The results reported in Table 3 suggest that fitting heavier tailed distributions, such as Pareto and Weibull, results in lower values of the GOF statistics, which leads to acceptance of the null for practically all loss types and all criteria of the goodness of fit, viewed from the high p-values. Since the analysis of operational losses deals with estimating the operational Value-at-Risk (VaR), it is reasonable to determine the ultimate best fit on the basis of the ADup and AD2 up statistics, introduced in this paper. As can be seen from Table 3, these proposed measures suggest a much better fit of the heavier-tailed distributions.
Moreover, for the Pareto distribution, while the statistics that focus on the center of the data (Kolmogorov-Smirnov, Kuiper, Cram´er-von Mises) do not show a good fit, the ADup and AD2
up statistics indicate that the fit in the upper tail is very good. It should be noted that the Pareto and Weibull distributions very often suggest a superior fit in the upper tail to the Lognormal distribution. Yet is it the Lognormal distribution that was suggested in 2001 by the Basel Committee, BIS (2001).