The accuracy of Value-at-Risk estimates - The effect of the amount of historical data used

Sammanfattning: Value-at-Risk(VaR) has become the leading risk measurement technique in finance. It's important both from a regulatory and internal perspective that the calculated VaR measures are accurate and therefore they should be backtested with appropriate methods. This thesis reviews the effect that the number of historical returns used to calculate VaR measures have on the accuracy on the measurements. Two methods are considered, the parametric normal method and the historical simulation method. The accuracy is reviewed by backtesting, primarily by the Kupiec test.The thesis shows both that it is important to choose an appropriate time window and to adapt the time window used to the method and asset class used as the effect is very different between these.