Sensitivity Analysis of Input Variables in ANN-Based Option Pricing Models

Parametric models in option pricing have consistently failed to provide results truly consistent with the observed market data, due to the fact that market participants change their option pricing attitudes from time to time.Furthermore, most of these models are based on questionable assumptions such as the assumption of constant volatility, yet empirical evidence strongly suggests otherwise. This has led to the development and the use of alternative non-parametric models such as the ANN models. However, in spite of the well-known significance of the sensitivity indices in option pricing, little attempts have been made to provide a mechanism for computing these indices from the ANN model. Consequently, this study sought to provide a mechanism for and conduct sensitivity analysis from the ANN model outputs and conclude by comparing the performance of the hybrid model and the conventional ANN model developed. All the indices developed lied within their desired range of values and this, being a resultant of the derivatives of the developed ANN output with respect to its various inputs, demonstrated how reliable the model was. Results not only demonstrate how the indices could be computed from the ANN models, but also play a significant role in the validation of the model.