Uncovered interest parity tests and exchange rate expectations

We show that in a representative agent model with a constant risk premium, the uncovered interest parity (UIP) test coefficient can be expressed as a function of the variables and parameters of the prevailing exchange rate expectations mechanism. Taking into account the market microstructure, the robustness of this relationship is confirmed by simulations with a multi-agent model containing a time-varying risk premium. Distributed lag expectations are able to explain the often large negative values for UIP-test-coefficients usually found in empirical studies, while bandwagon expectations are able to explain more recent findings of UIP-test-coefficients larger than one. Regressive expectations generate positive values, both smaller and larger than one. Adaptive expectations are able to generate the whole spectrum of empirical values.