We present theory and evidence that challenges the view that forward premia contain little information regarding subsequent spot rate movements. Using weekly dollar-mark and dollar sterling data, we find that spot and forward exchange rates together are well represented by a vector error correction model; that there exists exactly the number of cointegrating relationships predicted by a simple theoretical framework and that a basis for this cointegrating space is the vector of forward premia. Dynamic forecasts indicate that the information in the forward premia can be used to reduce the root mean squared forecast error for the spot rate (relative to a random walk forecast) by at least 33 percent at a 6-month horizon and by some 50 to 90 percent at a 1year horizon.

We develop a framework to extract information regarding subsequent spot rate movements from the term structure of forward exchange premiums while admitting possible deviations from rationality and the presence of risk premiums. Using weekly dollar-sterling, dollar-mark, and dollar-yen data, the restrictions implied by our framework are not rejected, and spot and forward exchange rates together are well represented by a vector error correction model (VECM). Dynamic out-of-sample forecasts up to one year ahead indicate that the VECM is strikingly superior to a range of alternative forecasts, including a random walk and standard spot-forward regressions.