In this paper, we study how different choices of loadings affect forecasting in the exponential term structure model proposed by Diebold e Li (2006). The loadings are defined through a specific parameter lambda which controls both the decaying speed of the slope as well as the maximum of the curvature factors. In particular, adopting a database including Brazilian fixed income future contracts (ID future), we analyze four different rules of choices depending on metrics that minimize forecasting errors, for different forecasting horizons. We conclude that the optimal rule changes for different regions of ID future maturities/different forecasting horizons. This fact indicates that the choice of how movements will be parameterized in the Diebold e Li (2006), model should be done with care, tailored for each particular application.
