Quantile autoregression neural network model with applications to evaluating value at risk

A novel quantile autoregression neural network (QARNN) model is proposed that combines an artificial neural network (ANN) structure with the quantile autoregression (QAR) method for time series data. It has two advantages. First, the QARNN model is flexible and can be used to explore potential nonlinear relationships by taking advantage of the powerful nonlinear processing capacity of ANN. Second, the QARNN model provides more information for decision-making by using the ability of QAR to discover the entire conditional distribution of time series. Thus, it achieves high prediction accuracy and is able to evaluate value at risk in practice. 36