In January 2005 southern Sweden was struck by the storm Gudrun which caused economic damagage much in excess of SEK 3 billion. Can the size of losses due storms like Gudrun be predicted? Should Gudrun cause insurance companies to change their thinking about risks associated with storm losses?
In a collaboration with the insurance company LFAB we try to answer questions like this. Our metods are based on statistical extreme value theory. They give estimates of the probable maximum loss during future time periods and possibilities to detect changes in frequency and severity of storms, as caused e.g. by climate changes.
Credit derivatives are used to manage default risk. The market for simple instruments, e.g. CDS-s, is growing very rapidly and more complicated instruments such as CDO-s which depend on a basket of bonds are now also liquidly traded. To price the risk associated with the more complicated instruments one needs dynamic models for the dependence between defaults of different firms. We are developing a class of Markov based models for this, together with methods for making the models computationally tractable.
Standard Markowitz-like methods for optimizing portfolios of stocks rest on unrealistic assumptions about the behaviour of stock prices. A further research area is more realistic models for portfolio optimisation. Our methods will be tried out in practice in a collaboration with the Second Swedish National Pension Fund, AP2, and have been presented at a workshop on robust portfolio optimization and the software RoPOX on 21-22 May 2007 in Stockholm.
Efficient management of all these risk simultaneously lies at the heart of the insurance industry. But complexity rapidly increases with the interplay between all of a company's assets and liabilities. For solvency purposes, e.g. as stipulated under pillar one of the Solvency II-accord, full Asset Liability Modeling is needed.
Given an ALM model, it is natural to want to tweak operations parameters to improve risk exposure and thus there is a need to connect the model to an optimization engine. In addition to the research listed above we are developing optimization algorithms suitable for ALM models, including methods to specify the economic environment for the insurance company, the choice of relevant risk measures, and methods to ensure implementability and maintainability.