Using BS-PSD-LDA approach to measure operational risk of Chinese commercial banks

a r t i c l e i n f o The research of operational risk management among Chinese commercial banks is still in its preliminary stage. Operational risk events are rare and data is hard to collect. This leads to very small data samples. Besides, a large number of empirical researches show that the distributions of operational losses are often skewed with fat tails. To address these issues, this paper puts forward a loss distribution approach (LDA) based on bootstrap sampling and piecewise-defined severity distribution (BS-PSD-LDA). The approach divides data samples into two distinct parts (high-frequency low-severity losses and low-frequency high-severity losses), and fits the two parts by lognormal distribution and Generalized Pareto distribution respectively. Using hand-collected samples of 426 operational losses in Chinese commercial banks during 1994–2010, we estimate the magnitude of operational losses using the BS-PSD-LDA method. We show that our method provides a better fit than the traditional parametric methods. Besides, the method using historical simulation of nonparametric method seems to offer a good fit to the sample as well. However, we believe that the BS-PSD-LDA approach offers improvement from the perspective of satisfying risk control requirement of the regulatory authority and ensuring the efficiency of funds' utilization. According to the Basel II Accord, operational risk is defined as the risk of loss resulting from inadequate or failed internal processes, people and systems, or from external events. This definition includes legal risk but excludes strategic and reputational risk (BCBS, 2006). Research on operational risk management in Chinese commercial banks has started only recently. There is no comprehensive database that systematically records operational risk loss events. The low-frequency high-severity feature of operational risk events results in small data samples. It is especially hard to collect operational loss data from individual banks, due to banks' reluctance to disclose such data. As such, given the small sample, traditional distributions no longer provide a good fit to the modeling of operational losses. Besides, empirical studies have shown that the distributions of operational loss are often skewed with fat tails. Therefore, it has long been a challenge to the bank as well as the research community to come up with a model that can accommodate both the small sample, and the " low-frequency high-severity " nature of operational losses. Li (2009) proposes a loss distribution approach based on piecewise-defined severity distribution (PSD-LDA) to accommodate the low-frequency high-severity feature of operational risk, …