Is the forward bias economically small? Evidence from European rates

For the purpose of testing uncovered interest parity (UIP), rates of European currencies against the Mark offer a distinct advantage: the admissible band of the Exchange Rate Mechanism (ERM) induces statistically significant mean-reversion in weekly rates. Thus, unlike for freely floating rates, there is an expectation signal that has non-trivial variation and is sufficiently traceable for research purposes. When running the standard regression tests of the unbiased-expectations hypothesis at the one-week horizon, we nevertheless obtain essentially zero coefficients for intra-EMS exchange rates (and the familiar negative coefficients for extra-EMS rates). Even more puzzlingly, lagged exchange-rate changes remain significant when added to the regression, a feature that seems hard to explain as a missing-variable effect. The deviation from UIP is significant not just statistically but also economically: trading-rule tests reveal that for sufficiently large filters the average profit per trade exceeds transaction costs, and that cumulative gains can be quite impressive. The size of the profits and the patterns from buy versus sell decisions also allow us to reject the hypotheses of either a risk premium or peso issues about realignments as sufficient explanations.