Does the forward premium puzzle disappear over the horizon?

This paper provides the first comprehensive study of the horizon effect in tests of the forward rate unbiasedness hypothesis. It estimates Fama regressions employing 1-month through to 10-year horizon data for the five most heavily traded US dollar currency pairs pre-crisis 1980?2006. In contrast with extant studies, it fully deals with the econometric problems of long horizon regressions by means of a novel heteroskedastic- and autocorrelation-consistent bootstrap. The regression results confirm a clear horizon effect in that the slope coefficient approaches unity as the forward contract maturity is extended. The puzzle disappears at the 3-year horizon and beyond for all currencies.