Studies

This dissertation investigates the implications of using inappropriate distributions when modelling data sets with extreme data. It has been found that in practice it has been assumed that financial data follow a normal and T Distribution even in cases where these assumptions are inappropriate. Such assumptions when modelling extreme data can lead to gross understatements in risk estimates. The more suitable approach of EVT is introduced as a more prudent approach to model extreme risks. The research was facilitated by models built in EXCEL. Risk estimates derived from a Danish data set of insurance losses under the Normal, T and Generalised Pareto Distribution (EVT) were estimated and compared to determine the degree of error in making wrong distributions assumptions in risk modelling. The dissertation also discusses the usefulness of EVT in the context of regulatory capital charges and explores the limitations of the EVT approach through sensitivity testing. The Findings of the research highlight that financial institutions are set to incur significant understatements in risk estimates if traditional Normal and T distributions are used as the basis of modelling data with extremes. EVT is considered a necessary complement to existing internal and regulatory risk measurement processes