Alternative volatility models for risk management and trading: Application to the EUR/USD and USD/JPY rates

Abstract

This paper examines the forecasting ability of several alternative models of currency volatility applied to two foreign exchange rates: EUR/USD and USD/JPY which, according to the Bank for International Settlements (BIS), represent 45 per cent of the $1.9 trillion daily trading volume on the world currency markets. Benchmarked against two naive ‘random walk’ models and a RiskMetrics volatility model, the predictive abilities of the autoregressive (AR(p)); generalised autoregressive conditional heteroscedasticity (GARCH(p,q)); new modelling approaches such as stochastic variance (SV) and neural network regression (NNR) models; and two different model combinations are assessed at the one-day, five-day and 21-day horizons not only in terms of traditional forecasting accuracy measures but also in terms of risk management efficiency under the value-at-risk (VaR) framework and trading performance with a volatility filter strategy. These daily models are developed for the period from 2nd January, 1998, to 13th May, 2002 (1,116 observations) and tested out-of-sample from 14th May, 2002 to 28th March, 2003 (223 observations). The essence of the contribution is three ‘forecasting’ competitions using the same forecasts, some obtained from new modelling techniques, for three different purposes: the first is statistical accuracy, the second VaR and the third is simulated trading. Although no single volatility model emerges as an overall winner in terms of forecasting accuracy, risk management efficiency and FX trading performance, ‘mixed’ models incorporating market data for currency volatility, NNR models and combinations of models perform best most of the time.