Optimization of investments strategies in the life insurance business

This paper focuses on supporting methods used to manage financial cash flows between assets and liabilities for life insurance companies – Asset Liability Management. Asset Liability Management may take its importance especially for products with profit sharing. The purpose of the insurance business is to transfer the risk from policyholders to insurance companies that are obligated to fulfil their commitments even for large and highly unlikely events. Therefore, the process of proper management of assets and liabilities called Asset Liability Management (ALM) is introduced. ALM modelling is designed to optimize financial cash flows to achieve profit and fulfil all commitments, but it faces many technical or theoretical limitations that need to be addressed with an adequate solution for both assets and liabilities. The prediction of assets' cash flows will be provided by a vector autoregressive model with a high focus on automatization. The dynamic version of Nelson-Siegel model will be used to predict interest rates. The challenge in liability valuation is to correctly predict the size and timing of claims events. The practical limitation of the standard valuation techniques is the calculation time because to consider and evaluate all the risks is very time demanding and a high number of scenarios need to be tested. The outdated results may not bring any benefit to the management decisions. The alternative approach for faster liability valuation based on clustering analysis is presented. This approach proves as a good alternative to traditional cash flow modelling with very high accuracy and low calculation time. In general, the ALM model has limited application. The two standard ALM models are presented – static with low computational time but low accuracy and dynamic with high accuracy but high computational time. The main goal of this thesis is to create a new alternative ALM model called the mix model. The mix model combines both static and dynamic models and provides the results with very high accuracy and low computational time. The applications in this thesis are presented on the artificial portfolio. The design of process generating artificial portfolio is also described in this thesis.