The most recent financial upheavals have cast doubt on the adequacy of someof the conventional quantitative risk management strategies, such as VaR (Valueat Risk), in many common situations. Consequently, there has been an increasingneed for verisimilar financial stress testings, namely simulating and analyzingfinancial portfolios in extreme, albeit rare scenarios. Unlike conventionalrisk management which exploits statistical correlations among financialinstruments, here we focus our analysis on the notion of probabilisticcausation, which is embodied by Suppes-Bayes Causal Networks (SBCNs), SBCNs areprobabilistic graphical models that have many attractive features in terms ofmore accurate causal analysis for generating financial stress scenarios. Inthis paper, we present a novel approach for conducting stress testing offinancial portfolios based on SBCNs in combination with classical machinelearning classification tools. The resulting method is shown to be capable ofcorrectly discovering the causal relationships among financial factors thataffect the portfolios and thus, simulating stress testing scenarios with ahigher accuracy and lower computational complexity than conventional MonteCarlo Simulations.
展开▼
