Stima prospettica delle misure finanziarie e di rischio mediante reti neurali dinamiche: un'applicazione al mercato statunitense

The purpose of this article is to explain how a technology based on dynamic neural networks is used for prices forecasting in order to calculate risk measures, such as the Expected Shortfall (CVar). The paper is focused on US market and is divided into four parts: First section identifying and analyzing the sixteen most representative indexes of the US market from 1999 to 2019. Three sub-periods are thus defined, corresponding to financial relevant time-windows. In the second part, starting from the complete set of indexes, a portfolio analysis has been carried out in order to determine the weights of each asset in the reference time periods. This study is preliminary to identifying which of these are significant in terms of asset allocation. Third part discusses the working principles of NAR and NARX dynamic neural networks used as predictors for the subset of the more significant indexes. Moreover the performance measures obtained by the forecaster during the training phase (in-sample) and during the back-testing phase (out-of-sample) are provided. Once assessed of the statistical and econometric reliability of the forecasting instrument, the paper focuses on the future values of the indexes. Starting from these predictions, financial and risk measures have been estimated in a forward-looking perspective.