An Empirical Approach

In this study we forecast the term structure of FIBOR/EURIBOR swap rates by means of recursive vector autoregressive (VAR) models. In advance, a principal components analysis (PCA) is adopted to reduce the dimensionality of the term structure. To evaluate ex–ante forecasting performance for particular short, medium and long term rates and for the level, slope and curvature of the swap term structure, we rely on measures of both statistical and economic performance. Whereas the statistical performance is investigated by means of the Henrikkson–Merton statistic, the economic performance is assessed in terms of cash flows implied by alternative trading strategies. Arguing in favor of local homogeneity of term structure dynamics, we propose a data driven, adaptive model selection strategy to ’predict the best forecasting model’ out of a set of 100 alternative implementations of the PCA/VAR model. This approach is shown to outperform forecasting schemes relying on global homogeneity of the term structure