Are VIX futures prices predictable?

An empirical investigation*

Eirini Konstantinidia and George Skiadopoulosb **
a b

Department of Banking and Financial Management, University of Piraeus, Greece

Department of Banking and Financial Management, University of Piraeus, Greece, and Financial Options Research Centre, Warwick Business School, University of Warwick, UK

Abstract
This paper investigates whether volatility futures prices per se can be forecasted by studying the fast growing VIX futures market. To this end, alternative model specifications are employed. Point and interval out-of sample forecasts are constructed and evaluated under various statistical metrics. Next, the economic significance of the obtained forecasts is also assessed by performing trading strategies. Only weak evidence of statistically predictable patterns in the evolution of volatility futures prices is found. No trading strategy yields economically significant profits. Hence, the hypothesis that the VIX volatility futures market is informationally efficient cannot be rejected.

JEL Classification: C53, G10, G13, G14. Keywords: Interval forecasts, Market efficiency, Predictability, VIX, Volatility futures.

We would like to thank Ales Cerny, Alexandros Kostakis, Joëlle Miffre, Allan Timmermann and Halbert White for useful discussions and comments. Any remaining errors are our responsibility alone. ** Corresponding author. Tel: +30-210-4142363; fax: +30-210-4142341. E-mail addresses: ekonst@webmail.unipi.gr (E. Konstantinidi), gskiado@unipi.gr (G. Skiadopoulos).

1

1.

Introduction

Volatility derivatives have attracted much attention over the past years since they enable trading and hedging against changes in volatility. Brenner and Galai (1989, 1993) first suggested derivatives written on some measure of volatility that would serve as the underlying asset. Since then, a number of volatility derivatives have been trading in the overthe-counter market. In March 26, 2004, volatility futures on the implied volatility index VIX were introduced by the Chicago Board Options Exchange (CBOE)1. Volatility futures on a number of other implied volatility indices have been also introduced since then. The liquidity of volatility futures markets is steadily growing, with the VIX futures market being the most liquid one2. This paper focuses on the VIX futures market and addresses for the first time the question whether VIX futures prices per se can be predicted3. Answering the question whether volatility futures prices can be predicted is of importance to both academics and practitioners. This is because it contributes to understanding whether volatility futures markets are efficient and helps market participants to develop profitable volatility trading strategies and set successful hedging schemes. There is already some extensive literature that has investigated whether the prices of stock index, interest rate, currency, and commodity futures can be forecasted. The significance of the results has been evaluated under either a statistical or economic (trading profits) metric. A number of studies have documented a statistically predictable pattern in futures returns. In particular, Bessembinder and Chan (1992) found that the monthly nearest maturity commodity and currency futures returns can be forecasted within sample in a statistical sense. They concluded that this predictability could be attributed to an asset pricing model with time-varying risk-premia. Similar findings were documented by Miffre (2001a) for the FTSE 100 futures and by Miffre (2001b) for commodity and financial futures. On the other hand, the empirical evidence on the predictability in futures markets

2

under an economic metric is mixed. For instance, Hartzmark (1987) found that in aggregate, speculators do not earn significant profits in commodity and interest rate futures markets; daily data of all contract maturities were employed. On the other hand, Yoo and Maddala (1991) studied commodity and currency futures and found that speculators tend to be profitable; daily data for a number of futures maturities were considered. Similar findings were reported by Taylor (1992), Kho (1996), Wang (2004) and Kearns and Manners (2004). In particular, all these studies found that economically significant profits can be obtained by employing various trading rules in currency futures markets; daily data were used by Taylor (1992), and weekly by Kho (1996), Wang (2004) and Kearns and Manners (2004). A number of futures maturities were examined by Taylor (1992) and Kearns and Manners (2004), while Kho (1996) and Wang (2004) focused on the shortest maturity series. Significant profits were also reported in Hartzmark (1991) and Miffre (2002) who examined the commodity and financial futures markets; the latter study focused only on the shortest maturity contracts. Regarding the source of the identified trading profits, Taylor (1992) and Kearns and Manners (2004) attributed them to the inefficiency of the currency futures market. On the other hand, Yoo and Maddala (1991), Kho (1996), Wang (2004) and Miffre (2002) found that the reported profits were not abnormal and Hartzmark (1991) found that profitability is determined by luck rather than superior forecast ability; hence, the considered markets were efficient á la Jensen (1978). In contrast to the number of papers devoted to the topic of predictability in the previously mentioned futures markets, the research on whether there exist predictable patterns in the evolution of volatility futures prices is still at its infancy. The literature on volatility futures has primarily focused on developing pricing models (see e.g. Grünbichler & Longstaff, 1996; Zhang & Zhu, 2006; Dotsis et al., 2007; Brenner et al., 2008; Lin, 2008) and assessing their hedging performance (see e.g. Jiang & Oomen, 2001). On the other hand,

3

To check the robustness of our results. We test the statistical significance of the obtained forecasts by a number of tests and criteria. This is because interval forecasts have been found to be useful for volatility trading purposes. The authors developed trading strategies with VIX and VXD volatility futures based on point and interval forecasts that were formed for the corresponding underling implied volatility indices. this has been done indirectly and only under a financial metric. Section 2 describes the data set. Section 4 discusses the results concerning the in-sample performance of the models under consideration. both point and interval out-of-sample forecasts are considered. To this end. the out-ofsample predictive performance of the various models is evaluated in statistical and economic 4 . In contrast to Konstantinidi et al. the analysis is performed across various maturity futures series and by employing a number of alternative model specifications. Poon and Pope (2000) found that profitable volatility spread trades can be developed in the S&P 100 and S&P 500 index option markets by constructing certain intervals. (2008).to the best of our knowledge. Konstantinidi et al. (2008) is the only related study that has explored the issue of predictability of volatility futures prices. we investigate the predictability of the VIX volatility futures prices per se without resorting to the underlying implied volatility index4. The remainder of this paper is structured as follows. This study extends the literature on whether the evolution of volatility futures prices can be forecasted. In addition. Next. They found that the obtained Sharpe ratios were not statistically different from zero and hence the volatility futures markets are efficient. However. The latter is necessary since the question of predictability is tested inevitably jointly with the assumed forecasting model. Section 3 presents the forecasting models to be used. their economic significance is investigated by means of trading strategies. This is the ultimate test to conclude whether the recently inaugurated volatility futures market is efficient.

respectively. calculated as the difference between the prices of the ten-year U. 2005 to March 13. The Data Set Daily settlement prices of CBOE VIX volatility futures and a set of economic variables are used. 5 .S. 2008. The underlying asset of these contracts is VIX. The contract size is $1. 2004 by the CBOE. the shortest. namely. second shortest. The subset from March 18.terms in Sections 5 and 6. 2004 to March 13. These are exchange-traded futures contracts on volatility and may be used to trade and hedge volatility. the Libor one-month continuously compounded interbank interest rate. The last section concludes. and third shortest maturity series. government bond and the one-month continuously compounded interbank rate. 2008 is used for the out-of-sample evaluation. settlement prices corresponding to a trading volume less than five contracts are excluded.000 times the VIX5. The VIX futures prices are obtained from the CBOE website. The data set of economic variables consists of the return on the S&P 500 stock index. On any day. The final settlement date is the Wednesday that is thirty days prior to the third Friday of the calendar month immediately following the month in which the contract expires. The sample period under consideration is from March 26. and the slope of the yield curve. By ranking the data based on their time to expiration. 2. For the same reason. To minimize the impact of noisy data. the CBOE Futures Exchange (CFE) may list for trading up to six near-term serial months and five months on the February quarterly cycle for the VIX futures contract. three time series of futures prices are constructed. These data are obtained from Datastream. The VIX futures contracts are cash settled. VIX futures were listed in March 26. we roll over to the next maturity contract five days before the contract expires.

2002. 2008).3). 2008. 2001b. Rt the log-return on the S&P 500 stock index between time t-1 and t. 2004. with prices being higher for longer maturities (see also Brenner et al. this is not the case when they are measured in first differences.. This model specification tests the semi-strong form efficiency (Fama. Based on the BIC criterion.T (1) where ΔFt . and 6 . 1990) of the volatility futures market (see also Bessembinder & Chan.T + a2 ΔFt − 2. The average volume decreases for longer maturities. but stationary in the first differences. [Insert Figure 1 about here] [Insert Table I about here] 3. respectively). The term structure of futures prices appears to be upward sloping. 2004 to March 17. 1970. Miffre.Figure 1 shows the evolution of the three maturity VIX futures series and the VIX index over the period from March 26. Kearns & Manners. All variables measured in levels are positively (first order) autocorrelated.1 The Forecasting Models Economic Variables Model We use a set of lagged economic variables to forecast the evolution of futures prices. Table I shows the summary statistics for the three series of futures’ prices and the economic variables in levels and first differences (Panel A and B. 2001a. c a constant.2.. The ADF test indicates that most of the variables are non-stationary in the levels. for applications of similar predictive specifications to futures markets). 2005. 3.T denotes the daily changes in the futures price between time t-1 and t for a given maturity T (T=1. Konstantinidi et al. it the one-month continuously compounded Libor rate in log-differences.T = c + a1ΔFt −1. the following regression is estimated: ΔFt . 1992.T + a3 Rt −1 + a4 Rt − 2 + a5it −1 + a6it − 2 + a7 Δyst −1 + a8 Δyst − 2 + ε t .

1970.T Furthermore.2 Univariate Autoregressive. the following VAR(1) model is estimated: ΔFt = C + Φ1ΔFt −1 + ε t (4) (3) (2) where ΔFt is the ( 3 ×1) vector of changes in the three futures prices series that are assumed to be jointly determined. Φ1 is a ( 3 × 3) matrix of coefficients and ε t is a ( 3 ×1) vector of residuals. This model specification has been employed for forecasting purposes in various settings.T = c + ϕ1ΔFt −1..g.g. The number of lags employed are chosen on the basis of the BIC criterion and to avoid over-fitting the data. 3.. The following AR(2) model is estimated: ΔFt . such as to predict macroeconomic variables (see e.T + ε t .1) model: ΔFt . ARIMA and VAR models are employed to investigate the extent to which past volatility futures prices can be exploited for predictive purposes and to examine whether there are spillovers between the three futures series.Δyst the changes of the slope of the yield curve. 3.3 Principal Components Model Principal components (PCs) extracted from Principal Component Analysis (PCA) are used as predictors in a regression setting. The employed variables have been shown to have forecast power in equity markets (see e.T + ϕ2 ΔFt − 2. ARIMA and VAR Models Univariate autoregressive.1. Goyal & Welch. Stock & 7 .T = c + ϕ1ΔFt −1. C is a ( 3 ×1) vector of constants.T + θ1ε t −1 + ε t . 1991). These model specifications set up tests of weak form market efficiency (Fama. 2007) and hence they may also have predictive power in futures markets.T We estimate also an ARIMA(1.

T = c1 + ∑ r1 j PC1t − j + ∑ r2 j PC 2t − j + ∑ r3 j PC 3t − j + ∑ r4 j PC 4t − j + ε t .T j =1 j =1 j =1 j =1 2 2 2 2 (5) where rkj denotes the coefficient of the i-th PC lagged j times (k=1.. Focusing on the correlation loadings that show the impact of each PC to the daily changes of each one of the three futures series. among others). we can see that almost all PCs affect the changes of futures prices by a similar magnitude across maturities. 4 and j = 1.Watson. The only exception is PC2 that seems to affect the steepness of the term structure of VIX futures. In the first step.. Forecasting is performed in two steps. the term structure of petroleum futures (Chantziara & Skiadopoulos.2). Note that this model specification has implications for the semi-strong form of efficiency of volatility futures markets. 3. 2002b. This implies that the respective shocks move the term structure of futures prices in a parallel way. since volatility futures prices and economic variables are both employed to extract the PCs. 2005. 2008). In the second step. we apply PCA to the block of daily changes in the three futures series and the economic variables. The first four principal components are retained. the retained principal components are used as predictors in a regression setting and the following model specification is estimated: ΔFt . they explain 94% of the total variability of the system of variables employed. Stock and Watson (2002a) showed that the PCs are consistent estimators of the true latent factors and that the forecasts based on PCs converge to those that would be obtained if the latent factors were known. Figure 2 shows the correlation loadings of the retained principal components. 2. 2008) and implied volatility indices (Konstantinidi et al. [Insert Figure 2 about here] 8 . Artis et al.

Bates & Granger. To fix ideas. the weights are chosen so as to minimize the mean squared forecast error (see Granger & Ramanathan. we also consider combination forecasts. ˆ ˆ c is a constant.3. Clemen. 1989.T + ε t i =1 5 (7) where ΔFt .1. Combination forecasts aggregate the information used by the individual forecasting models. Two alternative linear combination forecasts are considered.T = Ft|it −1.1) model).T denotes the |t equally weighted combination forecast of the futures price constructed at time t-1 for t for a given maturity T. 1969. and FtU−1.4 Combination Forecasts Apart from model based forecasts. 2 (AR(2) model). the weights are obtained by estimating the following OLS regression recursively: ˆ ΔFt . This is a simple average of all model based forecasts and is hereinafter referred to as “unweighted combination forecast”. 5 (PCA regression model)]. 4 (VAR model).g. there is evidence that simple combinations frequently outperform more sophisticated ones (see e. 1984).T is the realized futures price change between time t-1 and t for a given maturity T. Clemen..T = c + ∑ ai ΔFt|it −1.T − Ft −1.T is the forecasted futures price constructed at time t-1 for t for a given maturity T by using the i-th model specification [i = 1 (economic variables model). First. an equally weighted combination forecast is employed: 1 ˆ ˆ FtU−1.T = ∑ Ft|it −1.T |t 5 i =1 5 (6) ˆ where Ft|it −1.g. Second. Then: 9 .T is the forecasted futures price change between time t-1 and t for a given maturity T and ε t is the error term. They have been found to be more accurate than individual forecasts (see e. 1989). a weighted average of the individual forecasts is used. ΔFt|it −1. 3 ˆ (ARIMA(1. for a review)..

ˆ ˆ FtW1|t . To this end. Finally. To start constructing the weighted combination forecasts recursively.1-3. III and IV show the in–sample performance of the economic variables.1. 2004 to March 17. the insample data (from March 26. The estimated coefficients. (3). the t-statistics within parentheses and the adjusted R2 are reported for each one of the implied volatility indices. Then. First.T + i =1 5 (8) ˆ where FtW1|t . the individual forecasts over the “pseudo” out-of-sample period are used to estimate regression (7) and the first out-of-sample weighted combination forecast (corresponding to March 18.1)/VAR and PCA regression models. one needs an “initial” time series of individual forecasts to estimate regression (7). AR(1)/ARIMA(1. 2005) are divided into a “genuine” insample period (from March 26. One and two asterisks indicate that the estimated parameters are statistically significant at 1% and 5% level.T denotes the weighted combination forecast of the futures price constructed at t + for t+1 for a given maturity T. In-Sample Evidence Tables II. respectively. 2004) and a “pseudo” out-of-sample period (from September 29. (2). 2005). 2004 to September 24. To form the remaining out-of-sample combination forecasts equation (7) is estimated recursively by adding each individual forecast to the sample as it becomes available. the “genuine” in-sample data are used to estimate the model specifications described in Section 3. forecasts are formed recursively over the “pseudo” out-of-sample period by adding each observation of the “pseudo” out-of-sample data set to the “genuine” insample data set as it becomes available.3 [equations (1). respectively. 4. 2005) is constructed as the fitted value of regression (7).T + c + ∑ ai ΔFt i+1|t .T = Ft . 2004 to March 17. respectively. 10 . and (5)]. (4).

1%).8%). Out-Of-Sample Evidence: Statistical Significance Point and interval forecasts are used to assess the out-of-sample performance of the models described in Section 3. 2001b. 1992. the models are re-estimated recursively by adding each observation to the in11 . Panel A and Panel B. Panel C]. 2008). The out-of-sample period is from March 18. 2008.. In the case of AR(2) and VAR models [Table III. To sum up. the in-sample goodness-of-fit depends on the model specification and the maturity of the futures series under consideration. in the case of the PCA regression model [Table IV].2% across the three futures series. the strongest pattern appears in the third shortest series (adjusted R2 equals 3. we can see that the largest values of the adjusted R2 are obtained for the second shortest series (1. 2005).In the case of the economic variables model [Table II]. Next. 2005 to March 13. To construct the remaining out-of-sample point forecasts. the out-of-sample performance is assessed so as to provide a firm answer to the question whether volatility futures prices can be forecasted. Finally. 2005) and the first out-of-sample point forecast is obtained (corresponding to March 18. reveals a strong predictable pattern in the case of the shortest futures series (adjusted R2 equals 5. the models are initially estimated over the in-sample period (from March 26. we can see that the adjusted R2 ranges from 1% to 3. respectively]. 2004 to March 17. Miffre. This is similar to the values of R2 documented by the previous related literature in various futures markets (see Bessembinder & Chan. respectively). Konstantinidi et al. 2001a. [Insert Table II about here] [Insert Table III about here] [Insert Table IV about here] 5. To form the point forecasts.7% and 3%. The application of the ARIMA model [Table III. 2002.

2 (AR(2) model). the null hypothesis is that no model outperforms the random walk. three alternative metrics are employed.. Moreover.000 simulation runs on each time step (i. respectively. 1997) and a ratio test for the RMSE/MAE and MCP metrics.T n t =1 denote the forecasted futures price based on the i-th model [i = 1 (economic variables model).sample data set as it becomes available.e. Let ˆ {F } i t |t −1. To this end.1) model). To fix ideas. day). we perform pairwise comparisons based on the modified Diebold-Mariano test (see Diebold & Mariano. averaged over the number of observations. The forecasts are compared to those obtained from the random walk that is used as the benchmark model. The first metric is the root mean squared prediction error (RMSE) calculated as the square root of the average squared deviations of the actual volatility futures prices from the model based forecast.T n t =1 and ˆ {F } RW t |t −1. 1995. The null hypothesis is that the model under consideration and the random walk perform equally well. The second metric is the mean absolute prediction error (MAE) calculated as the average of the absolute differences between the actual volatility futures price and the model based forecast. The interval forecasts are formed by generating 10.. Harvey et al. averaged over the number of observations. 5 (PCA 12 . 4 (ARIMA(1.1 Point Forecasts: Statistical Testing To assess the statistical significance of the obtained out-of-sample point forecasts. In this case. 3 (VAR model).1. 5. we use White’s (2000) test (also termed reality check) to compare jointly all forecasts to the benchmark model under the RMSE and MAE metrics6. the two tests are described as follows. The third metric is the mean correct prediction (MCP) of the direction of volatility futures price changes calculated as the average frequency (percentage of observations) for which the predicted by the model change in the volatility futures price has the same sign as the realized change.

T ) and g ( etRW ) and the loss .T ) = 0 . n t =1 and {etRW } .T is defined for the i-th model and for a given maturity T : ˆ fi . i. the modified i Diebold-Mariano test statistic MDM T for the ith model specification and the T-maturity futures series is given by: i MDM T = n dTi var(dTi ) (9) with dTi = ∑ dti.. In the case of one-step ahead forecasts. The first alternative hypothesis is that the random walk outperforms the respective model.e. k 13 ..T ) > 0 . In the case of the modified Diebold-Mariano test.T = −dti. At any ˆ point in time t the performance measure fi . 7 (weighted combination forecast)] and the random walk.T ) < 0 7. respectively. we make accept/reject decisions by comparing the calculated test statistic to the critical values from the Student’s t distribution with (n-1) degrees of freedom..T t =1 n n and var(dTi ) = ∑(d t =1 n i t .T (10) So.T n t =1 being the respective forecast errors for the ith model specification and the T-maturity futures series.T ) ≤ 0 .t .T ) − g ( etRT ) .T = g ( eti. 6 (unweighted combination forecast).T } ..T − dTi ) 2 n −1 .t . i. the null hypothesis is H 0 : E ( dti. 1999). The second alternative hypothesis is that the model under consideration outperforms the random walk. The idea of White’s (2000) test is as follows see also Sullivan et al. with {eti. (1997).T W differential dti. the null hypothesis is H 0 : max E ( f i .regression model). Following Harvey et al. Define a loss function g ( eti. The test statistic for the observed sample is: i =1. We test this against two alternative hypotheses.e. H 2 : E ( dti.. H1 : E ( dti.

We choose B = 10.T ) } (12) where j = 1.t .t . … B..T (1 − a ) and U ti/ t −1. The idea of the test is as follows. 2.T }t =1 . For each bootstrap sample. respectively. 1998). i.. { Ft .T = ∑ fi . A sample path.T 8.e. 5. and n is the number of forecasts made.. The test can be applied for any assumed underlying stochastic process. j − fi . White’s (2000) reality check p-value is then obtained by comparing V and the obtained V j for j = 1.. the following statistic is calculated: V j = max i =1.T (1 − a )} T t =1 is constructed.T (11) ˆ where fi .T . let B generated bootstrapped ˆ samples of fi .T (1 − a ) denote the lower and upper bound of an (1-a)%-interval forecast for time t constructed at t-1 for a given maturity contract based on the i-th model. We test whether the (1-a)%-interval forecast is “efficient”.t .T n . 2. To this end. White (2000) suggests that t =1 the null hypothesis can be evaluated by applying the stationary bootstrap of Politis and ˆ Romano (1994) to the observed values of fi .T is defined: 14 . In particular. an indicator function I ti. k { n( f * i .000. since is not model dependent (Christoffersen. Lit / t −1. U ti/ t −1. k { n ( f )} i .T . of futures prices for a given maturity is observed and a n series of interval forecasts {( Lit / t −1.V = max n i =1.2 Interval Forecasts: Statistical Testing Christoffersen’s (1998) likelihood ratio test of unconditional coverage is used to evaluate the constructed interval forecasts..T (1 − a ) .. whether the percentage of times that the realized future price at time t falls outside the interval forecast for time t constructed at time t-1 is a% for a given maturity. … B...

if Ft.T ⎧0. Hence. walk model (Panel A).1) model (Panel E) and the PCA regression model (Panel F).e. Monte Carlo simulated p-values are generated to assess the statistical significance of our results. respectively. MAE and MCP obtained for point forecasts based on the random.U ti/ t −1. the power of this test may be sensitive to sample size.T (1 − a ) ⎤ ⎪ ⎣ ⎦ =⎨ i i ⎪1. a = n1 ( n0 + n1 ) is one minus the observed coverage probability. the economic variables model (Panel B).U t / t −1. respectively) are also reported. Results for the unweighted and weighted combination of point forecasts (Panel G and H. Christoffersen’s test statistic is given by a likelihood ratio test: ˆ LRunc = −2 ln[ L(a) / L(a)] ∼ X 2 (1) (14) where n0 is the number of times that the futures prices falls within the constructed interval forecast (i. 5.e.T (1 − a ) ⎤ ⎣ ⎦ ⎩ (13) Thus. if Ft. We construct 99% and 95% interval forecasts to assess the robustness of the obtained results across different levels of significance. Under the null hypothesis. I t = 0 ). ˆ ˆ ˆ L ( a ) = (1 − a ) 0 a n1 is the likelihood under the null hypothesis and L ( a ) = (1 − a ) 0 a n1 is the n n likelihood under the alternative hypothesis. the VAR model (Panel D).T ) = α is tested against the alternative Η1: E( I ti.1.T (1 − a ) . the null hypothesis of an efficient (1-a)% interval forecast Η0: E( I ti.T ) ≠ α. n1 is the number of times that the futures price falls outside the interval ˆ forecast (i. We can see that there are 15 (out of 63 possible combinations in total) combinations of futures series and predictability metrics in which the 15 . the AR(2) model (Panel C).T (1 − a ) .T ∉ ⎡ Lt / t −1.3 Point and Interval Forecasts: Results In the case of point forecasts.T ∈ ⎡ Lit / t −1. However. Table V shows the RMSE. the ARIMA(1.I i t . One and two asterisks (crosses) denote rejection of the null hypothesis in favor of the alternative H1 (H2) at significance levels 1% and 5%. I t = 1 ).

e.999 (0. the unweighted combination interval forecasts and the weighted combination interval forecasts (Panels A. respectively...9532) for the shortest. One and two asterisks denote rejection of the null hypothesis at 1% and 5% significance levels. we accept the null hypothesis in all cases. the reality check p-value for the RMSE (MAE) is 0.random walk beats one of the models (i. D. 0. there is weak evidence of a statistically predictable pattern in the evolution of the shortest futures series. E. and Christoffersen’s (1998) test statistic value obtained by the economic variables model.963 (0. Thus. results are reported for each one of the three futures series. F and G. the AR(2) model.. All of these occur under the MCP measure and most of them are observed for the shortest series (4 out of 5). This implies that even the best performing model specification under the RMSE (MAE) metric does not outperform the random walk. second shortest and third shortest series.888 (0. 16 .1. the PCA regression model.e. In the case of White’s (2000) test.1) model. This holds for both the 99% and 95% interval forecasts. On the other hand. C.998) and 0. We can see that the null hypothesis of efficient interval forecasts is rejected in all instances.916). respectively. the VAR model. one would expect the models to beat the random walk only in roughly 3 out of 63 cases (i. the ARIMA(1. B. respectively). Note that under the assumption of independence of accept/reject decisions. Table VI shows the percentage of observations that fall outside the constructed 99% and 95%-interval forecasts. 24% of the cases).e. 8%) the model under consideration outperforms the random walk. Thus. [Insert Table V about here] Regarding interval forecasts. 5% of the cases) at a 5% significance level. in 5 out of 63 cases (i.

1 Testing for Economic Significance: Measures The profitability of the trading strategies is evaluated in terms of the Sharpe Ratio (SR). This is because the statistical evidence does not always corroborate a financial criterion (see e. 6.g.[Insert Table VI about here] 6. Transaction costs have been taken into account. The continuously compounded one month Libor rate is used as the risk free rate in the calculation of both measures of performance.5$ per transaction. Leland’s (1999) alpha is employed in order to account for the presence of nonnormality in the distribution of the trading strategy’s returns.. Moreover. The trading strategies are performed despite the fact that there is no evidence of a statistically predictable pattern. respectively. 2003). It is based on the asset pricing model of Rubinstein (1976) and Breeden and Litzenberger (1978). To provide a definite answer on the issue of predictability in volatility futures markets. the economic significance of the obtained forecasts is assessed by performing trading strategies based on point and interval forecasts. none of the constructed 99% and 95%-interval forecasts were found to be efficient. Ferson et al. who assumed that the 17 . The bootstrap samples have been generated under the null hypothesis of a zero SR and Ap. the standard transaction fee in the VIX futures market is 0. Leland’s (1999) alpha (Ap) and their bootstrapped 95% confidence intervals9. Out-of-Sample Evidence: Economic Significance The previously reported results on point forecasts suggest that there is a weak evidence of a statistically predictable pattern in the evolution of the shortest futures series based on the modified Diebold-Mariano test. The trading strategies involve a single volatility futures contract.

1. B p ( = cov ( r cov rp . respectively. the following regression is estimated: rpi .t and Aip are the return on the trading strategy and Leland’s (1999) alpha.t − B ip . 3 (VAR model).1) model). 7 (weighted combination forecast)]. 6 (unweighted combination forecast).t = Aip + ε t ⎣ ⎦ (17) where rpi . We use the one month continuously compounded Libor rate and the return on the S&P 500 as proxies for the rf and rmkt .i. − (1 + rmkt ) ) is a measure of risk similar to the ) CAPM’s beta and γ = ln ⎡ E (1 + rmkt ) ⎤ − ln (1 + rf ) ⎣ ⎦ var ⎡ ln (1 + rmkt ) ⎤ ⎣ ⎦ is a measure of risk aversion. Leland’s (1999) alpha is defined as: Ap = E ( rp ) − B p ⎡ E ( rmkt ) − rf ⎤ − rf ⎣ ⎦ (16) A two step procedure is employed to calculate Leland’s (1999) alpha. − (1 + rmkt ) mkt −γ −γ . If Aip > 0 then we conclude that the trading strategy offers an expected return in excess of its equilibrium risk adjusted level. Then. 18 . Under these assumptions. First. rf is the risk-free rate of interest.d. respectively.t ⎡ rmkt . at any point in time and that markets are perfect. 4 (ARIMA(1. 5 (PCA regression model). γ and Bstr are computed for each time step. the equilibrium expected returns satisfy the following single period relationship: E ( rp ) = rf + B p ⎡ E ( rmkt ) − rf ⎤ ⎣ ⎦ (15) where rp is the return on the trading strategy.t − rf . 2 (AR(2) model). Second. rmkt is the return on the market portfolio.returns on the market portfolio are i.t ⎤ − rf . that are based on the forecasts from the i-th model [i = 1 (economic variables model).

6. then do nothing.T . and 0.09)] for the shortest series. We can see that the SR and Ap are insignificant in all cases.1.T = Ft|it −1. Results are reported for trading strategy based on point forecasts derived by the economic variables model (Panel A).1) model (Panel D). and their respective bootstrapped 95% confidence intervals (95% CI) for the three VIX futures series. 0. the VAR model (Panel C).03.11)] for the second shortest series. then the investor takes no action and maintains his/her position.T < (>) Ft|it −1. the PCA regression model (Panel E) and the weighted (Panel F) and unweighted (Panel G) combination point forecasts.3 Trading Strategy and Results Based on Interval Forecasts 19 .13)] for the third shortest series. the ARIMA(1. Ap. The rational of this trading rule is as follows: If the current futures price is higher (lower) than the forecasted futures price.06.02. then go long (short). If the current futures price is equal to the forecasted futures price. 0. This implies that all trading strategies based on point forecasts do not yield economically significant profits. the AR(2) model (Panel B).0532 [95% CI = 0. Table VII shows the annualised SR.0178 [95% CI = (-0. ˆ If Ft −1. 0. then the price is anticipated to decrease (increase) and the investor goes short (long).2 Trading Strategy and Results Based on Point Forecasts The economic significance of the constructed point forecasts is evaluated in terms of the following trading rule: ˆ If Ft −1.0421 [95% CI = (-0. [Insert Table VII about here] 6. The results are similar to those obtained for a naïve buy and hold strategy in VIX futures that yields a SR equal to 0.T . 0.

then go long (short).09)] for the shortest series. 0.02.11)] for the second shortest series.03. The rational behind this trading rule is as follows: If the futures price is closer to the lower (upper) bound of next day’s interval forecasts. the VAR model (Panel C).0421 [95% CI = (-0.06. and 0. Table VIII shows the annualised SR. the AR(2) model (Panel B). the ARIMA model (Panel D). the trading strategies based on interval forecasts do not yield significant profits. If Ft −1. the SR and Ap are insignificant in all but one cases.T = U ti|t −1.T < (>) U ti|t −1.T (1 − α ) + Lit|t −1.0178 [95% CI = (0.T (1 − α ) 2 . then do nothing. then we anticipate the index price to increase (decrease) and as a result the investor should go long (short).The economic significance of the constructed interval forecasts is evaluated in terms of the following trading rule: If Ft −1. [Insert Table VIII about here] 20 . 0. This means that overall.the SR equals 0.13)] for the third shortest series. just as was the case with the trading strategies based on point forecasts. 0.T (1 − α ) 2 . We can see that the results are similar for the strategies based on the 99% and 95%-interval forecasts.T (1 − α ) + Lit|t −1. In particular. 0.0532 [95% CI = -0. and their respective bootstrapped 95% confidence intervals (95% CI) for the three VIX futures series. Ap. the PCA regression model (Panel E) and the weighted (Panel F) and unweighted (Panel G) combination interval forecasts. The results are similar to those obtained for a naïve buy and hold strategy in VIX volatility futures. Results are reported for the trading strategy based on 99% and 95%-interval forecasts derived by the economic variables model (Panel A). In particular .

A number of alternative model specifications have been employed: the economic variables model. we found weak evidence of a statistically predictable pattern in the evolution of the shortest futures series. Overall. 1993). the most liquid volatility futures market (futures on VIX) has been considered. The fact that the VIX futures market is found to be efficient does not invalidate the trading of VIX futures though. These findings are consistent with Konstantinidi et al. our results are in contrast to those found about the efficiency of other futures markets (stock. In the case of interval forecasts no model specification had predictive power.7. Point and interval forecasts have been constructed and their statistical and economic significance has been evaluated. Regarding the statistical significance of the obtained forecasts. the ARIMA(1. On the other hand. Weighted and unweighted combination forecasts have also been considered. (2008). 1989. who had studied the efficiency of the VIX futures market indirectly. To this end. the constructed forecasts did not yield economically significant profits. After all. interest rate and commodities) where predictability in either statistical or economic terms has been documented. This has implications for the efficiency of the VIX volatility futures market. Regarding the economic significance of the obtained forecasts.1. in the case of point forecasts. Conclusions This paper has investigated for the first time whether the volatility futures prices per se can be forecasted. This is because VIX futures can also be used for hedging against changes in volatility.1) model and the PCA regression model. The latter is assessed by means of trading strategies using the VIX futures. currency. our results imply that one cannot reject the hypothesis that the VIX volatility futures market is informationally efficient. the AR(2) model. this was the main motivation for their introduction (see Brenner & Galai. the VAR model. 21 .

J. M. (1989).J. H. 22 . Brenner... VIX futures. Time varying risk premia and forecastable returns in futures markets. M. (1992). Journal of Derivatives. D. 45.M.g. Working Paper. Brenner. Journal of Forecasting. M.W. (1993).E. (2008). The market for volatility trading. Hedging Volatility in Foreign Currencies. & Granger. Brenner. 1. this will be particularly useful for scalpers. 169-193. & Zhang.. Journal of Financial Economics. Fama & French. & Chan. It has been well documented that the predictability in asset returns increases as the horizon increases (see e. Banerjee.. C. R. Operational Research Quarterly. & Marcellino. Financial Analysts Journal. (1978). (2005). Breeden.J. 51.. The combination of forecasts. Bessembinder. New financial instruments for hedging changes in volatility. (1969). 53-59. K. 32. New York University. Inta-day data should also be used to test whether any predictable patterns may be detected within the day. D.Future research should investigate the issue of predictability in volatility futures markets at longer horizons. 621-652. A. Factor forecasts for the UK. J. it may be worth considering more complex model specifications given that the answer on the predictability question always depends on the assumed specification of the predictive regression. Poterba & Summers. D. as the VIX market operates only since 2004. a longer horizon study is beyond the scope of this paper due to data limitations. 1988b. However. 1988). 451-468. Bates. & Galai. 61-65. 279-298. 1988a. M. & Litzenberger. 20. Finally. J. Bibliography Artis. M. Prices of state contingent claims implicit in option Price. 24. & Galai. Shu. Journal of Business. Stern School of Business.

Goyal. Comparing predictive accuracy. An empirical comparison of continuous-time models of implied volatility indices. (1996). E. Hartzmark. Journal of Banking and Finance. R. (1988a). R. Improved methods of combining forecasts. K. (2008). International Economic Review. Returns to individual traders of futures: Aggregate results. Journal of Business and Economic Statistics. & Welch. 253-263. Clemen. (1988b). Journal of Banking and Finance.X.F. Combining forecasts: A review and annotated bibliography. D. 246-273. & French. Dotsis. G. (1995). 31. 30. G.J. Journal of Financial Economics. I. 962-985. G. 3584–3603. 985-1001. C. T. 22. & Skiadopoulos. (2008). 197-204. A. E.R.R. 3. (1984). forthcoming. 20. 559-583. F. & French. Journal of Forecasting. Journal of Futures Markets.Chantziara.A.. 39. Christoffersen.T. 841-862. Fama.W. Grünbichler. 96. Journal 23 .L. Diebold. 28. P. M.S. Energy Economics. Dividend yields and expected stock returns. 13. Journal of Political Economy. & Ramanathan. (1998). (1989). & Mariano. 5. & Skiadopoulos. & Longstaff. (2007). Review of Financial Studies. Can the dynamics of the term structure of petroleum futures be forecasted? Evidence from major markets. A.F. Permanent and temporary components of stock prices. Fama. (2007). A comprehensive look at the empirical performance of equity premium prediction.. Forecasting oil price movements: Exploiting the information in the futures market. Coppola. Granger. A. Evaluating interval forecasts. Psychyios. Valuing futures and options on volatility. International Journal of Forecasting. 34-56. (1987). F. K. F. 3-25. R.

55. (1981). R. 413-432. (2007).E. 64. 249-290. Rayner.of Political Economy. Working Paper. 36. (1997). Reserve Bank of Australia. Kearns. (2001). Luck versus forecast ability: Determinants of trader performance in futures markets. (1991). Journal of Financial Economics. N. & Oomen. Jobson. Some anomalous evidence regarding market efficiency... E. Leybourne. (2004). H. Newbold. Harvey. Performance hypothesis testing with the Sharpe and Treynor measures. and technical trading rule profits: Evidence from foreign currency futures markets. G. forthcoming. Skiadopoulos. P. The profitability of speculators in currency futures markets. A. 1292-1306. Hedging derivatives risks: A simulation study. 95-101. 41. Journal of Finance. P. (2008). 24 . implied volatility indices.S. Journal of Futures Markets. Journal of Futures Markets. M. P. & Tzagkaraki. J. J.J. Lin. 13. Journal of Business. Testing the equality of prediction mean squared errors.C. Journal of Banking and Finance. (1978). Newbold.. Kho. Beyond mean-variance: Performance measurement in a nonsymmetrical world. volatility.L. T. D.C. & Manners. Working Paper.I. Pricing VIX futures: Evidence from integrated physical and risk-neutral probability measures. Y. 49-74. International Journal of Forecasting. & Ennew. & Korkie. The relative efficiency of commodity futures markets. (1999). B. 6.. B. G. 1175-1217. Can the evolution of implied volatility be forecasted? Evidence from European and U.M. 95. 889-908. Jiang. C. Time-varying risk premia. Jensen. (1999). Financial Analysts Journal. 27.. (1996). Leland. S. Kellard. M. 27-35.. 19.D. Journal of Financial Economics. 281-291. Hartzmark. Konstantinidi . University of Arizona.

407-425. 7. Economic Letters. J. (1985). Journal of the American Statistical Association. H.W. 17. J. (1994). Taylor. M.H. technical trading rule performance. Lund University. 73. Taylor. 54. Summers (1988). Miffre. 89. Rubinstein. Data-snooping. Timmermann. Miffre. (2002b). European Financial Management. Journal of Finance. 1167-1179. P. 147-162. and the bootstrap. & Romano. Stock. 235-260. Nossman. Working Paper.J. Economic activity and time variation in expected futures returns.F. Is the VIX futures market able to predict the VIX index? A test of the expectations hypothesis. (1976). Journal of the American Statistical Association. J. Forecasting using principal component from a large number of predictors. 125-142. Rewards available to currency futures speculators: Compensation for risk 25 . Macroeconomic forecasting using diffusion indexes.H. Politis. The valuation of uncertain income streams and the pricing of options. & Vilhelmsson. Trading volatility spreads: A test of index option market efficiency. J. M. (2000). S. (2001a). R. D. & While. 7. 31. & L. S. 20. (2002). M. Efficiency in the pricing of the FTSE 100 futures contracts. (1999). & Pope. & Watson. 9-22. S. Economic significance of the predictable movements in futures returns. European Financial Management Journal. Journal of Financial Economics. The behavior of futures prices over time. Bell Journal of Economics.W. 97. Journal of Business Economics and Statistics. (2001b). Poon. A. & Watson. 713734. 1303-1313. Stock. (2002a). M. Sullivan. 6. J.. Poterba. Applied Economics. 73-79. H. J. 27-60. The stationary bootstrap. Mean reversion in stock returns: Evidence and implications. (2008). 22.. (1992).J. J. A.Miffre. Economic Notes by Banca Monte dei Paschi di Siena SpA. 1647-1691.

(2000). 1023-1041. 521-531. 28.S. 26.E. 68 (Supplement). Econometrica. 11. Journal of Banking and Finance. (1991). 105-116. & Zhu. Futures trading activity and predictable foreign exchange market movements. Wang. Whaley. Risk premia and price volatility in futures markets. J. Journal of Futures Markets. (2000). 1097-1126. J.E. (2006). 68. Zhang. The investor fear gauge. C. Yoo. Economic Record. VIX futures. Journal of Portfolio Management. A reality check for data snooping. 26 . Y. (2004). 165-177. White. 12–17. H.or evidence of inefficient pricing?. G. 26. & Maddala. R. Journal of Futures Markets.

2004 to March 17.8 0.8 -1 2nd Shortest 3rd Shortest i Δys R PC1 PC2 PC3 PC4 Figure 2: Correlation loadings of the first four principal components. Principal component analysis has been applied to the block of daily changes in the three futures series and the explanatory economic variables employed in equation (1) over the period March 26. 2004 to March 17.250 25 200 VIX Futures Prices 20 100 10 50 5 0 26/3/2004 17/5/2004 7/7/2004 25/8/2004 29/9/2004 11/11/2004 27/12/2004 VIX Index 11/2/2005 Shortest VIX Futures Series 3rd Shortest VIX Futures Series 2nd Shortest VIX Futures Series 0 Figure 1: Evolution of the three shortest maturity VIX futures series (left axis) and the VIX index (right axis) over the period March 26. 2005. 0.2 -0.6 -0. 27 VIX (%) 150 15 .6 0.4 -0.2 0 Shortest -0. 2005.4 0.

01 -15.70 24.64 19.792* -2.49 0.69 0.261751 4.18* Table I: Summary statistics.23 1.63* S&P 500 0.25* -6.82 17.87* 2nd Shortest 3rd Shortest -0. 2004 to Mar 17. The first order autocorrelation ρ1.71* 0.01 -0. and the S&P 500 stock index.05 0.22* -0. respectively.972* -1. One asterisk denotes rejection of the null hypothesis at the 1% level.000284 0.26 7. and the slope of the yield curve (calculated as the difference between the prices of a ten-year U.51* 0.966* -2. 28 .47* 0.08 1.004924 0.64 835.03 (5 – 865) 194 178. 2004 to Mar 17.66 8.003561 0. 2005 Shortest Mean Std.17 (5 – 1.05 2.86 12.17 22.243805 5.67 249 2.88 12.58* 0. Deviation Skewness Kurtosis Jarque-Bera ρ1 ADF Average Volume (min-max) 240 158.04 1. Entries report the summary statistics for each VIX futures series and the economic variables.40 278.49 0.921* -3. The null hypothesis for the Jarque-Bera and the ADF tests is that the series is normally distributed and has a unit root.41 Panel B: Summary Statistics of VIX futures and Economic Variables (Daily differences): Mar 26.20 0.75 104.18 0.19* 0.10 36.46 0.53 65.88* -0.18 239.80 0.26 1.28* 0.07 135.973* -3.S.18 5.77* -0.Panel A: Summary Statistics of VIX futures and Economic Variables (Levels): Mar 26.87 11.34 -0.912* -3. government bond and the one-month interbank rate).59 24.218) 2nd Shortest 3rd Shortest 234 169.63** S&P 500 254 1143.11 1. 2005 Shortest # Observations Mean Std.87* 0. the Jarque-Bera and the Augmented Dickey Fuller (ADF) test values are also reported.32 0.00 1.71* 1M interest rate Slope of yield curve 0. The economic variables under consideration are the one-month Libor interbank interest rate. Deviation Skewness Kurtosis Jarque-Bera ρ1 ADF -0.04 -10.26 1.67 21.08 -13.00* 0.04 -15.56 5.67** 186.01 -14.56* -0.79 41.93 0.13 -0.01 10.16 (58 – 974) 1M interest rate Slope of yield curve 249 1.70 0.89* 0.461538 3.13 2.

012 Dependent Variable: 2nd Shortest 181 Coeff. The entries report results from the regression of each VIX futures series on a set of lagged economic variables.022 (-0.932 (-0.262) -132.493) 128.341 (1.071) -0.013 (0.019 (-0.244) 39.466) -0.Included Obs.738) -7.828** (-2. 29 . government bond and the one month continuously compounded interbank rate.000) -33.694) 67.T + a2 ΔFt − 2.223) -31.158 (-1.032 Table II: Forecasting with the economic variables model: In-sample analysis.536) 5.745 (0. The following specification is estimated: where ΔFt .788 (0. One and two asterisks denote rejection of the null hypothesis of a zero coefficient at the 1% and 5% level.078 (1. ΔFt .136) -0.548) 98.930) 104.736 (-0.595 (0. R2 Dependent Variable: Shortest 194 Coeff.946 (0.919) -3.S.385 (-1.446 (1.730 (-0.723 (1.006) 12.099) 0.286) 0. (t-stat) -0.T = c + a1ΔFt −1. it : the one month continuously compounded Libor rate in log-differences and Δyst : the changes of the slope of the yield curve calculated as the difference between the prices of the ten year U. c: a constant. respectively.T + a3 Rt −1 + a4 Rt − 2 + a5it −1 + a6it − 2 + a7 Δyst −1 + a8 Δyst − 2 + ε t . The estimated coefficients.033 (-0.514 (0. augmented by an AR(2) term.783) 0.593 (-1. Rt : the log-return on the S&P 500 stock index between time t-1 and t. 2005.T .535 (-0. and the adjusted R2 are reported.211) -24.277) -74. c ΔFt-1 ΔFt-2 Rt-1 Rt-2 it-1 it-2 Δyst-1 Δyst-2 Adj. (t-stat) -0.233) 0.828 (-1. The model has been estimated for the period March 26.578) 0.782) 84.175 (-0.010 Dependent Variable: 3rd Shortest 113 Coeff.404) 0. 2004 to March 17. (t-stat) -0.662 (1.312 (0.563) 1.085 (1. Newey-West tstatistics in parentheses.T : daily changes in the futures prices between time t-1 and t for a given maturity T.403) 3.605 (-1.843) -0.

T = c + ϕ1ΔFt −1.497 (-1.196) -0. Coeff.238) 0.030 Panel C: ARIMA(1.070) 0.T .062 0.533** (26. for the three VIX futures prices series: ΔFt = C + Φ1ΔFt −1 + ε t .017 Panel B: VAR Model -0. R2 Included Obs. ARIMA and VAR models: In-sample analysis.926) -0.1) Model 226 217 -0.136 (-0.600) -0.012 156 -0.692) (-1. (t-stat) (t-stat) Panel A: AR(2) Model 200 187 -0.643** (-161.013 (-0.002) -0.245) (0.307 (1.933) 0.864* -0.285 (-0. The estimated coefficients. and εt. Panel B: The entries report the estimated coefficients of a VAR.002 0.464 -0. respectively.T .119 (-1.883) 0.T is a (7x1) vector of errors. Panel A: The entries report results from the estimation of a univariate AR(2) specification for the daily changes of each VIX futures series.352 (2.010 -0. 30 . One and two asterisks denote rejection of the null hypothesis of a zero coefficient at the 1% and 5% level.396 -0.162 -0. Φ1 is the (7x7) matrix of coefficients to be estimated.1.308) -0.517) 0.904) -0.T .984* (-58.064) -0.992* 0.413 (-1.Included Obs.966) (-1. C is a (7x1) vector of constants.1. c φ1 φ2 Adj.933) (1. where ΔFt is the (7x1) vector of changes in the three futures prices series.609) (1.121 (-0.906) 0.009 Dependent Variable: 3rd Shortest Coeff.373 -0.415 0.021 Table III: Forecasting with the univariate autoregressive.068** (-2.T + θ1ε t −1 + ε t .544) 0.018 0.028 0.051 0.T = c + ϕ1ΔFt −1.116 (0.015 (0. Newey-West t-statistics in parentheses and the adjusted R2 are reported.249) (2. 2005.262) (-1. 2004 to March 17. R2 C ΔF1t-1 ΔF2t-1 ΔF3t-1 Adj.469 (-1.465) 0.034* (-3.895* (18.T + ε t .928) -0.517) (-0. R2 Dependent Dependent Variable: Variable: Shortest 2nd Shortest Coeff.148 (-0. The models have been estimated for the period March 26. (t-stat) 118 -0.1) model: ΔFt .T + ϕ2 ΔFt − 2.907) -0. namely: ΔFt .093) 0. c φ1 θ1 Adj.370) (-2. Panel C: The entries report the estimated coefficients of a ARIMA(1.175) (-1.079 (-0.739) 0.537) 0.092 -0.

246 (0.356) 0.571 (1.632) 0.584) 0.Included Obs.003) -0.222 (0. (t-stat) -0. and the adjusted R2 are reported. One and two asterisks denote rejection of the null hypothesis of a zero coefficient at the 1% and 5% level.602) -0.038 Table IV: Forecasting with the principal components analysis model: In-sample analysis. (t-stat) -0.181 (-0.177 (-0.211 (-0. The model has been estimated for the period March 26.974) (-0.797** (2. PC2.445 (1.360 (-0.879) 0.007 Dependent Variable: 2nd Shortest 116 Coeff.392) -0.225) -0.278 (-0.606) -0.228 (1. 2004 to March 17. 2005.572) 0.008 Dependent Variable: 3rd Shortest 107 Coeff.116 (-0.975) 0.391) 0. respectively.598) 0.421 (1.157) -0.121 (0.777) 0.656) -0.372067 (-1.409 (1.T = c1 + ∑ r1 j PC1t − j + ∑ r2 j PC 2t − j + ∑ r3 j PC 3t − j + ∑ r4 j PC 4t − j + ε t . The entries report results from the regression of the futures price changes on the lagged first four principal components PC1.352 (1.038) -0.001 (0.489) -0.385) 0.160 (-0. 31 .234 (-0.636) (-1. R2 Dependent Variable: Shortest 116 Coeff.090) 0.T .364211 (-1. The j =1 j =1 j =1 j =1 2 2 2 2 estimated coefficients.169) -0. Newey-West t-statistics in parentheses. (t-stat) -0.669) -0.322149 (1.182) -0.385) 0. c PC1t-1 PC1t-2 PC2t-1 PC2t-2 PC3t-1 PC3t-2 PC4t-1 PC4t-2 Adj.005 (-0.342 (-0.275 (-0.109772 (-0.011) 0. PC3 and PC4 derived from the set of the three VIX futures prices series and the economic variables: ΔFt .520) 0.

respectively) are also reported.06 RMSE 8.1) Model Point Forecasts 5.70% MCP Panel G: Unweighted Combination Point Forecasts 7.24 4. 2008.68 RMSE 4.33% 51.76* 5. the mean absolute prediction error (MAE).09 MAE 51.77 RMSE 4. respectively.46 3. Namely H1: the random walk outperforms the model and H2: the model outperforms the random walk.86** 3. the VAR model (Panel D).05 MAE 52.35 3.41% Panel H: Weighted Combination Point Forecasts RMSE 8.50 MAE 4.05** 5. The root mean squared prediction error (RMSE).Shortest 2nd Shortest 3rd Shortest Panel A: Random Walk 7.23 4.17* 50.1.26 5.71%+ Panel E: ARIMA(1.33 3.60* 3. 32 .60 5.35 2.78%++ Panel D: VAR Model Point Forecasts 7.59% MCP 54.1) model (Panel E).36 2.26% MCP 53.13 5.32 3.80% 49.71% Table V: Out-of-sample performance of the model specifications for each one of the VIX futures prices series.76* MAE 5. and the mean correct prediction (MCP) of the direction of change in the value of each VIX futures price series.26** 3.01 5.49 4.16* 4.71% 49.16** 5.31 2. the economic variables model (Panel B).35** 49.04** MAE 5. the AR(2) model (Panel C).16% 52.25 3. to test the null hypothesis that the random walk and the model under consideration perform equally well.81 RMSE 4.42% 50. One and two asterisks (crosses) denote rejection of the null hypothesis in favour of the alternative H1 (H2) at significance levels 1% and 5%.74 3.37 4.90* 3.07%+ Panel C: AR(2) Model Point Forecasts 7.09% 50. The Modified DieboldMariano test (for RMSE and MAE) and the ratio test (for MCP) are employed.97 MAE 51.08% 47.49% 52.75% MCP 54.14 3. 2005 to March 13.98 MAE + 51. Two alternative hypotheses H1 and H2 are considered.1.41% 50.89 MAE Panel B: Economic Variables Model Point Forecasts RMSE 7.16 4. The models have been estimated recursively for the period March 18.26% MCP 55.55 RMSE 4. and the PCA regression model (Panel F) have been implemented.46 3.42* 4. The random walk model (Panel A). Results for the unweighted and weighted combination point forecasts (Panel G and H.56 3.68 RMSE 7.83** MCP 51.29%++ Panel F: PCA Regression Model Point Forecasts 5. the ARIMA(1.40* 5.45% MCP 54.

1.28% 6.70% 11.68% 9.43* 528.25% 14.80% 6.1) Model Interval Forecasts 17.33% 11.49% 65.52% 14.96* Panel C: VAR Model Interval Forecasts 5.42% 7.13* 44.00* 262.69* 13.07% 20.03* 221.31% 35. respectively) are also presented.25* Panel D: ARIMA(1. Entries report the percentage of the observations that fall outside the constructed intervals.91* 106.52% 32.57* 49.06% 11.65% 7.94* 83.18% 7.94* 395.12% 22.20* 77.1) model (Panel D).Interval Forecasts # Violations LRunc # Violations LRunc # Violations LRunc # Violations LRunc # Violations LRunc # Violations LRunc # Violations LRunc Shortest 99% 95% 2nd Shortest 99% 95% 3rd Shortest 99% 95% Panel A: Economic Variables Model Interval Forecasts 6.89% 12.44% 495.96% 81.19* 39.98* Panel G: Weighted Combination Interval Forecasts 5.94% 23.54% 9.93* 29.95% 11. and the PCA regression model (Panel E).94% 13.1.25* 13.11% 6.69* Panel F: Unweighted Combination Interval Forecasts 8.70% 6. The results are reported for daily 99% and 95%-interval forecasts generated by the economic variables model (Panel A). One and two asterisks denote rejection of the null hypothesis at 1% and 5% significance levels. 2008.07* 74.44% 76.93* 74.11% 52.15* 94. 2005 to March 13.23* 134.93% 75. the ARIMA(1.69% 7.67% 11.96% 290.05* 44.25* 29.87% 54.10% 6.64* 38. the VAR model (Panel C). The null hypothesis is that the percentage of times that the actually realized futures price falls outside the constructed (1-α)%-interval forecasts is a%.59% 57.19* 66.74% 12.87% 57.57* 15.14* 11. the AR(2) model (Panel B).02% 11.15* 39.59* Panel E: PCA Regression Model Interval Forecasts 5. The models have been estimated recursively for the period March 18.00% 6. and the values of Christoffersen’s (1998) likelihood ratio test of unconditional coverage (LRunc) for each VIX futures price series.40* 81.99* 97. Results for the unweighted and weighted combination 99% and 95%-interval forecasts (Panel F and G.11* 95.91* 102. respectively.39* 93.39* 11.87* Table VI: Statistical efficiency of the constructed interval forecasts.25* 131. 33 .86* Panel B: AR(2) Model Interval Forecasts 6.13% 48.24% 16.80% 39.50% 10.87* 40.61* 36.44* 11.48* 25.

025 95% CI (-0.14) Ap 0. 0. 0. 0.85) Panel F: Unweighted Combination Point Forecasts SR 0. 0.007 0.09. 0.59.07) (-0.38.415 (-0. 0. 0.057 95% CI (-0.014 -0. 0.03. 0.35) (-0.86) (-0.179 0. 0.02.064 SR (-0.39) (-0. The entries show the annualised Sharpe ratio (SR) and Leland’s (1999) alpha (Ap) and their respective bootstrapped 95% confidence intervals (95% CI) within parentheses. 0. 0. 0. 0. 0.78.82) (-0. 0.128 -0.83) (-0.90.009 -0. 0.10.10.39) Panel E: PCA Regression Model Point Forecasts SR 0.59.648 -0.0532 [95% CI = -0.099 -0. 0.06) Ap 0.51.040 0.064 -0.02.09. 0. 0.14) (-0.09) (-0. 0.0178 [95% CI = (-0. 0.12) (-0.001 95% CI (-0.09.007 95% CI (-0. 0.011 -0.013 -0.63.05.01.06.146 -0.025 95% CI (-0.89. 0.14. 0.65. 0.69.06.09) (-0.42) (-0.08) Ap 0.08.039 -0. the ARIMA(1.10) (-0. 0.296 0.04. 0. 0.07) Ap 0.26.072 0.030 0.09)] for the shortest maturity series.46) (-0.363 95% CI (-0.06. 0. 34 .50) Panel G: Weighted Combination Point Forecasts 0. 0.11.09) Ap 0. 0. 0. 0.85) (-0.09.098 0.06.05) (-0.024 -0.12.07. 1. the AR(2) model (Panel B).0.51.038 95% CI (-0.46) Panel C: VAR Model Point Forecasts SR -0.31) -0.008 -0.13)] for the third shortest maturity series.110 95% CI (-0.04) (-0. The strategy is based on point forecasts obtained from the economic variables model (Panel A). 0.05) (-0. 0.06) (-0.147 -0. 0.Shortest 2nd Shortest 3rd Shortest Panel A: Economic Variables Model Point Forecasts SR 0.98) (-0.08) (-0. Results for the unweighted and weighted combination point forecasts (Panel F and G respectively) are also reported. 0.86) -0.1) model (Panel D) and the PCA regression model (Panel E).56. the VAR model (Panel C).0421 [95% CI = (-0.091 95% CI (-0.016 95% CI (-0.51) (-0.04. 0. 2008.06) (-0.238 95% CI (-0.11) Ap -0.111 (-0.74. 0.47.06.08.11.06) (-0.72.240 -0. 0.1. The SR for a naïve buy and hold strategy in VIX volatility futures is 0.01) 95% CI Ap 95% CI 0. 0.08) Table VII: Trading strategy with VIX futures based on point forecasts from March 18.231 (-0.11)] for the second shortest maturity series and 0.1.36) (-0.034 -0.69.016 95% CI (-0.09) (-0.64.198 0.27. 0.1) Model Point Forecasts SR 0.015 0.026 0.023 -0.60) Panel B: AR(2) Model Point Forecasts SR 0.60) (-0. 0. 2005 to March 13.10.72) Panel D: ARIMA(1.

26) (-0.044 -0.05.07) (-0.11)] for the second shortest maturity series and 0.11) (-0. 0.363 0. 0. 0.07) (-0.006 -0.446 0. 0.10) 0. 0.0178 [95% CI = (-0.058 0.07.003 95% CI (-0.038 -0.11.016 0.08. 0.12) (-0.0421 [95% CI = (-0.15.09.83) Panel F: Unweighted Combination Interval Forecasts SR 0. 0.01.69) (-0.80) (-0.330 Ap 95% CI (-0.08) 0.011 -0.08) (-0.052 95% CI (-0.55.41.16) (-0.607 0. 0.08) (-0.103 Ap 95% CI (-0.49.95) (-0. the AR(2) model (Panel B). 2008.09) (-0.56. 0.07) (-0.019 -0. 0.03.88.07.40.04.47) (-0. 0. 0.05.07.08) (-0. 0.002 0.019 -0.04. 0.30.13) 0. 0.13.09.05) (-0.062 -0. 0.13) (0.006 0.06) 0.08.37) (-0. the ARIMA(1.01.07) (-0. 0.47.46) (-0. 0. 0.22) (-0.06) (-0.09.69.71.187 -0.00. 0.12.04.061 0.006 -0. 0.25. 0.004 0.047 -0. 0. 0.49.08.02) -0.012 95% CI (-0.061 0.12) Shortest Table VIII: Trading strategy with VIX futures based on interval forecasts from March 18.441 0. 0.036 0.037 -0.11) (-0.044 0.090 Ap 95% CI (-0.391 Ap 95% CI (-0.014 0.67) (-0.017 0. 1.55.62. 0. 0.18) (-0. 0.06. 0.05. 0. 0.005 0.04) (-0.83) (-0.07.14 -0.14) 0.09) (-0.55) (-0.038 Ap 95% CI (-0. 0. 0.045 0.11) (-0. 0.021 0.777 -0. 0.07.08) (-0.10.66.005 -0.52) (-0. 0.08) (-0. The entries show the annualised Sharpe ratio (SR). 0.10) (-0.069 0. 0.48) (-0.15) (-0.14) (-0. 0.026 95% CI (-0.37) Panel E: PCA Regression Model Interval Forecasts SR 0.07.09.09)] for the shortest maturity series.3274 -0.436 0. 1.72.13.1. 0.52.52) (-0.02.093 0.61) (-0. 0.06) (-0. 0.06) (-0.0532 [95% CI = -0.078 -0.89) Panel D: ARIMA(1.57.10.058 95% CI (-0.031 0. 0.33. 0.08.71) (-0.65. 1. 0.130 0.51) (-0.165 Ap 95% CI (-0.46) (-0. 0.073 0.05.10.06.13)] for the third shortest maturity series.87) (-0.06) (-0. 0.112 -0. 35 .64.64.017 -0.09) 0.009 0. 2005 to March 13.212 0.060 -0. 0.033 0.57.12) (-0.12) (-0.11.42. 0.07.05. Leland’s (1999) alpha (Ap) and their respective bootstrapped 95% confidence intervals (95% CI) within parentheses. 1.32. Results for the unweighted and weighted combination point forecasts (Panel F and G respectively) are also reported.314 0. 0. 0.52) Panel G: Weighted Combination Interval Forecasts SR 0. 0. 0.56) (-0.81) (-0.029 0. 0.009 0.1.51.064 95% CI (-0.058 95% CI (-0. 0.006 0.018 95% CI (-0. 0.29.07) (-0.11) (-0. 0.029 0.73. 0. 1.80) (-0. 0. 0.07.002 0. 0.02.37) (0.032 0.029 0. 0.004 -0.2nd Shortest 3rd Shortest Interval Forecasts 99% 95% 99% 95% 99% 95% Panel A: Macro Regression Point Forecasts SR 0. 0.08) (-0.37879 Ap 0.015 -0. 0.11) (-0.253 0. 0.023 0.03. 0. 0. 0. 0.24 -0.017 0.03.07. 0. 0.58) Panel C: VAR Model Interval Forecasts SR 0.092 0.57.04.36.09.021 -0.85.06.106 0. 0.44) (-0. 0.1) Model Interval Forecasts SR 0.59) (-0.08) (-0.158 0. 0.45.004 -0. 0.59) (-0. 0.66.29.64) Panel B: AR(2) Model Interval Forecasts SR 0.228 0.57.77) (-0.06. The SR for a naïve buy and hold strategy in VIX volatility futures is 0.1) model (Panel D) and the PCA regression model (Panel E).09) (-0.037 0.08) (-0.42. 0. 0.195 0.93) (-0. 0.02. 0. 1. 0.087 -0.09) (-0.41) (-0.08) (-0.99) (-0.03) (-0. The strategy is based on 99% and 95%-interval forecasts obtained from the economic variables model (Panel A).06.82) (-0. the VAR model (Panel C). 0.70.

It reflects the market participants’ consensus view about expected futures stock market volatility and can be considered as an “investor fear gauge” (Whaley. We adjusted the VIX futures series accordingly. 2007. the multiplier was increased from $100 to $1..g. 2000). we proxy the random walk with the naïve rule that “the predicted change in the futures prices has a 50% chance to be positive and a 50% to be negative”. 2007. for a study using VIX futures).Footnotes 1 VIX is an implied volatility index that tracks the implied volatility of a synthetic option on the S&P500 with constant time to maturity (thirty days). respectively. The VXD and VXN are implied volatility indices that track the implied volatility of a synthetic option on the Dow Jones Industrial Average and the Nasdaq 100. VIX futures were based directly on the underlying VIX volatility index instead on the “Increased-Value index” (VBI=10*VIX). 2 The CBOE launched the VXD and VXN volatility futures in April 25. the traded futures prices were reduced by a factor of 10. and Nossman and Vilhelmsson. As a result. this corresponds to a 59% increase from January 3. on January 2.3 billion in terms of market value.. This is because the corresponding loss function cannot be defined for the benchmark model. the VIX futures were rescaled in two ways. On the same date. Jensen’s (1978) definition of futures market efficiency is adopted. This is to say that the random walk case corresponds to an MCP equal to 50%. 36 . However. a market is efficient with respect to the information set It. as well. 5 On March 26. respectively. 1999. and the references therein. the H1 and H2 hypotheses are stated as H1: MCP < 50% and H2: MCP > 50%. but the $ value of each contract did not change. 2008. 4 Efficiency in the underlying implied volatility index market does not necessarily imply an efficient volatility futures market. Kellard et al. Regarding the liquidity of volatility futures.481 contracts or $57 million in terms of market value.792 contracts or $1. there may be other factors/information flow that affect volatility futures markets. White’s (2000) test is not applied to the MCP metric. 2008 the open interest for VIX futures was 55. 3 This question is distinct from the question whether futures markets are efficient in the sense that the futures price is an optimal forecast of the underlying spot price to be realized on the contract expiry date (see e. 6 Note that the MCP cannot be calculated for the random walk model. 2005 and July 6. in the case where it is impossible to make economic profits by trading on the basis of this information set. In addition. 2007. Coppola. First. 7 In the case of the MCP. the open interest of VXD and VXN futures was $19 and $4 million. The trading volume was 2. In our study. with constant time to maturity (thirty days).000. 2008. respectively. Second.

The block size follows a geometric distribution with mean block length 1/q.T in our case. It involves re-sampling blocks of random size from the original time series to form a pseudo time series (or a bootstrapped sample). The main feature of this procedure is that the re-sampled pseudo time series retain the stationarity property of the original series.1 that ˆ corresponds to a mean block size of 10. The nonnormality of volatility futures returns is consistent with previous findings in the related literature for other futures markets (see e.8 The stationary bootstrap is applicable to weakly dependent stationary time series. kurtosis and skewness range from 6 to 13 and from –1 to 2 respectively. 1985. In our case. 9 We do not employ Jobson and Korkie’s (1981) testing procedure to investigate the statistical significance of the SR. This is a reasonable block size given the low autocorrelation in fi . we choose q = 0.. stationarity has been found ˆ for fi .t . and the references therein). we use bootstrapped confidence intervals that are robust to non-normality. (1999). Following Sullivan et al. 37 .g.T (results not reported). Instead. This is because this test is based on the assumption that the returns of the strategy under consideration are distributed asymptotically normally. Taylor. the trading strategies’ returns exhibit excess kurtosis and skewness.t . across the three futures maturities.