Implied Realized Vol Cross Section
Implied Realized Vol Cross Section
Implied Realized Vol Cross Section
Manuel Ammann, David Skovmand, Michael Verhofen University of St. Gallen and Aarhus School of Business
Abstract Using a complete sample of US equity options, we analyze patterns of implied volatility in the cross-section of equity options with respect to stock characteristics. We nd that high-beta stocks, small stocks, stocks with a low-market-to-book ratio, and non-momentum stocks trade at higher implied volatilities after controlling for historical volatility. We nd evidence that implied volatility overestimates realized volatility for low-beta stocks, small caps, low-market-to-book stocks, and stocks with no momentum and vice versa. However, we cannot reject the null hypothesis that implied volatility is an unbiased predictor of realized volatility in the cross section. Keywords: Implied Volatility, Realized Volatility JEL classication: G10
Manuel Ammann ([email protected]) is professor of nance at the University of St. Gallen, Switzerland, David Skovmand ([email protected]) is assistant professor at the Aarhus School of Business, University of Aarhus and CREATES, and Michael Verhofen ([email protected]) is lecturer in nance at the University of St. Gallen, Switzerland. We thank Sebastien Betermier, Peter Feldhtter, Thomas Gilbert, Sara Holland, Peter Tind Larsen, Miguel Palacios, Hari Phatak, Ryan Stever, and Stephan Sss for helpful comments.
Introduction
In this paper, we contribute to the existing literature by analyzing the relation between implied and realized volatility in the cross-section of equity options. While existing studies use small samples of option price data or only index options, we use a complete, survival bias free database of US equity options. Moreover, our focus is to analyze the relation between corporate fundamentals, implied and realized volatility while almost all existing studies focus on time-series models of volatility (e.g., GARCH). Implied volatility can be regarded as the market forecast of future s volatility (Latan & Rendleman (1976), Poterba & Summers (1986), Day & Lewis (1988), Harvey & Whaley (1992), Sheikh (1989)). As for any forecast, the question arises how accurate this forecast is and whether this forecast is aected by any biases. A number of authors analyze the accuracy of this forecast and the results have primarily shown a dependence on the analyzed underlying asset price process and the time period under investigation. The a rmative case for implied volatility can be found in Christensen & Prabhala (1998) and Jorion (1995), for example. The latter uses data from the foreign exchange market and nds that implied volatility is the best available forecast for realized volatility. The result, however, is subject to a number of biases. For example, he reports that implied volatility itself is too volatile. Day & Lewis (1992) use time-series models of volatility as a starting point and analyze the additional forecast contribution provided by implied volatility. Using S&P 100 options, the results suggests that implied volatility contains additional information over GARCH models. On the contrary, Canina & Figlewski (1993) analyze the forecasting performance of implied volatility. They nd that implied volatility has no material correlation with realized volatility. Poon & Granger (2005) review 93 studies that conduct tests of volatility-forecasting methods on a wide range of nancial asset returns. They nd evidence that option-implied volatility is superior to time-series based forecasts. Within the group of time series models, they nd that historical volatility is superior to GARCH and stochastic volatility models. While most authors have analyzed index options, only very few have analyzed individual equity options. For example, Lamoureux & Lastrapes (1993) analyze the relation between implied volatility and realized volatility for 10 equity options between 1982 and 1984. They nd that time series models (such as GARCH) can improve forecasts of future realized volatility. Bakshi & Kapadia (2003a) analyze the volatility risk premium embedded in index options. A portfolio long in call options and short in a delta-hedged 2
portfolio of the underlying delivered a return below 0 in the sample period. Therefore, they argue that the volatility risk premium is negative. Similarly, Bakshi & Kapadia (2003b) argue that this nding also holds for 25 individual equity options albeit not as signicantly as in the case of index options. This study diers from existing studies in two ways. First, we use a complete, survival bias free sample of all US equity options over a time period of nine years (1997 to 2005) provided by OptionMetrics. Therefore, our sample size is much larger than the samples used in existing studies that usually focus on index options. Second, the focus of our study is on the relation between stocks characteristics, implied volatility and historical volatility. Since most existing studies focus on time series models of volatility (GARCH-type models), this is a perspective neglected so far. We focus on stock characteristics (such as beta, size, market-to-book, and momentum) for a number of reasons. First, little is known about the interaction of stock characteristics, implied volatility, and historical volatility although there are good reasons why there might be a relation. For example, stocks with certain characteristics might be perceived more or less risky by market participants after controlling for historical volatility. Consequently, cross-sectional dierences in the predictive power of implied volatility might be priced. Second, we seek an underlying economic explanation of volatility beyond pure statistical models. Time-series models of implied volatility might capture the dynamics of volatility, but usually neglect the economic foundations of volatility, e.g., why similar stocks might have dierent historical or implied volatilities. Our main ndings can be stated as follows. We nd that high beta stocks, small stocks, stocks with a low market-to-book ratio, and non-momentum stocks trade at higher implied volatilities than justied by historical volatility. We nd evidence that implied volatility overestimates future realized volatility for low beta stocks, small caps, low market-to-book, and stocks with no momentum. Implied volatility underestimates realized volatility for stocks with the opposite characteristics. However, we cannot reject the null hypothesis that implied volatility is an unbiased predictor of realized volatility in the cross section. Section 2 starts by presenting the research design approach. Section 3 presents the empirical results. Section 4 concludes.
2
2.1
Research Design
Data
The option price data in this study are provided by OptionMetrics via WRDS. The raw data set consists of implied volatilities for all US equity options. In this study, we use implied volatilities for standardized call options with a maturity of 91 calendar days and a strike price equal to the forward price. The sample starts in January 1996 and ends in April 2006. Implied volatilities are computed as described in OptionMetrics (2005). Beside implied volatilities, return data and historical volatilities are retrieved from OptionMetrics. Historical volatility is also computed over a time period of 91 calendar days. To form portfolios on size and market-to-book, the dataset was merged with Compustat data. To compute the market values, the data elds 24 (Price - Calendar Year - Close) and 25 (Common Shares Outstanding) are multiplied. To compute market-to-book-ratios, the information in data eld 60 (Common Equity) was used. To compute the beta, the market risk premium as provided by Fama and French (via WRDS) is used.
2.2
Portfolio Formation
To explore the relation between historical and implied volatility, we use a portfolio formation approach as widely used in empirical asset pricing (see e.g., Fama & French (1993), Fama & French (1996)). In particular, we use two dimensional (5 5) sorted portfolios based on historical volatility and a second variable. We have selected the 5 5 sorting because it ensures that a su cient number of stocks is included in any portfolio at any point in time. This research design enables us to assess cross-sectional eects, i.e., we can capture dierent patterns for stocks with specic characteristics. For example, we can explore whether small and large stocks with similar historical volatility trade at dierent implied volatility or have dierent predictive power. Portfolio construction can be illustrated by using the size portfolio for 1997. At the end of 1996, we check which stocks have market value information available in Compustat as well as which stock options have implied and historical volatility data available in OptionMetrics. Using all stocks for which all information is available, we compute the breakpoints for historical volatility and market value. Then, we assign each stock into the corresponding portfolio. As is standard in asset pricing, we keep the portfolio constant over the year. We compute the average implied and historical volatility in 4
each portfolio on a daily basis. At the end of 1997, the procedure is repeated to nd the portfolio for 1998. We drop the rst year of data (1996) for the purpose of portfolio formation because we need the data in this year for the formation of the momentum portfolio and the beta portfolio. Moreover, historical volatilities in the rst three months of 2006 are included to assess the accuracy of implied volatilities in the last three months of 2005. This gives us a time series of implied volatility for each portfolio in the time period between January 1997 and December 2005. For the historical volatility, data is available until March 2006. Table 1 shows the breakpoints for beta, size, market-to-book, and momentum. Overall, the results are as expected. The breakpoints re the ect development of the US stock market during the sample period. For example, the market value breakpoints rise until 2000, drop between 2000 and 2003, and rise again after 2003. The market-to-book ratio shows the same pattern. For the momentum breakpoints, this evolution is re ected in the number of negative breakpoints. For example, only one breakpoint (Q1 to Q2) is negative in 1997, while three breakpoints are negative in 2003. Table 2 shows the average number of stocks in each portfolio over the sample period. For the momentum premium and the market-to-book ratio, there is always a su ciently high number of stocks in each portfolio (at least 30). However, a few low numbers can be observed. As expected, the combination high-beta, low-volatility (Q5 and Q1) and low-market value, low volatility gives somewhat low numbers (5 and13). But since we generally observe a su ciently high number of stocks in most portfolios, the results are not likely to be negatively aected by the few small-number portfolios.
2.3
Spread Calculation
To capture the time series behavior of implied volatility in the cross-section, we construct four auxiliary time series. The construction of these variables is similar to the construction of risk premia in a number of asset pricing papers by Fama & French (1993) and Carhart (1997). They introduced four risk premia now widely used in asset pricing. The market risk premium (MRP) measures the excess return of the market portfolio over the risk-free rate, the small-minus-big (SMB) or size premium the return of small stocks over big stocks, the high-minus-low (HML) or value premium the return of stocks with a high book-to-market ratio in excess of stocks with a low bookto-market ratio, and the up-minus-down (UMD) or momentum premium the return of stocks with of a high momentum in excess of stocks with a low
momentum. In a similar manner, we construct four variables which measure the dierences in implied volatilities for stocks with dierent characteristics. We refer to these variables as the market-risk spread (MRPS), the small-minus-big spread (SMBS), the high-minus-low spread (HMLS), and the up-minus-down spread (UMDS). The market-risk spread (MRPS) measures the dierence in implied volatilities between high beta stocks and low beta stocks, i.e.,
B 1 X (beta) M RP St = IVB;b;t B b=1 (beta)
"
"
(1)
where IVa;b;t is the implied volatility at time t for the portfolio formed on beta and historical volatility where beta is in the ath quantile out of a total of A quantiles and the historical volatility is in the bth quantile out of a total of B quantiles. Since we use in this study 5 5 portfolios, A and B are equal to 5. The small-minus-big spread (SMBS) measures the dierence in implied volatilities between stocks with a low market value and stocks with a high market value, i.e.,
B 1 X (size) IVB;b;t SM BSt = B b=1 (size)
"
"
(2)
where IVa;b;t is the implied volatility at time t for the portfolio formed on size and historical volatility, where the market value is in the ath quantile and the historical volatility is in the bth quantile. The high minus-low spread (HMLS) measures the dierence in implied volatilities between stocks with a high and low book-to-market ratio, i.e.,
B 1 X (btm) HM LSt = IVB;b;t B b=1 (btm) IVa;b;t
"
"
(3)
where is the implied volatility at time t for the portfolio formed on the book-to-market ratio and historical volatility, where the book-tomarket ratio is in the ath quantile and the historical volatility is in the bth quantile. The up minus-down spread (UMDS) measures the dierence in implied volatilities between stocks with a high momentum and stocks with a low momentum, i.e.,
"
"
(4)
where is the implied volatility at time t for the portfolio formed on momentum and historical volatility, where the momentum (one year return) is in the ath quantile and the historical volatility is in the bth quantile.
2.4
Forecasting Error
To assess the forecasting accuracy, we compute the mean relative error, M RE, as the average dierence between realized volatility, HV; and implied volatility, IV; divided by the implied volatility M REa;b
(x) (x) T IVa;b;t 1 X HVa;b;t+ = (x) T IV t=1 a;b;t
(5)
where T denotes the sample size, the time to maturity of the options (91 calendar days), and (x) denotes the grouping criterion for the portfolio (e.g., mom for momentum). The test statistic can be interpreted as follows. Values smaller than 0 mean that future realized volatility was smaller than implied volatility. Values larger than 0 mean that future realized volatility was higher than implied volatility. Beside the mean relative error, we use a number of other approaches (such as the root mean squared error, forecasting regressions, qq-plots) to assess the predictive accuracy. However, as the M RE turned out to capture a high proportion of the information contained in the forecasting error, we report only the M RE. Jorion (1995) suggests a regression-based approach the for evaluation of the performance of implied volatility. The predictive power of a volatility forecast can be estimated by regressing the realized volatility on forecast volatility: HVa;b;t+ = c0;a;b + c1;a;b IVa;b;t + "t;T
(x) (x)
(6)
where implied volatility at time t is used as the volatility forecast. If forecasts are perfect, we would expect the intercept, c0;a;b , to be 0 and the slope coe cient, c1;a;b , to be unity. Since overlapping periods introduce autocorrelations in the error terms, "t;T , we correct for this eect using the Newey & West (1987) correction with the appropriate number of lags.
3
3.1
Empirical Results
Implied Volatility
In this section, we analyze the relation between implied volatility, IVt , the historical volatility, HVt , and the rm characteristics (market value, M Vt , the market-to-book ratio, M T Bt , the cumulative return over one year, M OMt , and the beta of a stock, BET At ). We focus on sample averages over the whole whole sample period. Figure 1 shows the relation between historical volatility and the beta on implied volatility. Numerical results are displayed in Table 3. As expected, implied volatility is in general increasing in historical volatility, i.e., portfolios with higher historical volatility have also higher implied volatilities. Within each quantile, we nd generally a positive relation between beta and implied volatility. Stock options with similar historical volatility trade at a higher implied volatility if the beta is higher. Market participants perceive stocks to be riskier if the stock is a high-beta stock. The dierences are of considerable magnitude. For example, for portfolios in the rst quantile of historical volatility, the average implied volatility is 26:12% for low beta stocks (Quantile 1) and 41:88% for high beta stocks (Quantile 5). However, for portfolios with higher historical volatility, we nd a non-monotonic relation between size and implied volatility. Stocks with a very low beta (Q1) show a higher implied volatility than stocks with a beta in the center (e.g., in Q2 and Q3) for high volatility portfolios. Figure 2 shows the relation between historical volatility and the market value on implied volatility. We nd strong evidence for a size eect in implied volatility. Within each quantile of historical volatility, large stocks (market value in Q5) have a substantially lower implied volatility than small stocks (Q1) or medium-size (Q2-Q4) stocks after controlling for historical volatility. As shown in Table 3, the dierence in implied volatility for large and small stocks is 6.77% (34.95%-28.19%) for stocks in the lowest quantile of historical volatility (Q1) and 21.67% (84.78%- 62.91%) for stocks in the highest quantile of historical volatility (Q5). Figure 3 shows the relation between historical volatility and the marketto-book-ratio on implied volatility. In contrast to the clearly observable patterns for beta and market value on implied volatility, the market-tobook ratio has a dierent impact on implied volatility. For the portfolio in the quantile 2 to 5 formed on MTB, the implied volatility is very similar after controlling for historical volatility. However, stocks with a very low market-to-book ratio (Q1) have substantially higher implied volatility than
all other stocks. As shown in Table 3, the dierence in implied volatility for stocks with a very low (Q1) and low (Q2) market-to-book ratio is 3.51% (32.71%-29.20%) for stocks in the lowest quantile of historical volatility (Q1) and 4.27% (75.50%-71.21%) for stocks in the highest quantile of historical volatility (Q5). Figure 4 shows the relation between historical volatility and the momentum on implied volatility. Similar to the ndings for the market-to-book ratio, we nd a higher implied volatility for stocks with a low momentum (Q1), i.e., stocks with an annual return in the lowest quantile, than for stocks with a higher momentum. For stocks in the rst quantile of historical volatility (Q1), the dierence of implied volatility between stocks with a momentum in Q1 and Q2 is 8.70% (40.00%-31.34%). For stocks with a high historical volatility (Q5), this dierence is 8.79% (79.63%-70.84%) and therefore of similar magnitude. For the remaining portfolios, we nd a smaller eect between momentum stocks in Q4 and Q5. Options on stocks with a very high momentum (Q5) trade at slightly higher implied volatility than options on stocks with a weaker momentum (Q4) after controlling for historical volatility. Overall, the ndings can be summarized as follows. We nd that the equity options of high beta stocks, small stocks, low market-to-book stocks and non-momentum stocks trade at a higher implied volatility over the whole sample period. The ndings are robust for dierent levels of historical volatility.
3.2
In this section, we analyze the time series behavior of dierences in implied volatilities for various portfolios to check for robustness and time-variation of previous ndings. For this purpose, we use the spreads introduced in section 2.3. Figure 5 shows the time series of the market-risk spread (MRPS), the small-minus-big spread (SMBS), the high-minus-low spread (HMLS), and the up-minus-down spread (UMDS). MRPS measures the dierence in implied volatility for high and low beta stocks, SMBS between small and big stocks, HMLS between stocks with a high and low book-to-market ratio, and UMDS between stocks with a high and low momentum. The time series average values are displayed in the title of each plot. The MRPS is 11.21% on average, i.e., high-beta stocks have on average a 11.21% higher implied volatility than low-beta stocks. The spread is positive at almost all times. However, there is considerable uctuation over time.
Between 2000 and 2002, the spreads is in general between 20% and 45%, in 2004, between 15% and 20%. In the remaining years, MRPS is between 0 and 10%. The SMBS on average is 16.12%, i.e., small stocks have on average a 16.12% higher implied volatility than large stocks. Similar to MRPS, it is also positive most of the time, but in contrast to MRPS, it shows a higher stability. The HMLS is -3.05% on average, i.e., stocks with a high book-to-market ratio trade on average at a 3.05% lower volatility than stocks with a low book-to-market ratio. This nding is surprising. In empirical asset pricing a number of authors document (e.g., Fama & French (1993), Fama & French (1998)) that stocks with a high book-to-market ratio have higher average returns than predicted by the CAPM. Our analysis shows that these stocks are perceived to be less risky than stocks with a low book-to-market ratio. In other words, option markets perceive stocks with high valuation ratios (low book-to-market) to be riskier than stocks with low valuation ratios. The UMDS is on average -9.70%, i.e., stocks with a high positive momentum have on average a -9.70% lower volatility than stocks with a high negative momentum. In other words, stocks with a strong negative return in the previous year trade at substantially higher implied volatility. Except for 2003, the UMDS is negative in all years.
3.3
In this section, we analyze the forecasting power of implied volatility. Table 4 shows the numerical results. Figure 6 shows the mean relative error (MRE) for portfolios formed on historical volatility and beta. Negative values mean that realized volatility was lower than predicted and positive values mean than realized volatility was higher than expected. In the sample period, realized volatility was lower than expected for most portfolios. In particular, volatility was overestimated for stocks with a low beta (Q1 and Q2) and high historical volatility (Q4 and Q5). For stocks with a beta in the highest quantile, realized volatility was higher than implied volatility. Figure 7 shows the MRE for portfolios formed on historical volatility and the market value. We nd that implied volatility is a poor predictor for realized volatility for small caps. For almost all portfolios with a market value in Quantile 1 and Quantile 2, implied volatility overestimated future volatility. In contrast, for all medium and large caps (Q3 to Q5), historical volatility was higher than implied volatility.
10
In Figure 8, we show the MRE for portfolios formed on the marketto-book ratio and historical volatility. The analysis shows that realized volatility is higher than suggested by implied volatility for all portfolios in quantiles 4 and 5 for the market-to-book ratio. For all portfolios with a high valuation, realized volatility is higher than implied volatility. In contrast, for stocks with a low market-to-book ratio and high historical volatility, realized volatility is substantially lower than expected. Figure 9 illustrates the relation between the mean relative error, momentum, and historical volatility. For the momentum portfolios, we nd that implied volatility overestimates realized volatility in general. Only for stocks with a very strong momentum (Q5) and with a very low momentum (Q1), we nd that realized volatility is higher than expected based on implied volatility. Table 5 and 6 show the estimated intercepts and slope coe cients of predictive regressions of realized volatility on implied volatility. For the intercept, we can reject the null hypothesis of an unbiased forecast (c0 = 0) on an 95% level only in 4 out of 100 cases. For the slope coe cient, the results suggest that implied volatility seems to be a biased forecast for a few portfolios. We can reject the null hypothesis of a correct forecast (c1 = 1) in 13 out of 100 cases. In particular, we can reject the null hypothesis for high-beta stocks with high volatility, for stocks with a low book-to-market ratio, and high-momentum stocks with high volatility. On rst sight the few rejections of the null in Table 5 and 6 appear to stand in opposition to the results in Table 4 and Figures 6- 9.1 A positive intercept and a larger than 1 slope coe cient suggests that implied volatility underestimates realized volatility which is con with the negative signs in ict the corresponding buckets in Table 4. However all the statistically positive intercept coe cients in Table 5 are accompanied by a less than 1 slope coefcient in Table 6. Similarly the higher than one slope coe cients in Table 6 are accompanied by negative intercepts in Table 5. Therefore, underpricing need not be the case on an average percentage basis. Indeed the opposite is conrmed in Table 4. Overall, we nd considerable dierences in the predictive power of implied volatility for future realized volatility in the sample period. Implied volatility overestimates realized volatility for low-beta stocks, small stocks, stocks with low market-to-book ratios, and stock with no momentum (in both directions). Implied volatility underestimates realized volatility for high-beta stocks, for large stocks, stocks with high market-to-book ratios
1
11
and stocks with a strong positive or negative momentum. However, we cannot reject the null hypothesis that implied volatility predicts realized volatility accurately in the cross-section except for very few cases.
3.4
Discussion of Results
As highlighted in the literature review, only very few studies have analyzed the behavior of implied volatility in the cross-section of equity options. Additionally, the setting of this study is dierent from other studies. Therefore, only a partial comparison of the ndings is possible. For example, Bakshi & Kapadia (2003b) examine the relation between realized and implied volatility for 25 equity options. They nd that Black-Scholes implied volatility is on average higher than the realized volatility, or - in other words - that there is a negative volatility risk premium. They argue that buyers of options may be willing to pay a premium in volatility because a long position in volatility helps hedge market risk due to an observed negative relation between market returns and market volatility. Adapting the argument by Bakshi & Kapadia (2003b) to our study, the results would imply that stocks for which implied volatility overestimates realized volatility (low beta stocks, small caps, low market-to-book, and stocks without a stock positive or negative momentum) should provide a better hedge against changes in market volatility. This issue has recently been discussed by a number of authors. Ang, Hodrick, Xing & Zhang (2006) examine the pricing of aggregate volatility risk in the cross-section of stock returns and nd that - consistent with expectations - that stocks with high sensitivities to innovations in aggregate volatility have low average returns. They nd that size, book-to-market, and momentum eects can account for neither the low average returns earned by stocks with high exposure to systematic volatility risk nor for the low average returns of stocks with high idiosyncratic volatility. They argue that this is a puzzle requiring further investigation. Basically, our results go into the same direction, except that we use option price data.
Conclusion
In this paper, we analyze the relation between implied volatility, realized volatility and stock characteristics in the cross-section of equity options. Using a large dataset of US equity options, we form portfolios based on historical volatility and stock characteristics (such as beta, size, market-tobook, and momentum). We nd a number of patterns that have not been documented up to now. 12
We nd that high-beta stocks, small stocks, stocks with a low marketto-book ratio, and non-momentum stocks trade at higher implied volatilities after controlling for historical volatility. We nd evidence that implied volatility overestimates realized volatility for low beta stocks, small caps, low market-to-book, and stocks without a stock positive or negative momentum. Implied volatility underestimates realized volatility for stocks with contrarian characteristics. However, we cannot reject the null hypothesis that implied volatility is an unbiased predictor of realized volatility in the cross section.
13
References
Ang, A., Hodrick, R. J., Xing, Y. & Zhang, X. (2006), The cross-section of volatility and expected returns The Journal of Finance 61, 259. , Bakshi, G. & Kapadia, N. (2003a), Delta-hedged gains and the negative market volatility risk premium Review of Financial Studies 16, 527 , 566. Bakshi, G. & Kapadia, N. (2003b), Volatility risk premiums embedded in individual equity options: Some new insights Journal of Derivatives , 11, 45 54. Canina, L. & Figlewski, S. (1993), The informational content of implied volatility Review of Financial Studies 6, 659 , 681. Carhart, M. (1997), persistence in mutual fund performance Journal On , of Finance 52, 57 82. Christensen, B. J. & Prabhala, N. R. (1998), The relation between implied and realized volatility Journal of Financial Economics 50, 125 , 150. Day, T. E. & Lewis, C. M. (1988), The behavior of the volatility implicit in the prices of stock index options Journal of Financial Economics , 22, 103 122. Day, T. E. & Lewis, C. M. (1992), Stock market volatility and the information content of stock index options Journal of Econometrics 52, 267 , 287. Fama, E. F. & French, K. R. (1996), Multifactor explanations of asset pricing anomalies Journal of Finance 51, 55 , 84. Fama, E. & French, K. (1993), Common risk factors in the returns on stocks and bonds Journal of Financial Economics 33(1), 3. , Fama, E. & French, K. (1998), Value vs. growth: The international evidence Journal of Finance 52, 1975 , 1999. Harvey, C. R. & Whaley, R. E. (1992), Market volatility prediction and the e ciency of the s&p 100 index option market Journal of Financial , Economics 31, 43 73. Jorion, P. (1995), Predicting volatility in the foreign exchange market , Journal of Finance 50, 507 528. 14
Lamoureux, C. G. & Lastrapes, W. D. (1993), Forecasting stock-return variance: Toward an understanding of stochastic implied volatilities , Review of Financial Studies 6, 293 326. Latan, H. A. & Rendleman, R. J. (1976), Standard deviations of stock price ratios in implied option prices Journal of Finance 76, 369 , 381. Newey, W. K. & West, K. D. (1987), simple, positive semi-denite, A heteroskedasticity and autocorrelation consistent covariance matrix , Econometrica 55, 703 708. OptionMetrics (2005), DB: File and data reference manual, version 2.5 Ivy . Poon, S.-H. & Granger, C. (2005), Practical issues in forecasting volatility , Financial Analyst Journal 61, 45 56. Poterba, J. M. & Summers, L. H. (1986), The persistence of volatility and stock market uctuations American Economic Review 76, 1142 , 1151. Sheikh, A. M. (1989), Stock splits, volatility increases, and implied volatilities Journal of Finance 44, 1361 , 1372.
15
Table 1: Breakpoints
The table shows the breakpoints for portfolio construction for each factor and each year. The table is interpreted as follows: A value of 0.48 for quantile 1 to quantile 2 (Q1 to Q2) and of 0.72 for Q2 to Q3 for beta in 1997 means that all stocks with a beta lower than 0.48 were assigned to portfolio 1, all stocks with a beta between 0.48 and 0.72 to portfolio 2, and so on. Market values are in million USD. The momentum breakpoint is the average daily continuously compounded return.
Beta
16
Market Value
Market-to-Book
Momentum
Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4
to to to to to to to to to to to to to to to to
Q2 Q3 Q4 Q5 Q2 Q3 Q4 Q5 Q2 Q3 Q4 Q5 Q2 Q3 Q4 Q5
1997 0.48 0.72 0.98 1.42 142.32 351.76 781.73 2341.17 1.56 2.20 3.17 5.17 -0.17 0.07 0.23 0.42
1998 0.43 0.65 0.86 1.15 146.79 415.86 934.76 2798.52 1.64 2.38 3.33 5.24 -0.21 0.09 0.29 0.48
1999 0.59 0.82 1.07 1.37 123.92 368.49 901.92 2829.17 1.17 1.93 2.97 5.41 -0.51 -0.17 0.07 0.34
2000 0.25 0.42 0.61 0.94 169.90 473.57 1169.78 3518.05 1.14 1.94 3.49 7.72 -0.35 -0.06 0.20 0.69
Year 2001 0.32 0.55 0.86 1.61 109.53 377.38 1001.83 3280.36 0.89 1.64 2.59 4.68 -1.00 -0.30 0.09 0.39 2002 0.44 0.73 1.04 1.64 130.14 416.97 1029.20 3178.77 1.07 1.74 2.57 4.23 -0.45 -0.11 0.10 0.38
2003 0.54 0.78 1.00 1.32 90.15 308.88 838.13 2664.48 0.84 1.36 1.95 3.16 -0.87 -0.35 -0.11 0.10
2004 0.57 0.84 1.07 1.41 212.91 564.57 1365.89 3950.69 1.42 2.04 2.89 4.67 0.16 0.30 0.47 0.76
2005 0.81 1.08 1.41 1.83 262.55 691.77 1596.23 4793.30 1.52 2.16 3.02 4.57 -0.15 0.07 0.20 0.37
Historical Volatility Q1 Q2 Q3 Q4 Q5 Beta Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 113 127 100 42 13 5 18 48 118 209 50 101 103 76 59 30 86 129 106 40 64 103 110 81 33 7 26 68 123 173 52 88 96 88 68 43 91 104 94 58 49 73 95 105 63 21 63 95 105 109 70 80 79 83 76 68 90 80 81 66 35 49 63 102 134 47 94 102 84 64 71 70 76 85 84 98 84 64 63 75 28 34 45 84 181 81 120 87 62 35 78 62 62 80 98 141 66 42 47 77
Market Value
Market-to-Book
Momentum
17
Market Value
Market-to-Book
Momentum
18
Beta Q1
Market Value
Market-to-Book
Momentum
Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5
19
20
21
22
Figure 2: Relation between historical volatility and market value on implied volatility
The Figure shows the average implied volatilities over the sample period (daily data, 01/1997 - 12/2005) for portfolios formed annually based on the market value of equity (as provided by Compustat) and on the historical volatility (91 days) at the end of each year.
23
Figure 3: Relation between historical volatility and market-tobook ratio on implied volatility
The Figure shows the average implied volatilities over the sample period (daily data, 01/1997 - 12/2005) for portfolios formed annually based on the market-tobook ratio (as provided by Compustat) and on the historical volatility (91 days) at the end of each year.
24
Figure 4: Relation between historical volatility and the momentum on implied volatility
The Figure shows the average implied volatilities over the sample period (daily data, 01/1997 - 12/2005) for portfolios formed annually based on the total return (as provided by OptionMetrics) and on the historical volatility (91 days) at the end of each year.
25
26
Figure 6: Relation between historical volatility and beta on the prediction error
The Figure shows the mean relative error, i.e., the dierence of realized and implied volatility relative to implied volatility, over the sample period (daily data, 01/1997 12/2005) for portfolios formed annually based on the beta (OLS regression of returns on CRSP market return based on daily data over one year) and on the historical volatility (91 days) at the end of each year. Negative values mean that implied volatility overestimated realized volatility on average. Positive values correspond to an underestimation of realized volatility.
27
Figure 7: Relation between historical volatility and market value on the prediction error
The Figure shows the mean relative error, i.e., the dierence of realized and implied volatility relative to implied volatility, over the sample period (daily data, 01/1997 - 12/2005) for portfolios formed annually based on the market value of equity (as provided by Compustat) and on the historical volatility (91 days) at the end of each year. Negative values mean that implied volatility overestimated realized volatility on average. Positive values correspond to an underestimation of realized volatility.
28
Figure 8: Relation between historical volatility and market-tobook-ratio on the prediction error
The Figure shows the mean relative error, i.e., the dierence of realized and implied volatility relative to implied volatility, over the sample period (daily data, 01/1997 - 12/2005) for portfolios formed annually based on the market-to-book ratio (as provided by Compustat) and on the historical volatility (91 days) at the end of each year. Negative values mean that implied volatility overestimated realized volatility on average. Positive values correspond to an underestimation of realized volatility.
29
Figure 9: Relation between historical volatility and the momentum on the prediction error
The Figure shows the mean relative error, i.e., the dierence of realized and implied volatility relative to implied volatility, over the sample period (daily data, 01/1997 - 12/2005) for portfolios formed annually based on the total return (as provided by OptionMetrics) and on the historical volatility (91 days) at the end of each year. Negative values mean that implied volatility overestimated realized volatility on average. Positive values correspond to an underestimation of realized volatility.
30