Short selling during the time of an earnings announcement

(1)

Short Selling during the time of an Earnings Announcement

Amsterdam Business School

Name Pernille Deijlen

Student number 10747354

Program Economics & Business Specialization Finance & Organization Number of ECTS 12

Supervisor Ilko Naaborg Target completion 31/1/2018

Abstract

This paper examines short selling activity prior to earnings announcements in the Nasdaq market from 2012 until 2016. The tests show that there is a negative effect. In the five days prior to an announcement and on the day itself, the amount of short selling is lower than normal. This is in line with existing literature. Other results are that higher stock liquidity increases short sales and unexpected are the findings that both stock return and stock return volatility have a negative relationship with short interest.

(2)

2 Statement of Originality

This document is written by Student Pernille Deijlen who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

Table of Contents... 3

Introduction ... 4

Literature review ... 4

Hypothesis, methodology and data ... 6

Hypothesis and methodology ... 6

Data and descriptive statistics ... 7

Empirical results ... 10

Conclusion and discussion ... 12

References ... 13

(4)

4

Introduction

Short sales are very popular in the technology sector. Investors are short selling stock from big firms like Apple, Google and Facebook. There are multiple research papers about what kind of variables affect short sales but not a lot specifically about the technology sector. Investors use short sales for multiple reasons, for example to make a profit in situations where they expect the stock prices to fall. This could arise when firms announce their earnings. It is an interesting question to see whether the amount of short sales is higher than normal around the time of an earnings announcement. Christophe, Ferri and Angel (2004) did this kind of research, they studied whether there is a difference in amount of short selling in days preceding earnings announcements and in days where there are no

announcements. They found for most firms there are less short sales in days before an earnings announcement. They used data of short sales transactions of 913 Nasdaq firms in the fall of 2002. It would be interesting to see what the result would be if you used a larger recent time frame and specified it to the technology sector. How do earnings announcements of Nasdaq listed technology companies affect the amount of their stock shorted by investors in the period 2012-2016?

To answer this question I will look at data of all these technology companies and perform a regression. In this paper I will first discuss relevant literature on the topic, then I will describe the methodology and the data, then I will present the results and at last the conclusion.

Literature review

There are a lot of research papers on short selling. An important article for this research paper is the article from Angel, Christophe and Ferri (2003). They look at variables that could affect short sales in the Nasdaq market. There are a few important things they find which are used in this paper. Angel et al (2003) find that, ‘(1) overall, 1 of every 42 trades involves a short sale; (2) short selling is more common among stocks with high returns than stocks with weaker performance; (3) actively traded stocks experience more short sales than stocks of limited trading volume; (4) short selling varies directly with share price volatility; (5) short selling does not appear to be systematically different on various days of the week; and (6) days of high short selling precede days of unusually low returns’ (p. 66).

Angel et al (2003) have a dataset with daily data from September 13th_{2002 until December}

12th_{2002, so three months in total. This leads to a dataset of 89152 observations. They use quintiles to}

show their results for the relationship between short sales and stock returns and they find that stocks with high returns are shorted more often. They rank the stocks on how actively they are traded to see whether this influences short sales. They find that the liquidity of a stock positively affects the amount

(5)

5 of short interest. For the effect of share price volatility they also use quintiles. They find that short sales are higher when return volatility is higher. I use these findings for my own research.

Christophe et al (2004) did research on short selling prior to earnings announcements. They look at 913 Nasdaq firms in the fall of 2000 and check if short selling in the days before an earnings announcement differs from short selling without an announcement. They look at data from September 13th_{2003 until December 12}th_{2003 and have a sample of 913 Nasdaq traded stocks. They find that for}

most firms short selling is lower in the days before an announcement than normal. For a few firms, short selling is higher during the time of an announcement, this is probably caused by informed traders. What they have researched is similar to my research. A big difference is that I have data for multiple years instead of three months.

Woolridge and Dickinson (1994) look at short selling and common stock prices. A lot of people say that informed traders make money at cost of non-informed traders. They refute this by finding that short sellers do not earn abnormal returns. Short sellers do not drive prices down, in fact they make the market more liquid. They also find there is a positive relation between short selling changes and stock return. Woolridge and Dickinson (1994) use monthly data on short sales and stock return from the NYSE, Amex and Nasdaq from 1981-1991. Then they analyse this by using a random sample of 100 companies.

Other research done in this field is the research from Kim and Verrecchia (1994). They look at the effect of earnings announcements on market liquidity and trading volume. Because of informed traders there is more information asymmetry during the time of an announcements which leads to a decrease of market liquidity. On the other hand informed traders may cause for an increase in trading volume during the time of an announcement.

To summarize, Angel et al (2003) find that high returns, high liquidity and high return volatility positively influence the amount of short sales. Christophe et al (2004) find that for most firms short interest in the days prior to an earnings announcement is lower than in a

non-announcement period. Research by Woolridge and Dickinson (1994) shows that there is a positive relationship between short selling and stock return, short sellers do not earn abnormal returns and they do not drive stock prices down, instead they increase market liquidity. And at last there is the research of Kim and Verrecchia (1994), they state that during the time of an earnings announcement there is information asymmetry between traders which causes less liquidity but an increase of trading volume.

Using the result of research done by Christophe et al (2004), I expect the answer to the research question, how do earnings announcements of Nasdaq listed technology companies affect the amount of their stock shorted by investors in the period 2012-2016, to be that there is a negative effect. So I expect that the amount of short sales will be lower during the time of an earnings announcement than on other days.

(6)

6

Hypothesis, methodology and data

Hypothesis and methodology

To answer the research question, how do earnings announcements of Nasdaq listed technology companies affect the amount of their stock shorted by investors in the period 2012-2016, I need data on the dependent variable, the short interest, and on the variables that influence this dependent variable. So I need data for multiple companies and multiple years which is a case of panel data.

Angel et al (2003) have looked into short selling in stocks listed on the Nasdaq. They found that there are more short sales if a stock has a high return, a high liquidity and a high volatility. These are the control variables. The regression below is mostly based on this research paper. The variable earnings announcement is added to see if and how this variable influences the amount of short interest. 𝑙𝑜𝑔(𝑆𝐼𝑖𝑡) = 𝛽0+ 𝛽1𝑅𝐸𝑇𝑖𝑡+ 𝛽2𝑙𝑜𝑔⁡(𝑇𝑉𝑂𝐿𝑖𝑡) + 𝛽3𝑙𝑜𝑔⁡(𝑇𝑉𝐴𝐿𝑖𝑡) + 𝛽4𝑉𝑂𝐿𝑖𝑡+ 𝛽5𝐴𝑁𝑁𝑖𝑡+ 𝜀𝑖𝑡

The dependant variable is the logarithm of the amount of short sales⁡𝑙𝑜𝑔⁡(𝑆𝐼𝑖𝑡). The first

control variable is the stock return (𝑅𝐸𝑇𝑖𝑡). According to Angel et al (2003) and Woolridge and

Dickinson (1994) a higher stock return leads to a higher short interest so I expect the coefficient to be positive. The second and third control variables are measurements of stock liquidity. This means how easy it is to trade the stock. The first one is trading volume 𝑙𝑜𝑔(𝑇𝑉𝑂𝐿𝑖𝑡) and the second is trading

value 𝑙𝑜𝑔(𝑇𝑉𝐴𝐿𝑖𝑡). Short sales of a stock will be higher when the liquidity of the stock is high (Angel

et al, 2003). Also according to d’Avolio (2002) stocks with low prices are difficult to short. So a positive coefficient is expected for both variables. Because the variables short interest, trading volume and trading value are very large values compared to the stock returns and the stock return volatilities I use the logarithm of these values.

The next control variable is the stock return volatility (𝑉𝑂𝐿𝑖𝑡). Angel et al (2003) say that

stocks with higher volatility also have higher short sales, so this would mean that the coefficient should be positive. The last variable is the earnings announcement (𝐴𝑁𝑁𝑖𝑡). This is the variable of

interest and it is a dummy. The dummy is equal to one if there is an earnings announcement on that day or in the five days prior to the announcement and it is equal to zero otherwise. I choose to use a total of six days just like Christophe et al. (2004). They chose to use six days for two main reasons. The first one is that it is unlikely that short selling occurs only one or two days before the

announcement. Informed traders have an incentive to engage in multiple smaller trades instead of one large trade. The second reason is equity lending market loans. Short sellers need the equity lending market to borrow the stock they want to sell short. Equity lending market loans usually last from one to multiple days.

I need to regress the daily amount of short sales of a stock on its returns, liquidity, volatility and on whether there is an earnings announcement of the company on that day or in the upcoming five

(7)

7 days. I can use 𝛽5 to answer the research question. As I mentioned before I expect there to be a

negative effect so I expect β5 to be lower than 0. This leads to the following testable

hypothesis:⁡H0:⁡β5≥ 0⁡ H1:⁡β5< 0. If it is significantly equal to zero, earnings announcements have

no effect on how many stocks are sold short. If it is larger than zero there is a positive effect and if 𝛽5

is lower than zero there is a negative effect.

Data and descriptive statistics

To run the regression I need data of all Nasdaq listed technology companies for the years 2012-2016. There are 434 Nasdaq listed technology companies at this moment. I need information on short interest, stock prices, trading volume, and earnings announcements for each firm. Unfortunately data on short interest was only available for 85 of the 434 companies. So for these 85 companies I tried to find data on stock prices, trading volume and earnings announcements. For some combinations of companies and dates I could not find all the data I needed so these were omitted from the dataset. Sometimes the trading volume was equal to zero, which means that day was a no trading day, so these were also omitted. If there was no data on the end of the year earnings announcement for a company I omitted all data of that company from the previous end of the year earnings announcement until the next announcement.

By calculating the interquartile range I could find the major outliers in my dataset. After taking a good look at the outliers I decided to delete all of them. I used the following two formulas.

𝑜𝑢𝑡𝑒𝑟⁡𝑓𝑒𝑛𝑐𝑒 = 𝑄1− 3(𝑄3− 𝑄1)

𝑜𝑢𝑡𝑒𝑟⁡𝑓𝑒𝑛𝑐𝑒 = 𝑄3+ 3(𝑄3− 𝑄1)

So after looking thoroughly at the data there are 82 companies and 8194 observations left to work with.

Below I will explain a bit about every variable and give descriptive statistics. Because I have a large dataset I make use of quantiles. These quantiles are based on the dependent variable, the amount of short interest. Quantile 1 has the lowest short interest and quantile 4 the highest. Table 1 shows what the quantiles look like. Because 8194 observations cannot exactly be divided by four, quantile 1 and 4 contain 2049 observations and quantile 2 and 3 contain 2048 observations. For the regression I use the logarithm of short interest, trading volume and trading value but for the descriptive statistics I will use the real non-logarithmic values because this gives a better insight in what the data looks like. The dependent variable is the short interest. The information on the amount of stock shorted for each firm is available through Compustat. The data is available daily, twice a month, in the middle

(8)

8 and at the end. So the rest of the data is based on these dates. Table 1 below presents descriptive statistics about the short interest.

Table 1. Daily Short Interest by Quantiles based on the Short Interest in 2012-2016.

Quantile 1

(lowest short interest)

Quantile 2 Quantile 3 Quantile 4 (highest short interest) Mean 167619.4 1123515 3545185 23174801.9 Standard deviation 145749.1 490574.6 955941.3 27956160.1 Median 135253 1063469 3467262 12043984 Min 322 451223 2064012 5436989 Max 450901 2056005 5426398 157656160

Note: quantile 1 and 4 have 2049 observations and quantile 2 and 3 have 2048 observations.

The control variables in the regression are the stock returns, trading volume, trading value and stock return volatility. How stock returns are measured is based on the paper of Angel et al. (2003). They measure stock returns as the percentage price change of the closing price of today and the day before.

𝑠𝑡𝑜𝑐𝑘⁡𝑟𝑒𝑡𝑢𝑟𝑛⁡𝑎𝑠⁡𝑎⁡𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 =𝑁𝑝− 𝑂𝑝 𝑂𝑝

× 100%

Where 𝑁𝑝 is the new price, so the closing price of today, and 𝑂𝑝 is the old price, so the closing

price of yesterday.

Stock return volatility is computed by measuring the standard deviation of stock returns of 7 days after and 7 days before ‘today’. Information on closing stock prices for each firm comes from CRSP. If the closing price is not available CRSP gives the bid/ask average so than this is used instead. The two tables below show information on the daily stock returns and the daily stock return

volatilities.

Table 2. Daily Stock Return in Percentages by Quantiles based on the Short Interest in 2012-2016.

Quantile 1

Quantile 2 Quantile 3 Quantile 4 (highest short interest) Mean -0.1078 -0.0342 0.0940 0.1160 Standard deviation 2.5454 2.3052 2.1166 1.9013 Median -0.0811 0 0 0 Min -8.3333 -8.2467 -7.5061 -8.1448 Max 8.3333 8.3261 8.2540 8.2445

Table 3. Daily Stock Return Volatility in Percentages by Quantiles based on the Short Interest in 2012-2016.

Quantile 1

Quantile 2 Quantile 3 Quantile 4 (highest short interest) Mean 2.8341 2.3791 2.1904 1.8920 Standard deviation 1.4486 1.2873 1.2131 1.0303 Median 2.5132 2.0976 1.8638 1.6070 Min 0 0.3912 0.3211 0.4421 Max 7.9200 7.9463 7.9517 7.8455

(9)

9 Table 2 shows the mean stock return gets higher for each quantile. Which is expected because the literature told us that higher returns cause higher short interest. The standard deviation goes down with each quantile and the minimum and maximum stay about the same. Table 3 shows that the mean return volatility goes down for each quantile. Which is odd, because I expected the return volatility to be higher for higher short interest. The standard deviation also goes down for each quantile and the minimum and maximum remain approximately the same.

The next two control variables are trading volume and trading value. Data on trading volume comes from CRSP and is equal to the amount of stock sold on that day. The trading value is an interaction variable, it is equal to the trading volume times the stock price. As I mentioned before closing stock prices come from CRSP. The three tables below present descriptive statistics on trading volume, trading value and stock prices.

Table 4. Daily Trading Volume by Quantiles based on the Short Interest in 2012-2016.

Quantile 1

Quantile 2 Quantile 3 Quantile 4 (highest short interest) Mean 81351.64 356656.1 1245283 8220545.31 Standard deviation 147619.3 580659.6 1331848 13780483.8 Median 39107 197542.5 689261 2695218 Min 16 3454 41459 71137 Max 1973908 9954280 14280610 156257503

Table 5. Daily Trading Value in US dollars by Quantiles based on the Short Interest in 2012-2016.

Quantile 1

Quantile 2 Quantile 3 Quantile 4 (highest short interest) Mean 1166072 10938945.02 77168764.92 575594090.20 Standard deviation 3336104 26280753.49 148334784.40 1954382974 Median 225719.40 2902918.82 19388751.65 94355842.56 Min 56.25 3128.29 121185.81 523568.32 Max 65164240 459647324.20 1750843320 27835352528

Table 6. Daily Stock Closing Prices in US dollars by Quantiles based on the Short Interest in 2012-2016.

Quantile 1 (lowest short interest)

Quantile 2 Quantile 3 Quantile 4 (highest short interest) Mean 11.3256 22.9915 43.1606 52.5702 Standard deviation 14.2160 21.2345 38.7195 87.2679 Median 6.5600 14.6350 31.9400 28.4200 Min 0.3400 0.6465 0.8500 1.7200 Max 97.4900 217.9200 249.7000 691.2800

According to the literature higher liquidity would cause a higher short interest. Table 4 and 5 show that the mean trading volume and mean trading value indeed goes up with each quantile. The standard deviation, median, minimum and maximum all go up for both the trading volume as the trading value. Table 6 shows that for each quantile the mean stock price goes up, as well as the

(10)

10 standard deviation. It is remarkable that the median of quantile 4 is a bit lower than the median of quantile 3, you would expect it to be higher, but the high standard deviation of quantile 4 could

explain this. The minimum and maximum get higher for each quantile but for quantile 2 and 3 they are almost the same.

At last the variable of interest, the dummy variable earnings announcements. Data on the announcements comes from Compustat. There is no data on end of the year earnings announcements, so I use the 4th_{calendar year quarter announcement instead. Table 7 presents descriptive statistics}

about this dummy variable.

Table 7. Daily Earnings Announcements by Quantiles based on the Short Interest in 2012-2016.

Quantile 1

Quantile 2 Quantile 3 Quantile 4 (highest short interest) Mean 0.0107 0.0132 0.0127 0.0122 Standard deviation 0.1031 0.1141 0.1120 0.1098 Median 0 0 0 0 Min 0 0 0 0 Max 1 1 1 1

In table 7 the mean, minimum and maximum do not say much. The minimum and maximum are always [0,1] because it is a dummy variable. I expected the mean to be lower for each quantile but this is not entirely true. From quantile 2 it does go down, but it is actually lowest for quantile 1. The standard deviation behaves the same as the mean.

Empirical results

First I will do a Hausman test to see if I need to use the random effects model or the fixed effects model. Under the null hypothesis you prefer the random effects model and the alternative is the fixed effects model. The Hausman test is presented in table 1 in the Appendix. The table shows the null hypothesis is rejected so the fixed effects model is the best model for this panel dataset.

Table 8 below presents the output of the regression.

(11)

11 Table 8 shows that for the regression for the whole period the coefficient for the stock return is negative and not significant. This is not in line with the theory because I expected it to have a positive effect on short sales. A change in return will cause a 0.04% decrease in short sales. The coefficient of the logarithm of trading volume is positive and significant at the 5% level. I expected this because the more liquid a stock is the higher the short sales. Short sales will increase with 0.3258% with every increase in trading volume. The effect of trading value is positive but not significant. As I said before this is expected because liquidity should increase short interest. Short interest increases with 0.0097% with every increase in trading value. The coefficient for the stock return volatility is negative and not significant. I expected the coefficient to be positive so this is surprising. A change in volatility will cause a -1.28% decrease in short sales. At last the dummy variable earnings announcement. The coefficient is negative and is significant at the 10% level. I expected it to be negative and this is indeed the case. If there is an announcement in the five days to come or on the day itself, short sales will decrease with 5.13%.

The F-test tests if all coefficients together differ from zero. This is indeed the case, the p-value is equal to or lower than 0.0000 and is significant at the 1% level. So the model means something. The R-squared is a goodness-of-fit measure and it measures how much of the variance of the dependent variable short sales is explained by the variance of the independent variables. The R-squared is equal to 0.7582 so this means that 75.82% of the variance of the dependent variable is explained by the model.

Table 8 shows the coefficient of the variable announcement is negative for every period, so this is in line with theory, but it only is significant for the 2012-2016 period. Furthermore it is

noticeable that for almost every period a different coefficient is significant. This shows that the model

Table 8. Fixed Effects

2012-2016 2012 2013 2014 2015 2016 Intercept 4.3006*** (0.3024) 5.2161*** (0.3134) 5.3076*** (0.3035) 5.5976*** (0.5001) 6.2487*** (0.2294) 5.8025*** (0.3281) RET -0.0004 (0.0019) 0.0057** (0.0023) 0.0006 (0.0028) 0.0030 (0.0023) 0.0023 (0.0020) -0.0010 (0.0026) logTVOL 0.3258** (0.1246) -0.2218 (0.2058) 0.1207 (0.1659) 0.4216** (0.2064) 0.2148 (0.2096) 0.6669*** (0.1715) logTVAL 0.0097 (0.1198) 0.3286 (0.2060) 0.0093 (0.1651) -0.2525 (0.2234) -0.1922 (0.1927) -0.4883*** (0.1462) VOL -0.0128 (0.0080) -0.0030 (0.0103) 0.0037 (0.0128) -0.0329*** (0.0112) 0.0146** (0.0062) -0.0091 (0.0118) ANN -0.0513* (0.0259) -0.0187 (0.0251) -0.0704 (0.0513) -0.0694 (0.0656) -0.0542 (0.0434) -0.0210 (0.0328) Observations 8194 1484 1575 1627 1710 1798 Number of firms 82 68 69 72 76 82 Overall-R2 _0.7582 _0.4253 _0.7243 _0.3247 _0.1183 _0.0003 F-test 12.93 (0.0000)*** 4.51 (0.0013)*** 4.02 (0.0029)*** 5.68 (0.0002)*** 1.39 (0.2364) 4.64 (0.0009)***

Note: robust standard errors between parentheses below the coefficients. Note: p-value between parentheses below the F-test values.

(12)

12 is not very strong at predicting the amount of short sales. The F-test for the separate years is

significant at the 1% level for every year except for 2015. This means that the joint effect of all variables does not differ from zero, so for 2015 it is not a very good model. The R-squared, how much of the variance is explained, is very low for the year 2016, it is only 3% which is quite low compared to the other R-squared values.

Conclusion and discussion

In this paper I have researched whether earnings announcements of technology companies have an effect on the amount of short sales in the period 2012-2016. Data consisted of 82 companies and 8194 observations. I used a regression to answer the research question and I found a negative relationship between earnings announcements and short interest. In five days prior to the announcement and on the day itself, short sales are lower than normal. This result is significant at the 10% level. This is

consistent with the results of Christophe et al (2004).

Unexpected is that the coefficients for the stock return and stock return volatility are negative. According to Angel et al (2003) they both should have had a positive effect on short sales and

according to Woolridge and Dickinson (1994) stock return should indeed have had a positive effect. In the research from Angel et al (2003) they also found that liquidity has a positive effect. This is

something I also found, trading volume and trading value both have a positive coefficient. So this means a stock is sold short more often when the stock is easily tradeable.

A limitation of this research is that data on short interest is only available for 85 of the 434 technology companies and this data was only available twice a month, not daily. Also data on earnings announcements was only available quarterly, so the 4th_{quarter earnings announcement had to be used}

as the end of the year earnings announcement. So the research could be better if you have access to more data on short interest and the end of the year earnings announcements. A suggestion for new research could be to look into what effect the crisis has had. Whether the amount of short sales differ before, during and after. Or a suggestion could be doing the research elsewhere, this paper looks into Nasdaq listed companies, you could take a look at a developing market.

(13)

13

References

Angel, J. J., Christophe, S. E. & Ferri, M. G. (2003). A Close Look at Short Selling on Nasdaq. Financial Analysts Journal, 59(6), pp. 66-74.

d’Avolio, G. (2002). The market for borrowing stock. Journal of Financial Economics, 66(2), pp. 271-306.

Christophe, S. E., Ferri, M. G. & Angel, J. J. (2004). Short-Selling Prior to Earnings Announcements. Journal of Finance, 59(4), pp. 1845-1875.

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

Desai, H., Ramesh, K., Thiagarajan, S. R. & Balachandran, B. V. (2002). An Investigation of the Informational Role of Short Interest in the Nasdaq Market. Journal of Finance, 57(5), pp. 2263-2287. Kim, O. & Verrecchia, R. E. (1994). Market liquidity and volume around earnings announcements. Journal of Accounting and Economics, 17(1-2), pp. 41-67.

Saffi, P. A. C. & Sigurdsson, K. (2011). Price Efficiency and Short Selling. Review of Financial Studies, 24(3), pp. 821-852.

Woolridge, J. R. & Dickinson, A. (1994). Short Selling and Common Stock Prices. Financial Analysts Journal, 50(1), pp. 20-28.

(14)

14

Appendix

Table 1. Hausman test

Fixed effects (b) Random effects (B) Difference (b-B) srt(diag(V_b-V_B))S.E. RET -0.0004059 -0.0005857 0.0001798 . logTVOL 0.3258381 0.3224132 0.003425 0.0033412 logTVAL 0.0097166 0.0479579 -0.0382413 0.0044026 VOL -0.0128467 -0.0172483 0.0044016 . ANN -0.0513058 -0.0524055 0.0010997 . Chi2(5) 539.56 (0.0000)***

Note: p-value between parentheses below the Chi-test value.