Forecasting Properties of Neural Network Generated Volatility Estimates

Abstract

The price of an option, from an arbitrage pricing model, as an estimate of the option’s fair market value is only as good as the volatility measures being used. Investors having access to superior forecasts of future volatility are likely to devise trading strategies that would generate profits by identifying mispriced options. The extant literature considers informational efficiency and finds that implied volatility is not an unbiased forecast of future volatility. The analysis suggests that implied volatility is related to moneyness, time to maturity and the ratio of put-to-call trading volume, however, the exact functional relation is unknown. In the absence of a known functional form, artificial neural networks (ANN) can generate a model that captures this systematic relationship. Results using equity option data show that, in general, neural networks do not produce superior estimates of future volatility.