[Work in progress] Verification of forecast effectiveness for selected volatility estimators

Abstract

Aim: The aim of this study is to determine which volatility forecasting method produces results that are closest to the actual and whether the use of estimators using OHLC prices will affect the accuracy of forecasts.

Methodology: This study examines five models – a historical model, GARCH(1,1) and three GARCH models with selected volatility estimators, including Parkinson, Garman-Klass and Rogers-Satchell. The sample used daily prices, with each instrument having 2001 observations and a 20-day forecast horizon.  Prediction errors were calculated using two symmetric measures, RMSE and MAE.

Findings: The empirical results determined that no specific approach is universally regarded as superior. It is recommended that naive methods or the standard GARCH method be used, as they are simpler than the complex models with selected estimators and save operating time. An important finding is that volatility estimators improve results for stocks, but not for other types of instruments. An analysis of specific instrument types indicates that methods using volatility estimators obtain better results than classical methods when analyzing stock prices. However, when examining other instruments, classical methods prove to be better.

Implications: This work can assist researchers in selecting the appropriate model for specific data and indicate whether the use of a different estimator would enrich the results of forecasts. Further research could investigate the impact of higher frequency data on the performance of volatility estimators.

Originality/value: This study investigates whether the Polish market will respond to different estimators in a manner consistent with the responses observed in other markets around the globe. Another aspect that this work allowed verification of was whether the best model differs according to the type of instrument, and the results confirmed that these differences do exist (for stock prices, the model with the Rogers-Satchell estimator, for currencies the GARCH(1,1), and for commodities the historical method).