Estimation of operational risks using non-parametric approaches with an application to US business losses

Following the recent global financial crisis, many banks and other businesses in the industrialized countries incurred notably heavy losses. As a consequence, reliable estimation of operational risk (OR) is becoming increasingly important to all internationally active banks and other financial institutions. The OR is the unexpected loss, which is the difference between the 99:9 per cent quantile and the mean of the loss distribution. This paper adapts non-parametric methods based on heavy-tailed distributions and constructs point and 95 per cent confidence interval (CI) estimates for ORs. The main advantage of these nonparametric methods is that there are no assumptions made about the shape of loss distributions and that data determines their shapes, providing robust estimates for ORs. Employing these methods, we construct point as well as interval estimates for ORs for US businesses. The noteworthy observation is that the CIs are asymmetric with huge upper bounds, highlighting the extent of uncertainties associated with the point estimates of ORs. The estimates of expected shortfalls lie within these intervals. The nonparametric methods introduced in this paper will have much wider applications, for example, in estimating another popular measure of risk, credit risk.