Suchergebnisse

We investigate a novel database of 10,217 extreme operational losses from the Italian bank UniCredit, covering a period of 10 years and 7 different event types. Our goal is to shed light on the dependence between the severity distribution of these losses and a set of macroeconomic, financial and firm-specific factors. To do so, we use Generalized Pareto regression techniques, where both the scale and shape parameters are assumed to be functions of these explanatory variables. In this complex distributional regression framework, we perform the selection of the relevant covariates with a state-of-the-art penalized-likelihood estimation procedure relying on $L_{1}$-norm penalty terms of the coefficients. A simulation study indicates that this approach efficiently selects covariates of interest and tackles spurious regression issues encountered when dealing with integrated time series. The results of our empirical analysis have important implications in terms of risk management and regulatory policy. In particular, we found that high Italian unemployment rate and low GDP growth rate in the European Union are associated with smaller probabilities of extreme severities, whereas high values of the VIX and high growth rates of housing prices are associated with more extreme losses. Looking at firm-specific factors, low leverage ratio and high deposit growth rate are associated with a higher likelihood of extreme losses. Lastly, we illustrate the impact of different economic scenarios on the requested capital for operational risk. We find important discrepancies across event types and scenarios. ; Peer reviewed

AbstractResponse functions that link regression predictors to properties of the response distribution are fundamental components in many statistical models. However, the choice of these functions is typically based on the domain of the modeled quantities and is usually not further scrutinized. For example, the exponential response function is often assumed for parameters restricted to be positive, although it implies a multiplicative model, which is not necessarily desirable or adequate. Consequently, applied researchers might face misleading results when relying on such defaults. For parameters restricted to be positive, we propose to construct alternative response functions based on the softplus function. These response functions are differentiable and correspond closely to the identity function for positive values of the regression predictor implying a quasi-additive model. Consequently, the proposed response functions allow for an additive interpretation of the estimated effects by practitioners and can be a better fit in certain data situations. We study the properties of the newly constructed response functions and demonstrate the applicability in the context of count data regression and Bayesian distributional regression. We contrast our approach to the commonly used exponential response function.