Predicting Operational Loss Exposure Using Past Losses

Abstract

Operational risk models, such as the loss distribution approach, frequently use past internal losses to forecast operational loss exposure. However, the ability of past losses to predict exposure, particularly tail exposure, has not been thoroughly examined in the literature. In this paper, we test whether simple metrics derived from past loss experience are predictive of future tail operational loss exposure using quantile regression. We find evidence that past losses are predictive of future exposure, particularly metrics related to loss frequency.

Benchmarking operational risk stress testing models

“…The correlation of BHC stress loss projections to median, mean, and maximum nine-consecutive-quarter loss frequency is meaningfully higher than the correlation observed to median, mean, and minimum nineconsecutive-quarter total losses. So, just like Curti and Migueis (2017) found that historical loss frequency is a better predictor of future exposure than historical loss totals, BHCs' own stress loss projections tie closer to their historical loss frequency than to their historical loss totals.…”

“…To understand and compare banks' operational risk loss exposure, loss history is considered next. Past loss experience is a reasonable proxy for operational risk loss exposure going forward because past losses relate to the quality of risk controls and the riskiness of the business environment (Curti and Migueis 2017). However, caution should be applied when making comparisons across firms based on internal loss benchmarks because of their disparate data collection history and quality.…”

“…Among these, asset size and loss frequency are most correlated with BHC stress loss projections. In addition, Curti and Migueis (2017) showed that tail exposure is best predicted when loss history and size metrics are combined. For these reasons, we also produce benchmarks for BHC stress loss projections based on the quantile regressions where total assets and average loss frequency are used together as explanatory variables.…”

“…These benchmarks are simple and can be calculated with publicly available information. Also, previous research has documented the relation between scale of activity and operational loss exposure (Chernobai et al 2011, Curti and Migueis 2017. Thus, these size and business volume metrics are our starting point to benchmark banks' operational loss projections.…”

“…Finally, we present benchmarks based on both a financial statement metric (total assets) and a historical loss metric (average nine-consecutive-quarter loss frequency). Curti and Migueis (2017) showed that loss frequency is the historical loss metric that best predicts tail operational loss exposure and that prediction accuracy is maximized when loss frequency is combined with controls for firm size. Thus, following that paper, we generate benchmarks for multiple percentiles by using average loss frequency together with asset size within a quantile regression framework.…”