Score-Driven Nelson Siegel: Hedging Long-Term Liabilities

Due to its affine structure the Nelson-Siegel model for yield curves can be transformed to a factor model for excess bond returns. Hedging interest rate risk in this framework amounts to eliminating the factor exposure and minimizing the residual risk. Fitting the model directly on excess returns with constant factor loadings leads to large hedging errors caused by substantial and persistent time-variation in the shape parameter of the Nelson-Siegel factor loadings. To capture this variation we develop a Dynamic Conditional Score (DCS) model for the shape parameter. This dynamic model offers superior hedging performance and reduces the hedging error standard deviation by almost 50% during the financial crisis. Much of the improvement is due to the model for the shape parameter with some further reduction achieved by a GARCH model for the residual risk.