Article by: Juan Carlos Sosa
Published by: Boston College
Date: 1 Jan 2000
“Previous literature examines whether the volatility estimates implicit in option prices constitute accurate forecasts of future volatility for the underlying asset. Most studies address two questions. Do Black-Scholes implied volatilities predict future volatilities? Do they incorporate all past time series information? The empirical answers to both questions are conflicting. However, previous studies typically find that implied volatilities overestimate future volatilities. Christensen & Prabhala (98) provide evidence of a structural shift in the pricing mechanism of OEX options around the Crash of ’87. In contrast with the pre-Crash results of Canina & Figlewski (93), CP show that after the Crash implied volatilities predict future volatilities and dominate moving average forecasts. They also show that implied volatilities are unbiased, and that errors-in-variables is responsible for previous bias findings. The objective of this dissertation is fourfold. First, we implement encompassing regressions on a recent sample of OEX options. Our full-sample findings are consistent with other studies that support the informational efficiency of implied volatilities in recent years. However, the regressions are very sensitive to the degree of moneyness of implied volatility. Furthermore, we find that the degree of predictive power of implied volatility is strongly time-frame dependent. Second, we show that implied volatilities do overestimate realized volatilities and that CP’s unbiasedness finding is due to an ill choice of instrument, in addition to a low test power. Third, we attempt to correct for potential misspecification in the encompassing regressions and find that lagged implied volatility is important. Yet, we show that the significance of lagged implied volatility is not a symptom of market inefficiency, but of skewness-related model error in implied volatility estimates. Finally, we assess the economic value of informational efficiency via a trading analysis, and find it to be quite limited. In summary, we find that at-the-money implied volatilities are biased estimators of realized volatility, that they suffer from substantial model error, that their predictive power is time-frame dependent, and that moving average estimators perform just as well from an economic point of view.”
Full article (requires a subscription or payment): Link