Foreign exchange rates with the Taylor rule and VECMs

In this project, we challenge the conventional wisdom on exchange rate predictability with the Taylor rule (Molodtsova & Papell, 2009; Rossi, 2013) by employing the vector error correction model (VECM) when the cointegration (CI) rank of our multivariate model is greater than one and less than full. Even though our approach is quite bounded to the finding of a suitable CI rank, our predictions are quite good when compared to a driftless random walk as a benchmark in the long run, whilst the performance in the short run is not. Notwithstanding we claim that we could also obtain better results had we been able to perform a static forecast for three months ahead rather than one (the latter is the only case admitted by the gretl software).