Many life insurance products include options and guarantees which make the liability of the insurance company be linked to market risk factors. The replicating portfolio method is a simulation approach for the market risk capital calculation where the liabilities are represented by a pool of financial assets. Due to multicollinearity in high-dimensional data sets used in replication, instability problems often occur in the optimization procedure, where the distance between the values of liabilities and financial instruments in a set of calibration scenarios in minimized. In this Master's thesis we discuss possible techniques for reduction of the universe of instruments, such that only the most relevant financial instruments are selected. At the first stage of the selection we introduce the following approaches applied on the sensitivity scenario data: well conditioned basis, c-means clustering and portfolio with minimal 1-norm. Further we apply stepwise regression techniques on the calibration scenario data of the selected subset to determine the final replicating portfolio. We draw a comparison of various selection methods based on key characteristics used in validation of replicating portfolios and we make our suggestions on the optimal approach.      «

Many life insurance products include options and guarantees which make the liability of the insurance company be linked to market risk factors. The replicating portfolio method is a simulation approach for the market risk capital calculation where the liabilities are represented by a pool of financial assets. Due to multicollinearity in high-dimensional data sets used in replication, instability problems often occur in the optimization procedure, where the distance between the values of liabilit...     »