The Relative Option to Stock Volume (OS) and Market Response to Earnings Surprises

Transcription

1 The Relative Option to Stock Volume (OS) and Market Response to Earnings Surprises Atul Rai * Barton School of Business Wichita State University 1845 Fairmount Street Wichita, KS (316) atul.rai@wichita.edu and Semih Tartaroglu Barton School of Business Wichita State University 1845 Fairmount Street Wichita, KS (316) semih.tartaroglu@wichita.edu May 2014 We gratefully acknowledge comments from Rodney Boehme, Philipp Illeditsch, Joshua Livnat, Bharat Sarath, Shyam Sunder, and Wayne Thomas. We alone are responsible for errors that remain. Atul Rai wishes to acknowledge generous financial support from Jones Faculty Fellowship. *Corresponding author. Fax (316)

2 The Relative Option to Stock Volume (OS) and Market Response to Earnings Surprises ABSTRACT The study examines the effect of option volume relative to stock volume (O/S) on market response to earnings surprises. The market reaction per unit of earnings surprise is lower for firms that have high O/S prior to earnings announcement than for firms with low O/S prior to earnings announcement. The difference is exacerbated for higher levels of pre-announcement returns. Results suggest informed trading by option traders stimulates pre-emption of the information content of earnings releases and makes earnings surprises less of a surprise. Overall, results are consistent with the view that options improves informational efficiency. Results are robust to several controls. JEL Classification: D82, G12, G13, G14, M41 Keywords: Price Discovery, Relative Option to Stock Volume (OS), Earnings Surprises, Earnings Response Coefficient, Information and Market Efficiency

3 The Relative Option to Stock Volume (OS) and Market Response to Earnings Surprises 1. Introduction Several studies in option literature suggest that option activity may stimulate informed trading and lead to faster discovery of stock prices. Security analysts literature indicates that analysts produce information about future cash flows and earnings which improves the information environment of a firm. Other studies suggest that for firms with richer information environment, the markets anticipates (i.e., preempts) a larger portion of the information contained in the forthcoming earnings announcements. Since a larger portion of the information content is already reflected in the current prices, these firms are likely to have a lower level of information content at the time of earnings releases. The purpose of this paper is to examine the extent to which options stimulate preemption of the information content of earnings announcements in addition to the preemption due to other informational factors including security analysts following. We develop and test the hypothesis that at the time of an earnings announcement, ceteris paribus, higher options trading relative to stock trading prior to the announcement will lead to lower initial market reaction per unit of earnings surprise. We refer to this negative relation between the level of relative option trading and information content of earnings as the option preemption hypothesis. In perfect markets, options are redundant securities as they can be replicated by a portfolio of risk free bonds and stocks (Black and Scholes 1973). But in imperfect markets with frictions (e.g., short sale constraints, asymmetric information, and transactions costs etc.) options play an important role since options complete the markets, and trading in the options market offers various advantages to investors such as leverage, circumvention of short sale restrictions and lowering of implicit borrowing costs. A vast literature, both theoretical and empirical, is 1

4 driven by this recognition and many studies suggest that options may have an important informational role in markets with frictions. Theoretical models show that informed traders may prefer to trade in options markets rather than in stock markets due to incremental advantages options markets provide (Back, 1993; Biais and Hillion, 1994). Black (1975) argues that some informed traders would not even trade at all if options markets did not exist (p 61-62). Chakravarty et al. (2004) show that almost twenty percent of price discovery takes place in options markets rather than stock market. Finally, Roll et al (2009, p ) suggest that options may encourage investors to engage in information activities incremental to the information production by security analysts and other sources. In general, investors anticipate, to a large extent, information contained in an earnings announcement. Foster (1977) finds that approximately seventy percent of stock price adjustment associated with quarterly earnings occurs prior to earnings announcements. The theoretical model of Kim and Verrecchia (1991, p 311) predicts that prices react less to earnings surprises in the presence of a greater amount of pre-announcement information. To the extant options stimulate informed trading and/or incremental information production by investors, the stock prices will reflect more information prior to earnings announcements and market response conditional on new news (i.e., earnings response coefficients or ERCs) will be lower. 1 Earlier tests of the effect of options on ERCs are reported in Skinner (1990), Ho (1993), and Mendenhall and Fehrs (1999, henceforth MF). Using a dichotomous variable that identifies firms that have options on their stocks listed on the option exchange, these studies examine 1 The earnings response coefficient (ERC) is the regression coefficient of earnings surprise in a regression of cumulative abnormal returns on earnings surprise, where earnings surprise is the difference between actual earnings and analysts forecasts. 2

5 whether ERCs are lower for stocks with listed options, in comparison to stocks without listed options. These studies provide mixed results about the impact of the listing of options on informational efficiency. Skinner (1990) and Ho (1993) find that ERCs are lower for stocks with listed options in comparison to stocks without listed options. In contrast, MF show that the effect of options listing on ERC is opposite to that documented in Skinner and in Ho. Specifically, MF find that the decrease in ERCs can be explained by contemporaneous changes in listed firms information environment like size. After controlling for these contemporaneous changes, they find that listing of options cause an increase rather than a decrease in ERCs. Their results cast doubt whether option trading improves informational efficiencies of stock prices and do not support the option preemption hypothesis. A possible explanation for mixed results in Skinner (1990), Ho (1993) and MF is that the dichotomous variable that identifies whether a firm has listed options or not is not sufficient to examine whether option trading preempts information content of earnings. We expect that improvements in information efficiencies of stock prices will vary with the level of option trading activity. This variation is not captured in a dichotomous variable like listed versus notlisted option firms. Informed traders are more likely to be active in options that have high liquidity (Admati and Pfleiderer, 1988; Pagano, 1989). Recently, Roll et al 2009 recognize that advantages that options markets provide, if any, will not be observed if the options does not have sufficiently high liquidity. The authors also argue that the examination of the effects of options on informational efficiency should take into consideration different levels of option activity. If investors have private information, they can trade in both the option market and the stock market. To the extent that option market offers advantages to investors, informed traders are more likely to trade in the option market than in the stock market, implying that the option 3

6 trading volume is likely to be higher than the stock trading volume. Chakravarty et al. (2004) show that price discovery in option market occurs when option volume relative to stock volume is high. The trading in options markets may be informed, or driven by noise trading and hedging demand which increases before public announcements. Roll et al. (2010) show that option volume relative to stock volume (henceforth O/S) reflects informed trading prior to earnings announcements. These findings are also supported by Johnson and So (2012) who find that O/S contains private information around earnings announcements and stock prices with high O/S, on average, is followed by lower abnormal returns which is consistent with the recognition that options enables investor to circumvent short sale constraints besides other advantages. In this paper, we make use of cumulative O/S prior to earnings announcement, to measure incremental information impounded to stock prices prior to earnings announcements. Thus we use O/S to measure cross-sectional differences in pre-announcement information reflected in stock prices prior to an earnings announcement. In section II, we discuss this issue further. To test the option preemption hypothesis, we use Fama and Macbeth (1993) quarter-byquarter regressions to study stock price response per unit of earnings surprises (ERCs) of high O/S firms and low O/S firms. Our main results show that ERCs of high O/S firms are significantly lower than ERCs for low O/S firms. We also find that this difference is exacerbated for higher levels of pre-announcement returns. These findings are consistent with option trading improving pre-announcement information and reducing the information content of earnings announcements. Our main results are robust to alternative explanations and various controls. For example, additional tests show that our findings are not driven by the firm s size; this finding is important as the size is an important indicator of the information available about a firm, and could provide an alternative explanation for lower ERCs. The negative information does not 4

7 dominate our results; this finding is important because option trading mitigates short-sale constraints in stock markets. Further, we control for factors that are shown to affect ERCs including earnings predictability, persistence, loss and firm characteristics such as risk and growth. We find that our results are robust to these controls. Important contributions of our study are as follows. First, our study supports the option preemption hypothesis by showing that a higher level of option activity decreases ERCs and indicates that option trading improves pre-announcement information of a firm. It contradicts long standing results of MF, and supports findings of Skinner (1990). Second, our study contributes to the options literature by documenting the positive impact of options trading in improving informational efficiency prior to important public announcements. We also extend findings of Roll et al (2010) and Johnson and So (2012) by examining the market reactions conditional on earnings surprises for different levels of O/S. The negative relationship between O/S and ERC provides further evidence that O/S reflects private information and informed trading in options markets decrease markets' dependence on security analysts' information production or noisy public announcements. Third, we extend the literature on ERC by documenting that the flow of information from options market to stock market dilutes the information content of earnings. The study also documents that the dilution is exacerbated for higher level of pre-announcement returns. Billings and Jennings (2011) develops an anticipated information content (AIC) of earnings news from option prices and volatility and offers an option-market approach to study the responsiveness of stock prices to earnings information. While our study does not provide a refined or better approach based on options, it supports the arguments in Billings and Jennings (2011) by showing that ERCs based on financial analysts' forecast errors are influenced from option activity. 5

8 The rest of the paper is organized as follows. Section 2 discusses the empirical and theoretical background and develops the main hypothesis of the study. Section 3 describes sample and variable definitions. Section 4 provides results. Section 5 provides additional tests for robustness. We conclude in Section Background Literature and Hypothesis Development 2.1. Option trading and stock price discovery Previous findings generally support the notion that option trading facilitates stock price discovery. Options are redundant in complete and frictionless markets, because they can be replicated by a portfolio of stocks and bonds; this replication is not possible if markets are incomplete and have frictions like transaction costs, asymmetric information and transaction costs (Black and Scholes 1973; Ross, 1976). Trading in options, in lieu of trading in stocks, stimulates price discovery since options offer attractive venues for traders to trade on their private information and beliefs. For example, investors with limited wealth can exploit leverage embedded in options for higher returns for each dollar they invest (Black, 1975; Cao, 1999). Relative to trading in the stock market, option markets often provide lower transaction costs (Amin and Lee, 1977; Black, 1975). Under certain conditions, options provide lower implicit borrowing rates (Black, 1975; Cox and Rubenstein, 1985). Short sale restrictions in equity markets also encourage traders to prefer options markets as they can take positions by writing call options or by buying put options (Easley et al., 1998; Figlewski and Webb, 1993). The theoretical model of Easley et al. (1998) shows that under certain conditions, informed traders prefer to trade in option markets rather than in stock markets. Grossman and Stiglitz (1980) argue that information gatherers must earn a return in excess of their search and processing costs. If option trading is perceived by investors to provide more profitable trading opportunities, then 6

9 more information is likely to be produced. The additional information production by option traders is further likely to facilitate stock price discovery. Consistent with theses argument, prior findings suggest that security prices reflect more private information in the presence of options trading (Biais and Hillion, 1994; Cao, 1999; Cao et al., 2005; Chakravarty et al., 2004; Easley et al., 1998). Jennings and Starks (1986) find that the prices of stocks with listed options adjust quicker when earnings are announced. Amin and Lee (1997) examine the open interest in options prior to earnings announcements and conclude that option trading helps in stock price discovery and dissemination of earnings news. Roll et al (2009) find that option trading enhances information production that is incremental to information production by security analysts. Recently, Jin et al. (2012) examine the relationship between the option measures (volatility spread and skew) and future returns around information events like earnings announcements. They conclude that option traders have an advantage over equity traders in terms of information possession as well as in processing information. Experimental data also supports the notion that option trading improves price discovery. In laboratory experiments, Plott and Sunder (1988) and Klugger and Wyatt (1995) document that the introduction of option type securities results in faster convergence to informationally efficient equilibrium levels (see also Sunder 1995, Figure 6.7, p 497) Option volume relative to stock volume (O/S) and informed trading Liquidity is important to attract informed trading as it allows informed traders to successfully hide their trades (Admati and Pfleiderer, 1988; Glosten and Milogrom, 1985; Kyle, 1985; Pagano, 1989). Chakravarty et al. (2004) note that when option volume is higher relative to stock volume, the price discovery tends to be greater. When investors have private information, they may trade in options markets or in stock markets. If the private information 7

10 conveys good (bad) news, the informed traders will either buy (sell) call options or sell (buy) put options. To the extent that informed traders prefer options markets due to additional advantages offered in these markets, the option volume relative to the stock volume will be higher. Roll et al. (2010) find significant cross-sectional variation in O/S and examine whether the crosssectional variation is consistent with informed trading. The authors find that O/S significantly increases prior to earnings announcements, suggesting that O/S reflects private information on upcoming earnings news. Consistent with this interpretation, they find that O/S positively predicts the absolute magnitude of earnings news and that the effect is more pronounced when the earnings news is negative. In addition, they show the effect is attenuated by larger abnormal returns prior to earnings announcements, which suggests pre-emption of information content of earning announcements, whether the announcement is positive or negative. Johnson and So (2012) provide further empirical and theoretical evidence that O/S reflects private information. They argue that the total option volume alone is not helpful in inferring private information, because in the absence of the direction of the trade (whether buyer initiated or seller initiated), the sign and magnitude of private information cannot be easily inferred. 2 For example, call option volume can be good (bad) news if these trades are initiated by informed buyers (sellers). Similarly, put volume can provide ambiguous signals. Johnson and So (2012) note that main advantages of using O/S are that it uses publicly available information and that it does not require classification of trades as buyer or seller initiated. Johnson and So (2012) develop a model based on short sale and transaction costs in options and stock markets which suggests that, on average, high (low) O/S firms have low (high) future abnormal returns. The authors provide empirical evidence consistent with their model by 2 Pan and Poteshman (2006) use a private data that enables them to identify whether a trade is buyer initiated or seller initiated. They conclude that put-call data predicts future stock returns. 8

11 examining weekly O/S and future return relation. More importantly, the authors document that the levels of O/S is negatively related to abnormal returns associated with earnings announcements and earnings surprises and conclude that O/S reflects informed trading consistent with findings in Roll et al. (2010). Note that finding of Roll et al. (2010) and Johnson and So (2012) on the relation of O/S of abnormal returns around earnings announcements are seemingly consistent, but have different interpretation. Roll et al. (2010) arguments and empirical findings suggest that high levels of O/S contain information on upcoming news (positive or negative), and the effects of levels of O/S on market reactions to earnings news is larger for negative news due to well recognized short sale constraints in equity markets. On the other hand, in Johnson and So (2012) model, due to relative levels of short sale costs and transaction costs in options and equity markets, investors with negative (positive) information will prefer options (equity) markets which suggests high(low) O/S will be followed by low (high) abnormal returns. Both studies agree that high levels of O/S capture informed trading in options markets before earnings announcements but differ whether or not it holds for positive earnings news. 3 Furthermore, note that in case of positive news, informed traders take a preference to trade in the stock market rather than in the 3 Note that Roll et al. (2010) is silent on low levels of O/S which implies less or no private information of option investors. On the other hand, Johnson and So (2012) model suggests that low levels of O/S is associated with upcoming positive news, since investors with positive information will trade in equity markets rather than in options market due to different transactions costs in each market, which seems be to related to liquidity. Validity of these two different views on the relation between low O/S and returns for upcoming public news may be answered by examining relation between O/S and positive earnings. For earnings announcements with positive abnormal returns, Johnson and So (2012) model suggests a negative and significant correlation (i.e. low O/S is informed, low O/S implies larger positive abnormal returns) but Roll et al. (2010) suggests no or positive correlation (high O/S is informed, but not low O/S). In earnings sample with positive abnormal returns, we find that correlation between O/S and earnings returns is 0.07 but insignificant. This supports our interpretation of our results which is more consistent with Roll et al. (2010)'s interpretation. 9

12 option market only because the transaction costs in the option market in Johnson and So model are so high that they offset the leverage advantage offered by trading in the option market Option trading and price response to unexpected earnings Kim and Verrecchia (1991) model price change in response to an anticipated public announcement. Their model shows that the price change is proportional to the unexpected portion of the announcement and its relative importance to the posterior beliefs. The importance of the announcement to the posterior beliefs increases with precision of the announcement and decreases with precision of pre-announcement information. A prediction of the model is that when the pre-announcement information is greater, the price reaction per unit of earnings surprise is lower. Research has shown that analysts produce financial information (Bhushan, 1989; Brennan et al., 1993; Mikhail et al., 2004). Ceteris paribus, it implies that low analysts following will result in low financial information production about these firms. Assuming that option trading increases information production, Roll et al. (2009) hypothesize that the role of options in increasing informational efficiency will be higher for stocks with low analysts following. Roll et al. (2009) find results consistent with this view and conclude that option activity increases a firm s value, and a part of this phenomenon is linked to the increased information production in actively traded option markets (p 356). More recently, both Roll et al. (2010) and Johnson and So (2012) use O/S around earnings announcements. However, neither paper examines the effect of O/S on ERCs. 4 4 Jin et al. (2012) include earnings surprise in a regression only to examine whether option volatility skews have predictive ability beyond what can be explained by earnings surprise (p ). The focus of their analysis is not on the sensitivity of ERC to option activity. 10

13 Johnson and So (2012) examine the predictive ability of O/S by regressing earnings surprise (measured as analysts forecast errors) on ranks of O/S along with several control variables and find a negative and significant coefficient on the O/S rank; they do not investigate the impact of O/S on ERCs either. Truong and Corrado (2014) examine the impact of option volume on price reaction at the time of earnings announcement. However, as noted in Roll et al (2010) and Johnson and So (2012), option volume alone is not an indicator of informed trading and it is difficult to infer the information content from option volume alone, without the knowledge of the origination of the trade. For example, informed traders with good (bad) news can buy (write) call options. In both cases the option volume goes up, but it is difficult to infer the direction of the news without knowing whether the call volume was buyer initiated, an indication of good news or seller initiated, which would be an indication of bad news. Thus it is impossible to infer whether results of Truong and Corrodo (2014) are driven by informed trading or merely a reflection of increased liquidity. In our main tests, we use Amihud liquidity factor as a control for liquidity Option Preemption Hypothesis To summarize the evidence discussed in earlier paragraphs, we expect that crosssectional difference in the O/S, as noted in Roll et al. (2010), will lead to cross-sectional differences in the information production as well, with high O/S firms associated with higher information production. It implies that high (low) O/S firms are likely to have more (less) private information reflected in stock prices prior to an earnings announcement. Thus, consistent with Kim and Verrecchia (1991), the importance of the information content of an earnings release will be relatively lower (higher) for high (low) O/S stocks. 11

14 As stock price discovery increases with an increase in relative option activity, it is likely to increase the measurement error in earnings surprise that uses analysts forecasts as a proxy for market s expectation of earnings in the ERC regressions. One source of measurement error is the inherent optimistic bias in analysts forecast. 5 When option activity increases, the market s expectation moves closer to the true earnings but analysts forecasts do not. As a result, the optimistic bias is further exacerbated. Another source of measurement error arises because after the last analysts forecast has been made improvements in informational quality from option trading will continue to move the market expectation of earnings closer to the true earnings until the actual earnings are announced. This is likely to further misalign the analysts forecast from the market expectation that it is used to proxy and reduce ERCs. To summarize, the pre-announcement information, as well as the measurement error in analyst forecasts, will be higher for high O/S firms as a result of preemption of information content of earnings releases. This leads to the following testable implication of the option preemption hypothesis (stated in the alternate form): H 1 a: In comparison to firms with low O/S, firms with high O/S will have lower ERCs. In the short window before an earnings announcement, options trading can increase due to informed trading, noise trading, or hedging demands against the uncertainty in the earnings announcement. Hence the O/S is likely to be a noisier measure of private information in the short window prior to an earnings announcement. However, profit taking by informed traders prior to an earnings announcement can induce larger pre-announcement abnormal returns (Roll et al, 2010, p 14). It implies a higher preemption of the information content of earnings. Thus, the relationship between the O/S and the magnitude of announcements is likely to be attenuated 5 See Kothari (2001) for a review of literature on optimistic bias of analysts forecast. 12

15 when pre-announcement abnormal returns are larger. This implies that the difference between ERCs of high O/S firms and low O/S firms will increase in the magnitude of pre-announcement abnormal returns. We state this testable implication of the option preemption hypothesis (in the alternate form) as follows: H 2 a: As the magnitude of pre-announcement abnormal returns increases, the difference between ERCs of low O/S firms and high O/S will increase. We test hypotheses H1 and H2 using Fama-MacBeth (1973) cross-sectional regressions on quarterly basis. By using O/S, we are able to measure cross sectional differences in preannouncement information resulting from incremental option activities. On the other hand, the prior literature tests implications of options on ERCs by examining the change in ERC before and after options on a firm s stocks are listed in the options exchange (Skinner, 1990; Ho, 1993; Mendenhall and Fehrs, 1999). As noted earlier in Section 1, these studies find contradictory results: Skinner (1990) documents a decline in ERCs of firms after options are listed on their stocks while MF find opposite results after controlling for change in firm-size and market wide effects contemporaneous with the listing event. To the best of our knowledge, MF represents the current state of knowledge on the relationship between ERCs and option activity. III. Data and Methodology This study requires data on option volume, stock price, stock volume, analysts' quarterly financial forecasts, actual quarterly earnings, and institutional ownership. We get options data from the Ivy-DB Optionmetrics database, stock price and volume data from the University of Chicago Center for Research on Stock Prices (CRSP) database, quarterly earnings and other accounting data from Standard and Poor's Compustat database, financial analyst forecasts and 13

16 actual earnings per share information from detailed files of the Thomson Financial Institutional Brokers Estimate System (IBES) database, and institutional ownership data from the Thomson Financial Institutional Holdings (Form-13F) database. The Ivy-DB Optionmetrics database includes option data from the beginning of 1996; this determines the first year of our study. The final sample consists of 63,175 firm-quarters for which the intersection of these five databases produces non-missing variables used in this study (discussed below) during 59 announcement quarters from the beginning of 1996 to the end of We report the yearly number of firms and firm quarters in Table Insert Table 1 around here Roll et al. (2010) investigate the time series properties and cross sectional determinants of O/S during the period The annual number of firms and the annual trends that we report in the second column of Table 1 for the same period are similar to those reported in Roll et al. (2010). For , the period that is common to both papers, the number of firmquarters is comparable (48,626 firm-quarters in our sample, versus 48,243 firm-quarters in Table 10 of Roll et al., 2010). It is important to note that, unlike Roll et al. (2010), we require firms to have at least $5 stock price around earnings announcements since most institutions (such as mutual funds) are not allowed to invest in stocks with a price less than $5. We require this filter 6 We also compare our sample to other papers that have used the Ivy-DB Optionmetrics database. We have more observations than that used by Johnson and So (2012) in similar time period. The difference is due to the use of more restrictive filters to construct weekly O/S by them. We have slightly less observations than Jin et al. (2012) sample (68,437 to 71,482, depending upon their test design) due our requirement that each observation have at least two analysts forecasts for calculation of dispersion. Without this requirement, our sample is comparable to the upper range of their sample. Requiring at least two analysts is likely to introduce a conservative bias in our study. 14

17 because of the concerns on the information environment of firms that are highlighted in MF. 7 Since smaller firms have poorer information environment, eliminating firms with stock price below $5/share introduces a conservative bias in testing of our hypotheses. We measure the cumulative abnormal returns (CARs) over three trading days (t 1, t+1) around quarterly earnings announcements (t=0). We estimate abnormal returns with a market model estimated over a period ending forty days prior to the announcement date using the CRSP equally weighted index as market proxy. In robustness tests, we also used CRSP value weighted index as well as buy-and-hold abnormal returns (calculated as the difference between the stock return and the three factor Fama-French portfolio return). Our results are robust to different measurements of abnormal returns. We report results from equal-weighted market model to make our results comparable to Roll et al (2010), Skinner (1990) and MF. We calculate the relative option volume as the total option volume scaled by the total stock volume. /,,,, /,, 1 Where: Option Volume i,q,t = the total volume for all traded options on stock i on day t relative to an earnings announcement (q) Stock Volume i,q,t = the trading volume of stock i on day t relative to an earnings announcement (q) Closer to an earnings announcement, options trading increases due to higher volatility (noise trading) as well as due to hedging activities. Informed traders may book their profits by 7 The requirement, that stock prices should be at least $5/share, results in a loss of about 2,000 observations. Our results are similar if we do not apply this filter. 15

18 unwinding their positions before earnings announcements if stock prices have already moved consistent with their private information. As a result, the impact of O/S measured near the earnings announcement is likely to be different than O/S measured further away from the earnings announcement. To capture these differences, we measure O/S over two non-overlapping pre-announcement windows between quarterly earnings announcements. For Long-Window, t1 equals forty-days prior to an earnings announcement (t 40) and t2 equals eleven days (t 11) prior to the announcement. For Short-Window t1 is ten-days prior to an earnings announcement (t 10) and t2 is two-days (t 2) prior to the announcement. Figure 1 shows the measurement of main variables (relative option to stock volume (O/S) and abnormal returns) in event time. We calculate earnings surprise, i.e., the analyst forecast error (AFE), as the difference between the actual earnings-per-share and the latest analysts forecast before earnings are announced, scaled by stock price at t 2. 8 We use the Fama-MacBeth (1973) procedure to examine the relationship between the relative option activity and the sensitivity of stock price returns to the earnings surprise over fifty-nine quarters between Specifically, we run fifty-nine cross-sectional regressions, one for each quarter, and calculate the time-series average of the parameter estimates. We report the statistics based on the Newey-West (1987) procedure that corrects time-series auto-correlation for four lags; the exact number of lags is determined following Newey-West (1994) algorithm. In our cross-sectional regressions, we control for several factors shown by prior research to affect announcement returns, firms information environment, earnings response coefficients, 8 The choice of scaling is similar to Skinner (1990) and Mendenhall and Fehrs (1999). 16

19 and option trade volume. 9 A description of these variables is provided in Appendix A. The controls (used in all regressions) for factors that influence expected return include size (log of the previous quarter market capitalization), book-to-market (as of the previous quarter end), and price momentum. Size also affects information environment of the firm. Instances of options listings as well as options trading are likely to increase with firm-size. Additionally, we control for information environment of a firm by including a log of the number of analyst forecasts made during ninety days prior to an earnings announcement in our regressions. To control for analysts dispersion of opinions, we use the standard deviation of analyst forecast errors, scaled by the previous end-of- the-quarter stock price. We control for idiosyncratic volatility as options become more valuable when volatility is high. This may affect the choice of the trading venue (the options market versus the stock market). 10 Our regressions also include dummy variables for fiscal quarters and industries; we do not report them in tabulated results for the sake of brevity. 4. Empirical Results 4.1. Yearly Distribution of Relative Option Volume Table 1 reports statistics for O/S during our study period. It reports the annual mean, the 25th percentile, the median, the 75th percentile, and the skew for O/S measured over short as well as long windows prior to quarterly earnings announcements. 9 While selecting the control variables, we focus on variables that are likely to affect informational role of options and controls used in prior studies (i.e, Mendenhall and Fehrs, 1999) that examines the effects of option listing on ERCs. As further robustness check, we run cross sectional regressions that includes controls which recent studies control on changes of ERCs and discuss results in Section 5. Our results are robust to earnings persistence and firm characteristics' (such as growth and risk) which are shown to effect changes in ERCs in the cross section. 10 Mayhew and Mihov (2004) find that volatility of stocks is the most important factor that influences the listing decision of option exchanges. 17

20 Trends in annual mean/median O/S suggest that O/S is affected by important systemic events in the past few decades. For example, the low values of mean/median O/S occur in 2002 and 2009 which coincide with significant drops in market levels. These trends of O/S are similar to those reported in Johnson and So (2012). Comparing the skew of O/S (measured over nine or thirty days in our paper) with those reported in Johnson and So (measured over five trading days) we find that our measurements of O/S are smoother. Table 1 also reveals that, on average, the Short-Window mean and the median O/S are higher than the Long-Window mean and the median O/S, respectively. Un-tabulated test results for the difference between the average Shortand Long-window O/S show that the difference is significant at the one-percent level (t= 27.42). These results are consistent with findings of Roll et al. (2010) who report that O/S is higher before earnings announcements Correlation of Main Variables We conduct our main analysis on quartile ranks of O/S in order to have a reasonable number of observations in each group around earnings announcements. Our choice of forming quartiles of O/S is also consistent with the recent related literature. For example, Jin et al. (2012) use quartiles of option measures while investigating the information production around earnings announcements. Similarly, Johnson and So (2012) group the two highest and the two lowest deciles in their factor regressions for analysis of the relationship between weekly O/S and future abnormal returns. Table 2 provides Pearson correlation coefficients between the ranks of O/S, announcement returns (CARs), analyst forecast errors (AFE) and control variables used in this study. Since O/S quartile rank is unsigned, we also report correlations of O/S quartile ranks with absolute values of CARs and AFEs. Correlations of O/S quartile ranks with CARs and CARs 18

21 are negative and positive, respectively; this is true for both Short-Window and Long-Window. These correlations are significant at one-percent level. For Short-Window, correlations of O/S quartile rank with AFE and AFE are negative and positive, respectively, but neither is statistically significant. For Long-Window also, correlations of O/S quartile rank with AFE and AFE are negative and positive, respectively; only the correlation with AFE is significant (at ten-percent level). For both windows, the correlation between O/S quartile rank and Size is around 0.27 and significant (at one-percent level). This suggests that a portion of effects we capture in following analyses may be driven by size effects. Although we control for size in our cross sectional regressions, to further alleviate concerns related to size effects, we conduct additional tests in Section 5 to examine whether our results are driven by size Relative Option Volume and ERC Regressions For convenience, we repeat here the definition of earnings response coefficient (ERC), provided earlier in footnote 2. It is the coefficient of earnings surprise, (measured as AFE), in a regression of abnormal returns (measured over a three-day window, t = 1 to +1, where t=0 is the day of earnings announcement) on earnings surprise. We examine the ERCs, or β1, for the full sample as well as for four different quartiles O/S by estimating the following regression equation. 2 Following the Fama-MacBeth procedure, we first estimate the above equation using the OLS regression in each quarter (the quarter subscript is not shown for the sake of brevity). Table 3 reports the time-series average of coefficient estimates from these cross-sectional regressions 19

22 over fifty nine quarters. We control for several variables (shown as Xk in equation 2) including size, book-to-market, price momentum, volatility, analyst following, dispersion of analysts, institutional ownership, fiscal quarters, and the industry. Panel A (Panel B) of Table 3 reports results using O/S quartile break-points based on the Short- (Long-) Window Insert Table 3 around here Column I reports results for the full sample without considering the effects of relative option to stock volume. In this case, the coefficient for AFE, β1, is as reported in column I. It is statistically significant at one-percent level (t =8.92). Columns II, III, IV and V report results for the low O/S quartile, the low-mid O/S quartile, the high-mid O/S quartile and the high O/S quartile, respectively. Coefficient β1 are significant at the one-percent level for each quartile in Short- and Long- windows. Results also show a monotonic decline in β1, from the lowest O/S quartile to the highest O/S quartile. These results hold for both Short-Window and Long- Window, as shown in Panel A, and Panel B, respectively. To examine the statistical difference of ERCs between high and low O/S firms, we use an indicator variable (OSRank) that captures the differences between high and low O/S firms and estimate the following equation. 3 OSRank is an indicator variable that equals 0.5 for stocks in the lowest O/S quartile, equals +0.5 for stocks in the highest O/S quartile, and equals zero for the remaining two midquartiles. Hence, β3, the coefficient on AFE x OSRank, directly measures the difference between 20

23 ERCs for the high O/S and the low O/S stocks. The coefficient on AFE shows the ERC for midquartiles as a benchmark. We report the regression results of equation 3 following Fama-MacBeth procedure in Column VI of Panel A (Panel B) for Short (Long)-Window of Table 3. The coefficient β1 of AFE for mid-os quartiles based on Short- (Long-) Window O/S breakpoints is (2.015) and is statistically significant at the one-percent level. The coefficient of the indicator variable OSRank, β2, is negative but statistically insignificant. This result may be explained by the fact that OSRank is always positive, but CAR can be either positive or negative. The test of hypothesis H1 focuses on β3, the coefficient of the interaction term between AFE and OSRank, which measures the difference between the AFE coefficient for the highest and the lowest O/S quartile firms. The option preemption hypothesis predicts that β3 < 0. Coefficient β3 for Short- (Long-) Window O/S is ( 1.213). It is statistically significant at the five-percent level (t = 2.51) when O/S ranks are based on Short-Window and at onepercent level (t= 4.89) when O/S ranks are based on Long-Window. The coefficient β3 suggest that the ERCs for the high O/S firms are lower than ERCs for the low O/S firms by 29.5 (46.3) percent in case of Short- (Long-) Window. These results document an economically and statistically significant decline in the ERCs for the highest O/S quartile firms in comparison to the lowest quartile O/S firms. Overall, results reported in Table 3 suggest that ERCs are attenuated when there is relatively high volume in options compared to the volume in underlying stocks between quarterly earnings announcements. The findings are consistent with hypothesis H Non-parametric (un-tabulated) tests also confirm these findings. The probability that AFE (low O/S Rank) equals AFE (high O/S rank) is 0.5 under null hypothesis, assuming binomial distribution. The onetailed test rejects the null at p = for Short-Window O/S, and at p < for Long-Window O/S. 21

24 4.4. Relative Option Volume, Pre-announcement Returns and Preemption of Earnings Information This section tests the second hypothesis of our paper. Roll et al. (2010) examine whether preannouncement abnormal returns of large magnitude attenuate the relationship between O/S and the magnitude of announcement period returns. We follow their methodology but extend the investigation to examine effects of the magnitude of pre-announcement abnormal returns on the relationship between OSRank and ERCs. We estimate equation (4) to examine whether pre-announcement returns during short window exacerbate the difference between announcement returns of High O/S stocks and announcement returns of Low O/S stocks. 4 As before, we use Fama-Macbeth estimation procedure. The quarter subscript in the equation is dropped for the sake of brevity. Table 4 reports the coefficient estimate and t- statistics in two sub-columns of Column I. The coefficient on OSRank is small, positive, and significant, implying that the high O/S stocks are associated with larger announcement returns compared to the low O/S stocks. As predicted by Hypothesis 2, the coefficient β2 on the interaction of OSRank with PreCAR is negative and significant; furthermore, it is much larger in magnitude in comparison to the coefficient on OSRank. The large negative magnitude of this interaction term yields a negative combined effect, i.e., for a large magnitude of pre-car, stocks with high O/S are associated with smaller announcement returns compared to stocks with low O/S. Higher option trading brings prices closer to their full information value prior to an earnings announcement. Our findings are qualitatively similar to those reported by Roll et al. 22

25 (2010, p 15) on the relationship between daily O/S, pre-car, and the magnitude of announcement CARs Insert Table 4 around here Next, we examine the effect of pre-announcement returns on the relationship between relative option trading and ERCs. We expect pre-announcement stock returns during the O/S measurement period to exacerbate the difference in ERCs between high and low O/S stocks. To examine this, we use an interaction term of OSRank, AFE and the magnitude of pre-cars. First, we analyze this relationship by using magnitudes of CARs and AFEs in equation (5). (5) Equation (5) is estimated using the Fama-MacBeth procedure; the quarter subscript in equations is dropped for the sake of brevity. The results are reported in sub-columns of Column II in Table 4. Hypothesis 2 predicts that β4, the coefficient on AFE OSRank PreCAR should be negative. As predicted, β4, is negative and significant, indicating that for a given level of pre-announcement returns, the coefficient on AFE is lower for high O/S stocks. As pre-announcement returns increase, then the difference in coefficients on AFE between low O/S and high O/S is further exacerbated, as predicted by the option preemption hypothesis. equation (6). Next, we estimate equation (6) using CARs and AFE, rather than their magnitudes in. (6) 23

26 Equation (6) focuses on ERC regressions and uses Fama-Macbeth method as in earlier equations. Results for equation (6) are reported in sub-columns of Column III of Table 4. The coefficient on AFE is and is significant at one-percent level (t = 9.93). The coefficient on AFE OSRank PreCAR is and statistically significant at one-percent level (t= 3.21). This implies that when the preemption is high (as reflected by higher magnitude of PreCAR) the difference in ERCs for the low O/S and the high O/S stocks is exacerbated. Positive coefficients on OSRank, reported for the first two regressions in Table 4, warrant further comment. Although the coefficients are positive and statistically significant, they are nonetheless economically very small in absolute terms as well as relative to other regression coefficients of interactive terms. The overall impact of OSRank is negative and significant as it is dominated by the coefficients of interactive terms. MF also find that the magnitude of total announcement returns does not decrease after option listing, actually it becomes higher. The authors argue that perhaps option traders are successful at anticipating cases where total announcement returns are larger. We argue that the incentives for investors to engage in information activity will be larger in cases when large announcement returns are likely, since in these cases investors will be compensated via trading profits for costly information production. Hence, our findings on larger magnitude of total earnings announcement returns is not surprising and is consistent with the view that informed traders prefer to trade in options rather than underlying stocks to maximize their trading profits. 5. Sensitivity Analyses 5.1. Firm Size and Relative Option Volume Controlling for a firm s information environment, especially the size, is important according to previous literature as it directly impacts the option preemption hypothesis. Given 24

27 its importance for main findings of our paper, we control for size in our regressions while examining the effects of different levels of O/S on AFE. Correlation coefficients reported in Table 2 suggest that stocks with high O/S tend to be associated with a richer information environment, at least when information environment is measured by firm size. This evidence is especially relevant given the mixed evidence in prior literature. For example, MF find that after controlling for size effects, the decline in ERCs after option listing, as documented in Skinner (1990) either disappears, or reverses the direction of change. Note that earlier research uses an indicator variable for option listing status with a focus to examine the ERCs before and after the listing; however, the change in listing status is likely to be contemporaneous to changes in factors like size that influence an option exchange s decision to list. In contrast, our analysis is based on cross-sectional differences of option trading where we control for firm size. Thus, our analysis is not subject to criticisms of MF. In order to alleviate further concerns about size effects, we conduct two additional tests to show that O/S measure captures effects above and beyond size effects. In the first sensitivity test, we sort the sample independently into quartiles of O/S and size (Market capitalization). Table 5 reports the number of firm-quarters in various size quartiles for each O/S quartile. Rows where O/S and Size have the same rank are highlighted. The bottom row shows the percentage of firms that have the same rank of O/S and size Insert Table 5 around here Table 5 shows that only about 30 (43) percent of stocks in Low (High) O/S quartile are also ranked in the Low (High) Size quartile. The remaining 70 (57) percent of stocks that are classified as Low (High) O/S are not classified in the Low (High) Size quartile. In the overall sample, only around 32 percent of the firm quarters have the same O/S quartile ranks as the Size 25