Study On Rare Event Simulation Method And Operational Risk Management

In recent decades,the rapid development of the financial industry in the process,the operational risk events show high characteristics.Operational risk arises at the beginning of the development of the banking industry,but the relevant theoretical and empirical research is the last few decades.With the increasing impact of the loss of domestic operational risk,financial institutions are increasingly aware of the seriousness of operational risk,but the related research is not much,mostly stay in the policy discussion level,the risk perception and analysis are more emotional.In the empirical study of operational risk,Most of the related researches at home and abroad use the method proposed by the Basel ? to estimate the operational risk capital,The rare event simulation method has a great advantage in the estimation of operational risk loss,and more and more simulation algorithms are applied to the estimation of risk loss.In this paper,we mainly focus on the occurrence of risk events with extremely low probability but great loss.We hope to estimate the tail risk probability of the operational risk through rare event simulation.For the severity of losses with thick tails,the traditional exponential measure under the importance sampling algorithm can not be implemented,but the mixed distribution method can avoid the problem that the variance of estimating is not convergent and easy to implement.Based on the variance minimization,the auxiliary sampling distribution is determined by the cross entropy criterion,and the auxiliary sampling distribution parameters are solved iteratively using the EM algorithm in the classical statistical algorithm.In order to analyze the effect of these variables and verify the stability of the algorithm,we also set these variables in the update iteration of the parameters.The results show that the thresholds of the loss degree and the number of events have an influence on the auxiliary sampling distribution.In this paper,we give the estimation under the cross-entropy of other algorithms,and compare the simulation results.Finally,we estimate the tail probability of the total loss distribution according to the actual loss data.The results show that this algorithm outperforms other cross-Entropy algorithm.