NEURAL NETWORKS IN FINANCIAL RISK MANAGEMENT

Abstract

Models based on artificial neural networks has produced outstanding results in many applications (LeCun et al., 2015). There are several papers available in financial time series analysis, mostly focused on predicting return (Galeshchuk, 2016; Badics, 2014; Fischer & Krauss, 2018) or volatility (Lahmiri, 2017, Kristjanpoller etal., 2014). This paper introduces an expected shortfall estimation model based on an artificial neural network. Expected shortfall is a coherent (Artzner et al., 1999) risk measure first introduced by Rockafellar and Uryasev (2002). The Basel III Accord set expected shortfall as the base for capital requirements for the banking sector (Bugár & Ratting, 2016). Value at risk (VaR) was a very popular risk measure that was replaced by expected shortfall after it was shown that it is incoherent.