Forecasting volatility in New Zealand bank bill futures

Abstract

This paper compares volatility forecasts of four competing models based on historical data against implied volatility forecasts using New Zealand bank bill futures prices. Model forecasts are evaluated out of sample using mean absolute error and root mean square error and are tested for bias using regression techniques. GARCH was found to forecast volatility most accurately over 30-day and 60-day forecast periods; however, each of the five models gave similar errors over 90-day and 180-day forecast periods. Implied volatility forecasts gave the coefficient of determination for each period but systematically over forecasted volatility in all periods. When systematic bias was removed, the errors associated with this model fell below that of the GARCH model.