Forecasting Financial Risks By Modified Grid-Based Decomposition Algorithm For Normal Variance-Mean Mixtures

We describe an algorithm to forecast financial risks using parametrized models of normal variance-mean mixtures. The proposed method takes a set of vectors as the input, containing a fixed number of the distribution parameters – the final result of the modified two-step grid-based decomposition algorithm applied to a moving time window. In this article we use the class of generalized hyperbolic (GH) distributions as an example for method demonstration. Practical applications of the method proposed and processing speed are discussed in detail. We also describe the process of calibrating the method as well as provide detailed instructions on how to find the best fitting model. Using real market data we illustrate the accuracy of the resulting forecasts depending on the method settings, including long-term forecasts.