PCA-BASED EX-ANTE FORECASTING OF SWAP TERM STRUCTURES

In this study, we forecast the term structure of EURIBOR swap rates by means of rolling vector autoregressive (VAR) models. In advance, a principal component analysis (PCA) is adopted to reduce the dimensionality of the term structure. To statistically assess the forecasting performance for particular rates and the level, slope and curvature of the swap term structure, we rely on the Henrikkson-Merton statistic. Economic performance is investigated by means of cash flows implied by alternative trading strategies. Finally, a data-driven, adaptive model selection strategy to "predict the best forecasting model" out of a set of 100 alternative PCA/VAR implementations is shown to outperform forecasting schemes that rely on global homogeneity of the term structure.