| Artificial neural networks (ANNs), a generalized nonlinear nonparametric econometric method has been shown to perform very well in two applications.;In the first application, ANN has been applied to price S&P 500 index call options. Using current asset price, interest rate, time to maturity, exercise price, and open interest as inputs and call price as output, the ANN with five middle layer units outperforms the Black-Scholes formula (BSF) based on all four performance measures. The residual analysis shows that the ANN outperforms the BSF in reducing the residual autocorrelations, the extent of underpricing, the residual heteroscedasticity and the moneyness and timeliness biases. The economic implications of the ANN model are consistent with the properties of call option prices. Open interest has long been ignored in the option pricing literature, and is found here to be an important factor in determining option prices. This application provides an alternative to the traditional BSF that does not require the many unrealistic assumptions required by the BSF and gives much superior performance.;In the second application, ANN has been applied to forecast returns on S&P 500 index using five economic variables, inflation rate, unemployment insurance claim rate, price-to-earning ratio, real interest rate and money growth rate. The linear regression (LR) gives better fit than those reported in the existing return forecasting literature. The goodness of fit can be largely improved using an ANN by not imposing linear restrictions. The first order residual autocorrelation is smaller and the residual heteroscedasticity is less severe for the ANN than for the LR model. The substantial nonlinearity has been found by sensitivity analysis. The ANN forecasts can improve the average annual gains generated by a simple profit trading rule, though they do not statistically significantly improve the linear or the random walk (RW) forecasts in terms of squared forecasting errors. Moreover, the ANN outperforms the LR and RW in predicting the direction change of returns on S&P 500 index, and more importantly, the ANN is much better than the LR and the RW models in predicting the down turns of the S&P 500 index. |
