On Accuracy of Long-Term Risk Forecasts by Normal Variance-Mean Mixtures Decomposition Algorithm*

Abstract

This article provides an accuracy and applicability analysis of the approach to risk forecasting using parametric mixture models. The studied method is based upon results of the modified grid-based two-step decomposition algorithm for variance-mean mixtures. Instead of setting a fixed forecast interval, an approach is introduced to dynamically monitor relevant metrics for forecasts in a wide time frame, producing the basis for decision making regarding the quality and reliability of predictions for certain periods of time.