The Rolling Window Method: Precisions of Financial Forecasting

Sammanfattning: In this thesis we set out to study the prediction accuracy of statistical quantities related to portfolio analysis and risk management implied by a given set of historical data. The considered forecasting procedure rely on rolling-window estimates over varying horizons where the resulting empirical return distributions can be considered the corresponding stationary distributions. By using scenarios generated from a joint interest rate-equity framework the rolling-window method allows to, empirically, study the uncertainty of return statistics as well as risk measures related to market risk. The study shows that, given the chosen models, the method is valid in predicting future statistical quantities related to portfolio return of up to one year. For risk measures, the forecasting uncertainty is found to be too significant and highlights the difficulty in foreseeing extremities of future market movements.