Operational risk, uncertainty, and the economy: a smooth transition extreme value approach

We study the link between the distribution of extreme operational losses and the economic
context, a fundamental task to compute adequate risk measures over time. In
particular, we allow for time-varying dependencies due to structural changes, thanks to
a newly-introduced smooth transition Generalized Pareto (GP) regression model. In
this model, the parameters of the GP distribution are related to explanatory variables
through regression functions, which depend themselves on a predictor of structural
changes. Relying on this model, we study the dependence of the monthly loss severity
distribution at UniCredit, over the period 2005-2014. As indicator of structural
changes, we use the VIX, accounting for the general uncertainty on financial markets.
We show that both the goodness-of- fit far in the tail and Value-at-Risk estimates of
the total loss distribution obtained from such models are superior to a set of alternatives.
We also show that in periods of high uncertainty, conditions favorable to a
lax monetary policy are synonym of an increased likelihood of extreme losses. Finally,
we discover evidence of a self-inhibition mechanism, where a high number of losses
in a recent past are indicative of less extreme losses in the future, probably due to
improved monitoring.