Reexamining what survey data say about currency risk and irrationality using the cointegrated VAR

This paper reports new evidence of a time-varying risk premium, and against the usual interpretation of irrationality, in survey data for three major currency markets. Using the cointegrated VAR to better focus on the issue of persistence, the deviations from Uncovered Interest Parity are found to be non-stationary implying a time-varying risk premium. Further, the "relationship" between the forecast error and the lagged forward discount, which has been interpreted as implying irrationality, is a spurious regression, being non-stationary at the 1% level. In fact, the forecast error and forward discount do not even appear to share the same order of integration.