An alternative to SMA: Using through the cycle loss data to propose a ‘hourglass’ solution

The losses suffered across the industry in the aftermath of the financial crisis significantly undermined the credibility of operational risk modelling techniques and have led to the Basel Committee proposing the removal of Advanced Measurement Approach (AMA) models from Pillar 1 capital. This paper analyses systematically financial and loss information for the Global Systemically Important Banks (G-SIBs) that are in the public domain across economic cycles to answer three fundamental questions: (1) Why have operational risks proved so difficult to model? The analysis of historical loss data illustrates that some of the most significant operational risks are sensitive to economic cycles, and an explanation is provided for this observation that utilises real options theory. Further analysis of the impacts of the financial crisis demonstrates that losses across the three main risk types came in overlapping waves; that is, market risk losses peaked first, then credit risk and finally operational risk. The paper quantifies the varying lags between discovery and settlement of different categories of operational risks, and observes that a firm’s current loss profile may be more reflective of its previous cultures and controls. All of these characteristics of operational risk make it a difficult risk to model effectively. (2) Will Basel III’s new Standardised Approach (aka SMA) be any better than AMA? The paper identifies that AMA models were further undermined by an initial miscalibration of operational risk capital by the Basel Committee; the lack of standardisation in modelling techniques across the industry and the belief that operational risk could be modelled using insurance industry techniques. In striving for a methodology that delivers both greater simplicity and consistency for Pillar 1 capital, the paper argues that Basel III’s Standardised Approach does not address the various characteristics of operational risk described in the first section of the paper. Analysis is also provided, which calls into doubt the assumed relationship between size and a firm’s operational risk profile. (3) How could operational risk be modelled more effectively? Finally, the paper proposes an end to one size fits all approaches to modelling operational risk and instead splits the risk into three categories: (1) idiosyncratic and sudden losses; (2) idiosyncratic and lagging losses; and (3) cyclical and lagging losses. For each of these three categories modelling approaches are proposed that are appropriate for their specific characteristics. For idiosyncratic and sudden losses, the existing and proposed regulatory approaches are reasonable. While for idiosyncratic and lagging losses firms should primarily establish accounting provisions rather than set aside capital, reflecting the extended period between discovery and settlement. Finally, for cyclical and lagging losses Pillar 2B capital should be allocated due to the correlation of these risks with market and credit risk losses. The paper proposes techniques for conducting stress testing that are consistent with a real option theory explanation of correlation.