Causal analysis of operational risk for deriving effective key risk indicators

Abstract

Key risk indicators (KRIs) are intended to track operational risk exposure and provide early indications of potential severe losses. Guidance on establishing the most effective KRIs for financial institutions is, however, limited and, as a result, KRIs are typically derived from an institution’s available metrics, often leading institutions to compensate for a lack of effective KRIs by increasing the number of KRIs monitored. Strengthening the ability to identify and evaluate KRIs’ effectiveness could increase the value of each KRI, further reducing the number of KRIs necessary and increasing the overall value of institutions’ KRI framework. This paper proposes a theoretical foundation and method for identifying and evaluating effective KRIs. The proposed solution originates from research on causal analysis of operational risk, particularly using Bayesian networks. It was found that high-frequency and tail events can be related to a shared set of causes which can be exploited for the identification and evaluation of two categories of KRIs: (1) shared causes that constitute major risk drivers; and (2) high-frequency events providing a strong indication of changes in exposure to low-frequency, high-severity events. Applying the suggested method, financial institutions can map and evaluate current and potential KRIs, ensuring reliable monitoring of the operational risk exposure level.