literature discussed earlier, I expect that the effect of short interest on cumulative abnormal return needs to be bigger in the case of high short interest and a positive unexpected earnings announcement because this is not what was expected by the market and therefore price will have to adjust more after the announcement then when earnings are in line with what was expected, which I expect to be the case when there is high short interest and a negative unexpected earnings announcement. Using STATA I will also perform a second test to test whether the coefficients on short interest for the two groups are different from each other, I choose the SUEST option.

(9)

9

As efficient markets assume new information will incorporate itself almost immediately into prices I will use several event windows. [-1, +1] and [-2, +2] are used but also larger event windows are used to measure abnormal returns, following Christophe et al. (2004) and Daske, Richardson and Tuna (2005) I include [-5, +5], Christophe et al. (2004) conclude that short selling takes on extremely high values in the five days leading up to the (negative) event date. Also I use [-10, +10] and [-20, +20], because quarterly earnings announcements are the events of interest [-20, +20] will be my largest window, this is one month before the announcement until one month after the announcement. Since I assume efficient markets, and in an efficient market prices reflect all available information and adjust fairly quickly to news (Fama, 1991) I also add a [0, +1] window, to see whether the ‘unexpected’ announcement really is that unexpected. I would expect high absolute returns on these two days in case of an unexpected positive announcement, since Hong et al. (2012) show prices of the stock with high short interest and a positive announcement are overshooting their true value after this unexpected positive announcement.

The stock used in my sample are stock that are listed on the NASDAQ or on the NYSE. I randomly select 100 companies (see appendix A for an overview of selected stock), for all of these companies I compute the quarterly earnings surprise in the period 2007-2009 and filter out the observations for which there is no surprise. The earnings announcement dates and short interest dates are linked to each other in STATA and I now create two groups separating the companies with negative earnings from the group with positive earnings. The cumulative abnormal returns can be calculated using Eventus 8.0, which is an event study tool in WRDS. Benchmarks used to compute abnormal returns are the S&P500. Using Eventus 8.0 I calculate the cumulative abnormal returns for all companies and event dates separately and also I calculate the mean cumulative returns and weighted cumulative abnormal returns for all events together.

Results

Before I run my regressions, I test how the independent regression variables are correlated using Pearson correlation at a 5% significance level. The results are as follows:

(10)

10

F_ERROR 1

SI -0.0084 1

BM -0.0038 -0.0645 1

SIZE 0.0079 -0.04188* 0.0969* 1

Table 1: pearson correlation unexpected positive announcement. *significant at 5% level

UE=0 F_ERROR SI BM SIZE

F_ERROR 1

SI -0.2242* 1

BM 0.0725 0.0221 1

SIZE 0.0711 -0.2896* 0.0416 1

Table 2: pearson correlation unexpected negative announcement. *significant at 5% level

In the unexpected positive announcement dataset there is a very weak but significant negative relation between SIZE and SI, which suggests that short interest is a little higher in case of a smaller firm. In the unexpected negative announcement SI and F_ERROR have a significant but weak negative correlation, meaning the two variables move in opposite directions. If the size of the error goes up than short interest goes down. In this case if the error goes up it means that the error becomes smaller, the F_ERROR has negative values here since it is the negative announcement group and -0.06 is smaller than -0.05. So with higher short interest the forecasted error around the negative announcement becomes smaller, however with correlation I can say nothing about causal relation so I do not know if SI influences F_ERROR or the other way around.

I run two regressions and look at cumulative abnormal returns around the earnings announcements. First I look at total mean cumulative returns and weighted average abnormal returns for the two samples as a whole.

Mean CAR Weighted CAAR p-value

[-20, +20] 0.0313 0.0220 < 0.0001 [-10, +10] 0.0312 0.0233 < 0.0001 [-5, +5] 0.0321 0.0257 < 0.0001 [-2, +2] 0.0237 0.0196 < 0.0001 [-1, +1] 0.0219 0.0181 < 0.0001 [0, +1] 0.0206 0.0170 < 0.0001

Table 3: mean cumulative abnormal returns and weighted unexpected positive announcement. All statistically significant at 1%, 5% and 10% level. Also for one-sided t-test where H1 > 0

Mean CAR Weighted CAAR p-value

[-20, +20] -0.0482 -0.0394 0.0002 [-10, +10] -0.0455 -0.0337 < 0.0001 [-5, +5] -0.0359 -0.0271 < 0.0001 [-2, +2] -0.0347 -0.0274 < 0.0001 [-1, +1] -0.0227 -0.0168 < 0.0001 [0, +1] -0.0238 -0.0170 < 0.0001

(11)

11

Table 4: mean cumulative abnormal returns and weighted average unexpected negative announcement. All statistically significant at 1%, 5% and 10% level. Also for one-sided t-test where H1 < 0

The mean returns around positive earnings announcements are all significantly different from zero and using a one-sided test they are all expected to be bigger than zero at the 1%, 5% and 10% significance level. This is plausible based on Hong et al. (2012), I expect greater movements in abnormal returns for positive unexpected announcements. For the negative announcement a one sided test, mean cumulative returns are significant and smaller than 0. In a market where short selling is allowed, based on research by Boehmer and Wu (2013) I would expect no abnormal returns around the negative earnings announcement since short sellers would choose to sell the stock short leading to lower prices before the announcement.

The two multiple regressions show different results for the different windows. First I look at the positive unexpected earnings announcements. Here the regression coefficient of SI shows a significant and positive effect of short interest on cumulative abnormal returns for all the windows. The positive coefficient on SI shows that if short interest goes up, the cumulative abnormal returns go up as well, for the [-20, +20] window for example this means that if short interest goes up by 1%, this leads to a change in cumulative abnormal returns of + 0.468%. Below are the regression results of the [-20, +20] and [0, +1] windows, the other tables are included in appendix B. CAR [-20, +20] SI 0.468* (0.1718) BM -0.0001 (0.0007) SIZE -0.0001 (0.00001) Constant 0.0158 (0.0147) 0.0224 Adjusted 0.0170 No. observations 552

Table 5: regression unexpected positive announcement. Standard errors are given in parenthesis *Significant at 5% and 10% level.

(12)

12 CAR [0, +1] SI 0.217* (0.069) BM -0.000014 (0.0003) SIZE -0.000000008 (0.000000005) Constant 0.0137 (0.0059) 0.0364 Adjusted 0.0311 No. observations 552

Table 6: regression unexpected positive announcement. Standard errors are given in parenthesis *Significant at 1%, 5% and 10% level.

For the positive announcement group, higher short interest leads to higher returns but this higher return effect is not larger for the smaller windows. I expected the coefficient of the smaller windows and especially the [0, +1] window to be larger because at the event date (t=0) the positive announcement would make short sellers to close their positions preventing them from making greater losses (Hong et al. 2012). These results show no proof of that. The and adjusted show very low results, so the variables used in my regression do not explain a lot of the variation in the dependent variable (CAR). There may be other factors at work here that influence the abnormal returns. Only SI had a significant effect here, the SIZE and BM show no significant results.

The second regression concerns the group with the negative unexpected earnings announcement. None of the regression coefficients of the SI variable for the different windows show a coefficient with an effect significantly different from zero. Below are the results of two of the windows, the rest of the windows are included in appendix C.

CAR [-20, +20] SI 0.215 (0.348) BM -0.0016 (0.0034) SIZE 0.0000001 (0.000000015) Constant -0.061 (0.0301) 0.0034 Adjusted -0.008 No. observations 267

(13)

13 CAR [0, +1] SI 59.66 (108.49) BM -0.229 (1.06) SIZE -0.000004 (0.000047) Constant 2.4647 (9.411) 0.0016 Adjusted -0.0098 No. observations 267

Table 8: regression unexpected negative announcement. Standard errors are given in parenthesis

These results show no support for my second hypothesis. In this hypothesis I described that short selling might contribute to the price discovery process, which is based on research by Boehmer and Wu (2013), the presence of short selling would indicate that the stock is expected to go down. In an efficient market, when there is new information or knowledge of expectations, prices would respond to this new information and incorporate that into the price (Fama, 1991). Therefore I expect the abnormal returns to be absent or at least lower then when announcements are made that are different from what was expected, which I also based on research by Boehmer and Wu (2013) they also find that PEAD is no longer present here, as mentioned earlier. None of the control variables is statistically different from zero in the regression and the and adjusted are extremely low. In this case practically none of the variation in cumulative abnormal return is explained by the dependent variables included. Even though the SI variable turns out not to be significant in the negative announcement case, I still look at whether the size of this regression coefficient is different in the case of the separate announcement groups. After merging the datasets and creating the dummy variable for unexpected announcement and the interaction variable, as described in the research design, I run my third regression leading to the following results.

[-20, +20] [-10, +10] [-5, +5] [-2, +2] [-1, +1] [0, +1] UE*SI 0.037 (0.320) 0.334 (0.237) 0.0416* (0.179) 0.134 (0.137) 0.127 (0.126) -62.11 (72.59)

Table 9: test for . Standard errors are given in parenthesis

(14)

14

The only window for which the coefficient of the SI for positive unexpected earnings is significantly higher than the coefficient of the SI for negative unexpected earnings is [-5, +5]. This could mean that short selling concentrates itself in the five days leading up to the event, as Christopthe et al. (2004) find and that if this short selling turns out to be in conflict with actual earnings this would lead to a higher change in cumulative abnormal returns, as was suggested by Hong et al. (2012) for positive unexpected earnings. However, the coefficient is rather small so the effect might be there but the influence is still limited.

As I stated earlier, I also use a second test to compare the regression coefficient of short interest of the unexpected positive announcement and the unexpected negative announcements. Results are displayed below.

[-20, +20] [-10, +10] [-5, +5] [-2, +2] [-1, +1] [0, +1] p-value 0.63 0.4206 1.32 0.2504 3.25* 0.0714 0.37 0.5417 0.43 0.5130 0.96 0.3281

Table 10: test for .

*Significant at 10% level.

This test shows the same results as the first one, here the [-5, +5] window is the only window for which the coefficient is different for the two groups as well only in this case it has a lower statistical significance, since the result is only significant at the 10% level.

Conclusion & Discussion

The subject central in this thesis is how short selling influences markets. I focus on one specific element, which is how short selling might contribute to price discovery. Various literature I read -amongst others Boehmer and Wu (2013) and Christophe et al (2004)- before starting on my thesis, showed there might be a relation between negative earnings announcements, short selling and price discovery. My research question therefore is: ‘Does short selling contribute to price discovery in case of negative earnings announcements?’. To answer this question I use two hypotheses, which are: (1) High short interest in combination with a positive earnings announcement will have high cumulative abnormal returns since this was not expected. Short sellers will quickly close their positions, pushing prices up; (2) High short interest in combination with a negative earnings announcement will have lower cumulative abnormal returns since in a market where short selling is allowed, this is expected to contribute to price discovery leading prices to adjust before the negative announcement is made. These hypotheses are tested using three regressions, where the first two regressions are run to see the effect of short interest on cumulative returns and the last one is run to test

(15)

15

whether the coefficient of short interest for unexpected positive earnings announcements on cumulative abnormal returns is different from the coefficient of the short interest for unexpected negative earnings announcements on cumulative abnormal returns.

Results are that short interest has a small but significant positive effect on cumulative abnormal return in case of unexpected positive earnings announcements. This shows some evidence for the first hypothesis, the higher short interest the less a positive earnings announcement was expected and this might have led to higher cumulative abnormal returns since after the positive announcements short sellers quickly close their positions. No significant effect was found for short interest on cumulative abnormal returns in case of an unexpected negative earnings announcement. This shows no support for my second hypothesis and main idea that short selling contributes to price discovery, furthermore there are still significant negative cumulative abnormal returns around the negative earnings announcements. So I have found no evidence to believe prices adjust before the negative announcement due to higher short selling nor have I found evidence to conclude that short interest preceding negative earnings announcements leads to lower cumulative abnormal returns and helps the price discovery process. The last regression where I test whether the regression coefficients of short interest for unexpected positive earnings announcements are different from unexpected negative earnings announcements show that this is only the case for the [-5, +5] window, however this effect is very small and this is not enough evidence to completely accept the idea that the abnormal returns are higher around unexpected earnings announcements when short selling is allowed and present then the abnormal returns around unexpected negative earnings announcements when short selling is allowed and present.

I assume markets to be efficient, maybe they are not. Information on short sale volumes on stock might be available since the second half of 2007, but maybe not everyone uses this information or maybe the channel through which this information is made available is hard to find for investors. This problem of ‘new’ news incorporating into prices is also discussed in research, for example by Huberman and Regev (2001). They discuss how the channel through which information is made public could influence price reactions, a scientific magazine might have a smaller reach than a newspaper (Huberman and Regev, 2001). In this case, the information on short sell levels which have to be reported to FINRA might be difficult to find or people might not know it exists. Earlier research by Aitken et al. (1998) claims that the monthly short interest information in the American stock markets does not contribute to market transparency and therefore might not lead to price discovery. Even though short interest is has to be published bi-monthly now (FINRA, 2007) this might not be

(16)

16

frequent enough, for Aitken et al. (1998) argue that daily information on short sales would aid the price discovery process. Furthermore, the time period I used also included the short sell ban of 2008, with these restrictions markets might be less efficient as it slows price discovery as research by Beber and Pagano (2013) on the short sell ban of 2008 points out. However, this short sell ban was only active from September 18th 2008 until October 8th 2008 (SEC, 2008), in my sample of three year with 4 earnings announcements made per firm per year, I do not believe this three week ban to have an extremely high impact on my results. Maybe, having a period of financial crisis leads markets to act irrational and is that what influences the results of this study. Taking a different time period, one without financial crises or bubbles, might lead to different results. Also the control variables which are used show no significant results, maybe this is because the firms I used in my sample do not vary enough in terms of market capitalization or book-to-market ratio to show an effect. For further research more control variables could be included, I wanted to use more in this research as well. I thought about variables such as dividends announced and dividends paid, as Bali and Hide (1998) for example claim that prices change after dividends are paid, however I was not able to download that information from CRSP since it was not available. Furthermore volatility may influence prices, higher volatility in stock leads to higher risk and for people to accept this risk they might demand a higher return (Berk & Demarzo, 2011). The variable short interest in this research is based on bi-monthly data, I had no information available on daily short sell volumes, using daily numbers of short selling might give somewhat different results because they could paint a better picture of movements in prices and interest, which Boehmer and Wu (2013) also claim and they do use this daily short interest in their research. What also would be interesting to look at is why investors choose to go short in the first place, how they select their stock and to understand the mechanics behind this.

(17)

17 References

Aitken, M. J., Frino, A., McCorry, M. S., & Swan, P. L. (1998). Short sales are almost instantaneously bad news: Evidence from the Australian Stock Exchange. Journal of Finance, 53(6), pp. 2205-2223.

Bali, R., & Hite, G.L. (1998). Ex dividend stock price behaviour: discreteness or tax-induced clienteles? Journal of Financial Economics, 47(2), pp. 127-159.

Battalio, R., Mehran, H., & Schultz, P. (2012). Market Declines: What Is Accomplished by Banning Short-Selling? Federal Reserve Bank of New York: Current issues in economics and finance, 18(5), pp. 1-7

Beber, A., & Pagano, M. (2013). Short-Selling Bans Around the World: Evidence from the 2007–09 Crisis. Journal of Finance, 68(1), pp. 343-381

Berk, J., & DeMarzo (2011). Corporate Finance. Pearson, second edition.

Bodie, Z., Kane, A. and Marcus, A. (2011). Investments and portfolio theory. McGraco-Hill, 9th edition.

Boehmer, E., & Wu, J. (2013). Short selling and the price discovery process. Review of Financial Studies, 26(2), pp. 287-322

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. (2007). Efficiency and the Bear: Short sales and markets around the world. Journal of Finance, 62(3), pp. 1029-1079

(18)

18

Cheong, F. S., & Thomas, J. (2011). Why do EPS forecast error and dispersion not vary with scale? Implications for analyst and managerial behavior. Journal of Accounting Research, 49(2), pp. 359-401.

Christophe, S.E., Ferri, M.G., & Angel, J.J. (2004). Short-selling prior to earnings announcements. Journal of Finance, 59(4), pp. 1845-1875

Daske, H., Richardson, S.A., & Tuna, I. (2005). Do short sale transactions precede bad news events? Preliminary, the Wharton School, University of Pennsylvania.

Dechow, P., A. Hutton, L. Meulbroek, and R. Sloan (2001). Short-sellers, fundamental analysis and stock returns. Journal of Financial Economics, 61(1), pp. 77-106.

Desai, H., Krishnamurthy, S., & Venkataraman, K. (2006). Do short sellers target firms with poor earnings quality? Evidence from earnings restatements. Review of Accounting studies, 11(1), pp. 71-90.

Diether, K. B., Lee, K. H., & Werner, I. M. (2009). Short-sale strategies and return predictability. Review of Financial Studies, 22(2), 575-607.

Engelberg, J. E., Reed, A. V., & Ringgenberg, M. C. (2012). How are shorts informed?: Short sellers, news, and information processing. Journal of Financial Economics, 105(2), pp. 260-278.

Fama, E.F. (1991). Efficient capital markets: II. Journal of Finance, 46(5), pp. 1575-1617.

FINRA (2007). Short interest reporting.

http://www.finra.org/Industry/Compliance/RegulatoryFilings/ShortInterestReporting/

Hong, H., Kubik, J.D., & Fishman, T. (2012). Do arbitrageurs amplify economic shocks? Journal of Financial Economics, 103(3), pp. 454-470

(19)

19

Huberman, G., & Regev, T. (2001). Contagious speculation and a cure for cancer: a nonevent that made stock prices soar. The Journal of Finance, 56(1), pp. 387-396.

Livnat, J., & Mendenhall, R.R. (2006). Comparing the post-earnings announcement drift for surprises calculated from analyst and time series forecasts. Journal of Accounting Research, 44(1), pp. 177-205

Miller, E. M. (1977). Risk, uncertainty, and divergence of opinion. Journal of Finance, 32(4), pp. 1151-1168.

Sadka, G. (2007). Understanding stock price volatility: The role of earnings. Journal of Accounting research, 45(1), pp. 199-228.

Shiller, R. J. (2003). From efficient markets theory to behavioral Finance. Journal of Economic Perspectives, 17(1), pp. 83-104.

SEC (2008). SEC Halts Short Selling of Financial Stocks to Protect Investors and Markets. http://www.sec.gov/news/press/2008/2008-211.htm

(20)

20 Appendix A – Companies

NASDAQ FINANCIAL

Company Ticker Cusip

1. Zions Bankcorp ZION 98970110

2. American Financial Group AFG 02593210

3. Arch Capital Group ACGL G0450A10

4. Fifth third bancorp FITB 31677310

5. Iberiabank corp IBKC 45082810

6. World acceptance corp WRLD 98141910

7. Bank mutual corp BKMU 06375010

8. Associated Bank Corp ASBC 04548710

9. Cascade bancorp CACB 14715410

10. SEI investment comp SEIC 78411710

11. Capital city bank Group CCBG 13967410

12. Capitol Federal Fin. CFFN 14057C10

13. Center Financial corp CLFC 15146E10

14. Northern trust corp NTRS 66585910

15. Southwest bankcorp OKSB 84476710

16. Texas capital bancshares TCBI 88224Q10

17. Private bancorp inc PVTB 74296210

18. Signature Bank SBNY 82669G10

19. First Midwest Bancorp FMBI 32086710

20. First Merit Corp FMER 33791510

21. Huntington Bancshares Inc HBAN 44615010

22. SLM Corp SLM 78442P10

23. SVB Financial Group SIVB 78486Q10

24. E*Trade Financial Corp ETFC 26926410

25. West America Bancorp WABC 95709010

NASDAQ Non-financials Ticker Cusip

1. Amazon AMZN 02313510

2. Apple AAPL 03783310

3. Google GOOG 38259P50

4. Regeneron Pharm. REGN 75886F10

5. Adobe ADBE 00724F10 6. Microsoft MSFT 59491810 7. Yahoo YHOO 98433210 8. Ebay EBAY 27864210 9. RyanAir RYAAY 78351310 10. Intel INTC 45814010 11. Starbucks SBUX 85524410 12. Texas Instruments TXN 88250810 13. Costco COST 22160K10

14. Henry Schein Inc HSIC 80640710

15. Green Mountain Coffee Roasters GMCR 39312210

16. Automatic Data Processing ADP 05301510

17. Bed, Bath & Beyond BBBY 07589610

18. Mattel MAT 57708110

(21)

21

20. Vertex Pharmaceuticals VRTX 92532F10

21. Staples, Inc. SPLS 85503010

22. Amgen Inc. AMGN 03116210

23. Symantec SYMC 87150310

24. Western Digital WDC 95810210

25. Marriot International MAR 57190320

NYSE FINANCIALS

1. Citigroup C 17296710

2. Goldman Sachs GS 38141G10

3. Johnson & Johnson JNJ 47816010

4. Blackrock Inc BLK 09247X10

5. Moody’s Corp MCO 61536910

6. First American Bankcorp calif FAF 31852230

7. JPMorgan JPM 46625H10

8. Morgan Stanley Dean Witter MS 61744644

9. Metlife MET 59156R10

10. American Express AXP 02581610

11. American Equity AEL 02567620

12. Bank of America BAC 06050510

13. Istar Financial SFI 45031U10

14. Northstar Realty Finance NRF 66704R10

15. Allied Capital Corp ALD 01903Q10

16. Barclays PLC BCS 06738E20

17. Stancorp Financial Group SFG 85289110

18. Western Alliance Bankcorp WAL 95763810

19. Credicorp LTD BAP G2519Y10

20. BOSTON PROP BXP 10112110

21. Royal bank CA RY 7800871X

22. Avis business Group CAR 05377410

23. General growth Prop GGP 37002110

24. Lexington realty trust LXP 52904310

25. Amerigroup corp AGP 03073T10

26. Aspen Insurance holding AHL G0538410

NYSE Non-Financials

1. 3M MMM 88579Y10

2. McDonalds Corp MCD 58013510

3. IBM IBM 45920010

4. Wal-Mart WMT 93114210

5. Home Depot HD 43707610

6. Walt Disney DISN 25468710

7. Verizon VZ 92343V10

(22)

22

9. AT&T Inc T 00206R10

10. Caterpillar Inc. CAT 14912310

11. Cisco CSCO 17275R10

12. Dupont DD 26353410

13. Coca Cola KO 19121610

14. General Electric GE 36960410

15. Procter & Gamble PG 74271810

16. Exxon Mobile XOM 30231G10

17. Nike NIKE 65410610

18. Pfizer Inc PFE 71708110

19. Pepsi PEP 71344810

20. Coach COH 18975410

21. United techn. Corp. UTX 91301710

22. Sara Lee Corp SLE 80311110

23. Sony Corp. SNE 83569930

24. Nokia Corp. NOK 65490220

25. Oracle ORCL 68389X10

Appendix B – regression results different windows, unexpected positive announcement

SI 0.489* (0.1259) BM -0.0004 (0.0005) SIZE -0.0001 (0.0000009) Constant 0.011 (0.0108) 0.0352 Adjusted 0.0299 No. observations 552

Table 1: regression unexpected positive announcement. Standard errors are given in parenthesis *Significant at 1%, 5% and 10% level. CAR [-5, +5] SI 0.375* (0.1028) BM -0.00003 (0.0004) SIZE -0.000001 (0.0000007) Constant 0.0206 (0.0088) 0.0423 Adjusted 0.0370 No. observations 552

Table 2: regression unexpected positive announcement. Standard errors are given in parenthesis *Significant at 1%, 5% and 10% level.

(23)

23 CAR [-2, +2] SI 0.2017* (0.0781) BM -0.00009 (0.00032) SIZE -0.00000001 (0.000000005) Constant 0.0204 (0.0067) 0.0364 Adjusted 0.0311 No. observations 552

Table 3: regression unexpected positive announcement. Standard errors are given in parenthesis *Significant at 1%, 5% and 10% level. CAR [-1, +1] SI 0.156* (0.0725) BM -0.0000005 (0.0003) SIZE -0.00000001 (0.000000005) Constant 0.0205 (0.0062) 0.0288 Adjusted 0.0235 No. observations 552

Table 4: regression unexpected positive announcement. Standard errors are given in parenthesis *Significant at 5% and 10% level.

Appendix C – Regression results different windows, unexpected negative announcement

(24)

24 CAR [-5, +5] SI 0.0576 (0.1814) BM -0.00116 (0.0018) SIZE 0.0000005 (0.00000007) Constant -0.0392 (0.0157) 0.0031 Adjusted -0.0083 No. observations 267