A Statistical Neural Network Approach for Value-at-Risk Analysis

This study develops a new methodology based on ANN for Value-at-Risk (VaR) modeling. Specifically, we propose a statistical procedure for ANN model selection. The statistical ANN deals with each layer individually and estimates the weights of subsequent layer with those of preceding layers fixed. This allows the derivation of statistical theory for model selection, which reduces the need to fit a comprehensive set of models. Experiment results show that the statistical ANN approach performs well on stock index return series compared to traditional forecasting methods.