Predicting exchange rate volatility: genetic programming vs. GARCH and RiskMetrics

This article investigates the use of genetic programming to forecast out-of-sample daily volatility in the foreign exchange market. Forecasting performance is evaluated relative to GARCH(1,1) and RiskMetrics models for two currencies, DEM and JPY. Although the GARCH/RiskMetrics models appear to have a inconsistent marginal edge over the genetic program using the mean-squared-error (MSE) and R2 criteria, the genetic program consistently produces lower mean absolute forecast errors (MAE) at all horizons and for both currencies.