An error correction factor model of term structure slopes in international swap markets

The first two principal components in a vector of term structure slopes from IRS markets in eight major currencies explain above 90% of the fluctuations in the vector of slopes, and each of the eight slopes considered is cointegrated with these two factors. The implied error correction models are shown to be accurate for short-0 and medium-term slope forecasting for the eight currencies, as compared to univariate models, which allows for a drastic reduction of dimensionality, since we just need to use univariate forecasts for the two factors. Adding more factors to the model does not lead to a significant improvement in forecasting performance, while forecasts obtained using just one factor are not as good as those from two-factor error correction models.