The Joint Cross Section of Stocks and Options* Andrew Ang† Columbia University and NBER Turan G. Bali‡ Georgetown University Nusret Cakici§ Fordham University This Version: 15 February 2012 Keywords: implied volatility, risk premiums, predictability, short-term momentum JEL Classification: G10, G11, C13.                                                              * We thank Reena Aggarwal, Allan Eberhart, Larry Glosten, Bob Hodrick, Michael Johannes, George Panayotov, Tyler Shumway, Mete Soner, David Weinbaum, Liuren Wu, Yuhang Xing, the editor, associate editor, two referees, and seminar participants at the American Finance Association meetings, ETH-Zurich, Federal Reserve Bank of New York, and Georgetown University for helpful comments. Additional results are available in an internet appendix that can be obtained by contacting the authors. † Columbia Business School, 3022 Broadway, 413 Uris, New York, NY 10027. Phone: (212) 854-9154, Email: [email protected]. ‡ Corresponding author. McDonough School of Business, Georgetown University, Washington, D.C. 20057. Phone: (202) 687-5388, Fax: (202) 687-4031, E-mail: [email protected]. § Graduate School of Business, Fordham University, 113 West 60th Street, New York, NY 10023, Phone: (212) 636-6776, Email: [email protected]. The Joint Cross Section of Stocks and Options ABSTRACT Option volatilities have significant predictive power for the cross section of stock returns and vice versa. Stocks with large increases in call implied volatilities tend to rise over the following month and increases in put implied volatilities forecast future decreases in next-month stock returns. The spread in average returns and alphas between the first and fifth quintile portfolios formed by ranking on lagged changes in implied call volatilities is approximately 1% per month. Going in the other direction, stocks with high returns over the past month tend to have call option contracts that exhibit increases in implied volatility over the next month, but realized volatility for those stocks tends to decrease. 1. Introduction Options are redundant assets only in an idealized world of complete markets with no transactions costs, perfect information, and no restrictions on shorting. Not surprisingly, since in the real world none of these assumptions hold, options are not spanned by stock prices and option prices are not merely functions of underlying stock prices and risk-free securities. Many theoretical models jointly pricing options and underlying assets in incomplete markets have incorporated many of these real-world frictions.1 In addition, if informed traders tend to choose certain markets over others, information-based models such as Easley, O’Hara and Srinivas (1998) predict that those markets where informed trading takes place will lead other markets where informed trading does not predominate. Option markets have significant advantages for informed traders as enumerated by Black (1975), Grossman (1977), Diamond and Verrechia (1987), and others. Options offer an alternative way to take short positions when short positions in the underlying asset would be prohibitively expensive. Options provide additional leverage which may not be possible, or relatively expensive, to obtain in stock and bond markets (see Back, 1993; Biais and Hillion, 1994). Options also reduce transactions costs of making replicating trades in the underlying stocks. On the other hand, an informed trader may not always first choose to trade in options markets over underlying equity markets. Option trading volumes are much lower than trading volumes in the underlying stocks, which make hiding informed trades harder in option markets. As Easley, O’Hara and Srinivas (1998) show, only when the implicit leverage available in options is large and the option market offers sufficient liquidity will informed investors first trade in option markets. Conversely, if informed traders can more easily hide in equity markets, then equity returns will lead option prices. We document that the cross section of option volatilities contains information that forecasts the cross section of expected stock returns. Stocks with call options that have experienced increases in implied volatilities over the past month tend to experience high expected returns over the next month. Increases in put option volatilities predict decreases in next-month stock returns. While strongest for the next-month horizon, this predictability persists for several months. The strength and persistence of this predictability is remarkable. First, the innovation in implied volatilities can be considered to be a very simple measure of news arrivals in the option market. The predictability at the standard monthly horizon suggests the predictability is unlikely due to microstructure trading effects. In contrast, most of the                                                              1 See Detemple and Selden (1991), Back (1993), Cao (1999), Buraschi and Jiltsov (2006), and Vanden (2008), among others. 1 previous literature investigating lead-lag effects of options versus stock markets focuses on intra-day or daily frequencies. The predictability is statistically very strong and economically large. Quintile portfolios formed on past changes in call option volatility have a spread of approximately 1% per month in both raw returns and alphas computed using common systematic factor models. The difference between the top and bottom quintile portfolios in ranking stocks by past changes in put implied volatilities is approximately 60 basis points per month after controlling for the effect of call volatility innovations. The predictability of stock returns by option innovations is also robust in several subsamples. Whereas many cross-sectional strategies have reversed sign or become much weaker during the recent financial crisis, the ability of option volatilities to predict returns is still seen in very recent data. In particular, the predictive relation between large put volatility innovations and future low stock returns is very prominent in 2008. Although calls and puts contain different information, a consistent story for this predictability is that option prices change due to the action of informed investors, and this information is not contemporaneously reflected in equity markets. Thus, at least some informed traders are moving first in option markets. Bullish investors first trading in option markets could either buy call options or equivalently sell put options, and both higher call volatilities and lower put volatilities lead future increases in stock returns. Consistent with the actions of informed investors, the predictability of implied option volatility changes is strongest when they are accompanied by large contemporaneous changes in option volume. In addition, we find that the economic source of the return predictability by option innovations is almost all due to changes in idiosyncratic, not systematic, components of implied volatilities. This is consistent with the investors first trading option markets having better information about company-specific events or news. We also uncover evidence of reverse directional predictability from stock variables to option markets. Many of the variables long known to predict stock returns also predict option implied volatilities. A very simple predictor is the past return of a stock: stocks with high past returns over the previous month tend to have call options that exhibit increases in volatility over the next month. In particular, stocks with abnormal returns of 1% relative to the CAPM tend to see call implied volatilities increase over the next month by approximately 2%. There are no corresponding increases in realized stock volatilities over the next month. This predictability also persists for several months. The predictability of lagged stock returns and other stock characteristics for future changes in option implied volatilities is in addition to the well-known documented predictability of option market returns by option 2 market variables.2 Behavioral over-reaction models of option mispricing predict that option implied volatilities should increase together with other measures of uncertainty such as earnings dispersion. We find this is not the case. The cross-sectional predictability of option volatilities is stronger in stocks which exhibit a lower degree of return predictability and options which are harder to hedge, consistent with rational stories on option volatility predictability. 2. Related Literature We follow an older literature that debates whether options or stocks lead or lag each other. These earlier studies are all conducted at the daily or intra-day frequencies. Manaster and Rendleman (1982), Bhattacharya (1987), and Anthony (1988) find that options predict future stock prices. Fleming, Ostdiek and Whaley (1996) document derivatives lead the underlying markets using futures and options on futures. On the other hand, Stephan and Whaley (1990) and Chan, Chung and Johnson (1993) find stock markets lead option markets. Chakravarty, Gulen and Mayhew (2004) find that both stock and option markets contribute to price discovery. Our findings are very different from this literature because we find that option volatility innovations contain strong predictive power for the cross section of equity returns at the much lower monthly frequency. Similarly, we find predictability of stock characteristics, including past one-month stock returns, for future implied volatilities at the monthly frequency. The joint predictability of the cross section of option implied volatilities on stock returns and vice versa indicates that both options and underlying equities play an important role in the price formation of each other’s markets. Our empirical findings are partly consistent with the model of Easley, O’Hara and Srinivas (1998) where informed traders place orders in the equity market, the options market, or both. If at least some informed investors choose to trade in options before trading in underlying stocks, then some option prices will predict future stock price movements. Conversely, if stock markets are more liquid and informed traders can more easily hide their trades in equities, then stock markets may lead option markets. Easley, O’Hara and Srinivas find evidence that option volumes of certain types of trades forecast future stock prices within the next hour using intraday data, consistent with price updating occurring according to a microstructure model. The monthly frequency predictability we find, however, is probably not due to microstructure frictions. Our findings are related to a more recent literature showing that option prices contain predictive information about stock returns. Cao, Chen and Griffin (2005) find that merger information hits the call                                                              2 Obviously this predictability overwhelmingly rejects arbitrage-free option pricing models, which a long literature has also shown. The earlier papers in this literature include Figlewski (1989) and Longstaff (1995). More recently see Cao and Han (2009) and Goyal and Saretto (2009). 3 option market prior to the stock market, but focus only on these special corporate events. More recent studies such as Bali and Hovakimian (2009), Cremers and Weinbaum (2010), and Xing, Zhang, and Zhao (2010) use information in the cross section of options including the difference between implied and realized volatilities, put-call parity deviations, and risk-neutral skewness. Our option volatility innovation measures are simpler, and the predictive ability of changes in implied volatilities is different to option market information used by previous authors. We relate changes in option volatilities to contemporaneous changes in option volume, indicating that that informed trading is likely taking place in option markets and moving option prices before being captured in stock markets. Other related studies focus on predicting option returns, option trading volume, or the option skew in the cross section. Goyal and Saretto (2009) show that delta-hedged options with a large positive difference between realized and implied volatility have low average returns. We examine a much larger set of both stock and option market predictors of option-implied volatilities than they use and use option volatilities themselves, rather than option straddle returns. Roll, Schwartz and Subrahmanyam (2009) examine the contemporaneous, but not predictive, relation between options trading activity and stock returns. Dennis and Mayhew (2002) document cross-sectional predictability of risk-neutral skewness, but do not examine the cross section of implied volatilities. We find many of the “usual suspects” in the commonly used stock characteristics that predict stock returns also predict the cross section of option- implied volatilities, like book-to-market ratios, momentum, and illiquidity measures. We focus on the strong predictive power of the lagged stock return in the cross section, which to our knowledge has been examined only in the context of options on the aggregate market by Amin, Coval and Seyhun (2004). 3. Data 3.1. Implied Volatilities The daily data on option implied volatilities are from OptionMetrics. The OptionMetrics Volatility Surface computes the interpolated implied volatility surface separately for puts and calls using a kernel smoothing algorithm using options with various strikes and maturities. The underlying implied volatilities of individual options are computed using binomial trees that account for the early exercise of individual stock options and the dividends expected to be paid over the lives of the options. The volatility surface data contain implied volatilities for a list of standardized options for constant maturities and deltas. A standardized option is only included if there exists enough underlying option price data on that day to accurately compute an interpolated value. The interpolations are done each day so that no forward- 4 looking information is used in computing the volatility surface. One advantage of using the Volatility Surface is that it avoids having to make potentially arbitrary decisions on which strikes or maturities to include in computing an implied call or put volatility for each stock. In our empirical analyses, we use call and put options’ implied volatilities with a delta of 0.5 and an expiration of 30 days. For robustness we also examine other expirations, especially of 91 days, which are available in the internet appendix. Our sample is from January 1996 to September 2008. In the internet appendix, we also show that our results are similar using implied volatilities of actual options rather than the Volatility Surface. Table 1 contains descriptive statistics of our sample. Panel A reports the average number of stocks per month for each year from 1996 to 2008. There are 1292 stocks per month in 1996 rising to 2175 stocks per month in 2008. We report the average and standard deviation of the end-of-month annualized call and put implied volatilities of at-the-money, 30-day maturities, which we denote as CVOL and PVOL, respectively. Both call and put volatilities are highest during 2000 and 2001 which coincides with the large decline in stock prices, particularly of technology stocks, during this time. During the recent finance crisis in 2008, we observe a significant increase in average implied volatilities from around 40% to 54% for both CVOL and PVOL.3 3.2. Predictive Variables We obtain underlying stock return data from CRSP and accounting and balance sheet data from COMPUSTAT. We construct the following factor loadings and firm characteristics associated with underlying stock markets that are widely known to forecast the cross section of stock returns: 4 Beta: To obtain the monthly beta of an individual stock, we estimate market model regressions at the daily frequency: R r  (R r ) , (1) i,d f,d i i m,d f,d i,d                                                              3 There are many reasons why put-call parity does not hold, as documented by Ofek, Richardson and Whitelaw (2004) and Cremers and Weinbaum (2010), among others. In particular, the exchange-traded options are American and so put-call parity only holds as an inequality. The implied volatilities we use are interpolated from the Volatility Surface and do not represent actual transactions prices, which in options markets have large bid-ask spreads and non-synchronous trades. These microstructure issues do not affect the use of our option volatilities as predictive instruments observable at the beginning of each period. 4 Easley, Hvidkjaer, and O’Hara (2002) introduce a measure of the probability of information-based trading, PIN, and show empirically that stocks with higher probability of information-based trading have higher returns. Using PIN as a control variable does not influence the significantly positive (negative) link between the call (put) volatility innovations and expected returns. We also examine the effect of systematic coskewness following Harvey and Siddique (2000). Including coskewness does not affect our results. See the internet appendix. 5 where R is the return on stock i on day d, R is the market return on day d, and r is the risk-free i,d m,d f,d rate on day d. We take R to be the CRSP daily value-weighted index and r to be the Ibbotson risk- m,d f,d free rate. We estimate equation (1) for each stock using daily returns over the past month. The estimated ˆ slope coefficient  is the market beta of stock i in month t. i,t Size: Following the existing literature, firm size is measured by the natural logarithm of the market value of equity (stock price multiplied by the number of shares outstanding in millions of dollars) at the end of the month for each stock. Book-to-Market Ratio (BM): Following Fama and French (1992), we compute a firm’s book-to-market ratio in month t using the market value of its equity at the end of December of the previous year and the book value of common equity plus balance-sheet deferred taxes for the firm’s latest fiscal year ending in prior calendar year. To avoid issues with extreme observations we follow Fama and French (1992) and Winsorize the book-to-market ratios at the 0.5% and 99.5% levels. Momentum (MOM): Following Jegadeesh and Titman (1993), the momentum variable for each stock in month t is defined as the cumulative return on the stock over the previous 11 months starting 2 months ago to avoid the short-term reversal effect, i.e., momentum is the cumulative return from month t–12 to month t–2. Illiquidity (ILLIQ): We use the Amihud (2002) definition of illiquidity and for each stock in month t define illiquidity to be the ratio of the absolute monthly stock return to its dollar trading volume: ILLIQ |R |/VOLD , where R is the return on stock i in month t, and VOLD is the monthly i,t i,t i,t i,t i,t trading volume of stock i in dollars. Short-term reversal (REV): Following Jegadeesh (1990), Lehmann (1990), and others, we define short- term reversal for each stock in month t as the return on the stock over the previous month from t–1 to t. Realized volatility (RVOL): Realized volatility of stock i in month t is defined as the standard deviation of daily returns over the past month t, RVOL  var(R ) . We denote the monthly first differences in i,t i,d RVOL as ΔRVOL. 6 The final set of predictive variables is from option markets: Call/Put (C/P) volume: The relation between option volume and underlying stock returns has been studied in the literature, with mixed findings, by Stefan and Whaley (1990), Amin and Lee (1997), Easley, O’Hara, and Srinivas (1998), Chan, Chung, and Fong (2002), Cao, Chen, and Griffin (2005), and Pan and Poteshman (2006), and others. Following Pan and Poteshman (2006), our first measure of option volume is the ratio of call/put option trading volume over the previous month. Call/Put open interest (C/P OI): A second measure of option volume is the ratio of open interests of call options to put options. Realized-Implied volatility spread (RVOL-IVOL): Following Bali and Hovakimian (2009) and Goyal and Saretto (2009), we control for the difference between the monthly realized volatility (RVOL) and the average of the at-the-money call and put implied volatilities, denoted by IVOL, (using the Volatility Surface standardized options with a delta of 0.50 and maturity of 30 days). Bali and Hovakimian (2009) show that stocks with high RVOL-IVOL spreads predict low future stock returns. Goyal and Saretto (2009) find similar negative effect of the RVOL-IVOL spread for future option returns. Risk-neutral skewness (QSKEW): Following Conrad, Dittmar and Ghysels (2009) and Xing, Zhang and Zhao (2010), we control for the risk-neutral skewness defined as the difference between the out-of-the- money put implied volatility (with delta of 0.20) and the average of the at-the-money call and put implied volatilities (with deltas of 0.50), both using maturities of 30 days. Xing, Zhang and Zhao (2010) show that stocks with high QSKEW tend to have low returns over the following month. On the other hand, Conrad, Dittmar and Ghysels (2009) find the opposite relation with a more general measure of option skewness derived from using the entire cross section of options based on Bakshi, Kapadia and Madan (2003). 7 3.3. Measures of Volatility Innovations First Differences of Implied Volatility Levels The first, and simplest, definition of volatility innovations is the change in call and put implied volatilities, which we denote as ΔCVOL and ΔPVOL, respectively:5 CVOL CVOL CVOL , i,t i,t i,t1                                                            (2) PVOL  PVOL PVOL . i,t i,t i,t1 While the first difference of implied volatilities is a very attractive measure because it is simple, it ignores the fact that implied volatilities are predictable in both the time series and cross section. Our two other measures account for these dimensions of predictability. Time-Series Innovations Implied volatilities are well known to be persistent. To take account of the autocorrelation, we assume an AR(1) model for implied volatilities and estimate the following regression using the past two years of monthly data: CVOL a b CVOL c , i,t ci ci i,t1 i,t (3) PVOL a b PVOL p. i,t pi pi i,t1 i,t We define the current shock in call and put implied volatilities for stock i in month t as the monthly innovations in call and put implied volatilities. That is, we assign the time t value of c and p i,t i,t as the option innovations and denote them as CVOLshock andPVOLshock, respectively, with the “ts” ts ts subscript denoting that they are innovations derived from time-series estimators. Note that the ΔCVOL and ΔPVOL first difference measures implicitly assume that b b 1. ci pi Cross-Sectional Innovations We can alternatively estimate monthly innovations in volatilities by exploiting the cross-sectional predictability of implied volatilities. We denote the cross-sectional innovations as CVOLshock and cs                                                              5 As an additional robustness check, we also consider proportional changes in CVOL and PVOL and find very similar results. The results from the percent changes in call and put implied volatilities (%ΔCVOL, %ΔPVOL) are available in the internet appendix. 8