Enhancing the forecasting power of exchange rate models by introducing nonlinearity: Does it work?

It is demonstrated that the forecasting power of the flexible price monetary model of exchange rates can be enhanced by introducing dynamics through the use of a linear error correction specification. However, the introduction of nonlinearity, by using a polynomial in the error correction term, does not lead to any further improvement in forecasting accuracy and may even lead to deterioration. The results provide evidence against the proposition that the Meese–Rogoff puzzle can be explained in terms of failure to account for nonlinearity. It is also shown that the introduction of dynamics boosts the forecasting accuracy (in terms of the magnitude of the forecasting error) of the model relative to the static specification because dynamic specifications involve a random walk component. The empirical results lead to the conclusion that accounting for nonlinearity does not resolve the Meese–Rogoff puzzle.