Which model generates the best predictions on the future spot rate?– a comparison between three different forms of the UIP model

Sammanfattning: There are many empirical studies that reject the forecasting power of the uncovered interest rate parity condition. This might be due to different reasons, of which one often stated is the existence of a risk premium. In this paper three versions of the conventional UIP model are tested in order to investigate which one that generates the best predictions on the future spot rate. One model without a risk premium is used, one with a constant risk premium, and one with a time varying risk premium. The predictions are made out-of-sample and built on data on the Swedish exchange rate against the Danish krone, British pound, Japanese yen and the Norwegian krona, as well as the interest rates in these countries during the time period January 1993 until December 2004. The results indicate that the model with a time varying risk premium outperforms the other two models. Concerning the comparison between the model with no risk premium and the model with a constant risk premium, the result has in general a low significance level that might indicate some lack of credibility, but the former one seems to generate the most accurate predictions.