Forecasting Value-at-Risk Using Deep Neural Network Quantile Regression

Abstract In this article, we use a deep quantile estimator, based on neural networks and their universal approximation property to examine non-linear association between the conditional quantiles of dependent variable predictors. This methodology is versatile allows both different penalty functions, as well high dimensional covariates. We present Monte Carlo exercise where finite sample properties estimator show that it delivers good performance. forecast value-at-risk find significant gains over linear regression alternatives other models, which are supported by various testing schemes. Further, consider also an alternative architecture mixed frequency data in networks. article contributes interpretability network output making comparisons commonly used Shapley Additive Explanation values method partial derivatives.