A Hybrid Econometric-ANN Model for Value-at-Risk Estimation

In this study, we propose a hybrid model integrating econometric models, i.e. ARMA, EGARCH, andartificial neural network (ANN) models for Value-at-Risk (VaR) estimation. The ARMA and EGARCH modelsare fitted as benchmarks. Besides, they generate input and output variables for ANN models. One ANN modelis built based on the variables provided by ARMA model to re-estimate the conditional mean, while anotherANN model is built based on the variables provided by EGARCH model to re-estimate the conditionalvolatility. Based on these steps, the estimate of VaR could finally be constructed. The hybrid model achievesbetter efficiency in input variables selecting because they are selected and newly created by time seriesmodels. Repetitive trial and error process could be effectively eliminated to one time series process. On theother hand, the performance of traditional time series models could be further enhanced by the forecastingpower of ANN models. Empirical study shows that the hybrid model can improve the predictive power in theframework of both accuracy and reliability.