Forecasting sterling/dollar volatility: a comparison of implied volatilities and AR(FI)MA models.

Using high frequency intraday returns, we construct a proxy for the USD/GBP exchange rate return realized volatility. It is shown that they dynamics of the logarithms of realized volatilities can be captured be either a fractionally integrated long memory likelihood. The paper compares the forecasting ability of a short memory model, a long memory model and the implied volatilities from OTC foreign currency options, for realized volatilities over horizons of one month and three months. It is found that the out-of-sample results are sensitive to the sampling frequency used to construct the realized volatility. For a five-minute frequency, implied volatilities, a short memory model and a long memory model possess similar forecasting ability for both forecast horizons. For a thirty-minute frequency, the results show that implied volatilities perform better than both intraday models for the one-month forecast horizon and no worse for the three-month horizon. We also find that forecasts from long memory model do not contain information incremental to implied volatilities for both forecast horizons while a short memory model may contain information incremental to implied volatilities for the one-month forecast horizon.