On the Predictability of Stock Market Volatility: Implied vs. Historical Volatility
Ipatti, Jenni (2002)
Kuvaus
Kokotekstiversiota ei ole saatavissa.
Tiivistelmä
The purpose of this study is to examine the performance of various volatility forecasting models in the Finnish stock market. Moreover, this study aims to answer the question, whether the option implied volatility forecasts offer a better forecasting tool than past based approaches. The data used in the study consists of Finnish Option Index daily closing levels and index option quotes from 1992 to 1998. The set of models is limited to the most familiar ones, however, including the volatility forecasting model based on RiskMetricsTM, since its absence in the earlier literature.
Three hypothesis tested are: if (i) more complex models of historical volatility result better forecasts than simple ones, if (ii) option derived forecast are superior to historical estimates, and finally, if (iii) volatility forecastability decreases with the horizon. The task of assessing the best forecasting model is done both, with regression based method and with the conventional error statistics in 5- and 10-day forecasting horizons.
The first hypothesis, that more sophisticated models of historical volatility result better forecasts than simple approaches, is clearly rejected, since the predictive power of the GARCH model is not sufficient to beat the simple models. In the forecasting horizon of 5 days, we can confirm our hypothesis, that implied volatilities do provide forecasting tool superior to historical estimates. However, as the forecasting horizon is doubled implied volatilities loose practically all of the predictive power. Tests of our third hypothesis, give somewhat mixed results. The results vary both with the forecasting model, as well as with the evaluation approach.
Three hypothesis tested are: if (i) more complex models of historical volatility result better forecasts than simple ones, if (ii) option derived forecast are superior to historical estimates, and finally, if (iii) volatility forecastability decreases with the horizon. The task of assessing the best forecasting model is done both, with regression based method and with the conventional error statistics in 5- and 10-day forecasting horizons.
The first hypothesis, that more sophisticated models of historical volatility result better forecasts than simple approaches, is clearly rejected, since the predictive power of the GARCH model is not sufficient to beat the simple models. In the forecasting horizon of 5 days, we can confirm our hypothesis, that implied volatilities do provide forecasting tool superior to historical estimates. However, as the forecasting horizon is doubled implied volatilities loose practically all of the predictive power. Tests of our third hypothesis, give somewhat mixed results. The results vary both with the forecasting model, as well as with the evaluation approach.