Most operational losses occur and are detected nearly on a daily basis. Frequent operational losses are associated with minor errors committed by inexperienced employees or with product defects, credit card frauds, process slowdowns, delays attributed to computer problems, and so on. However, some losses may occur once in 5, 10, or 50 years, such as substantial property damage due to natural disasters or terrorist attacks. Moreover, some losses take place over an extended period of time, but may remain undetected for months or years, such as long-lasting unauthorized trading activities. In either scenario, observed losses arrive on an irregular fashion, with the interarrival times (i.e., the time intervals between the occurrences of events) ranging from several hours to several years or even decades. In this light, it appears appropriate to incorporate the specifics of the arrival process into the model for operational losses, and to model losses of every type as a process that is characterized by a random frequency of events and a random monetary magnitudes of their effect.
An essential prerequisite for developing a solid operational risk model is a systematized mechanism for data recording. In particular, a bank should be consistent in the way it records the date associated with every loss event. Currently, there does not seem to exist a standardized policy on this. In Chapter 4 we described three possible ways in which the date might be recorded. ...