In their UIP regressions, Huisman et al. (1998. Extreme support for uncovered interest parity, Journal for International Money and Finance 17, 211–228.) focus on extreme forward premia and find much higher coefficients. We show that, for such results, the expectation signal needs to be thicker-tailed than the missing variable. Transaction costs may produce the right sort of bias. It is (i) bounded (i.e. it has no tails at all), (ii) wide (i.e. it may generate betas below 1/2) and (iii) U-distributed, which makes an “extreme” sample quite effective. We derive theoretical and numerical results in the direction of what Huisman et al. observe. We also tighten Fama's moment conditions.