We present a macroeconomic market experiment on the financial determination of exchange rates, and consider whether the assumption that belief formation be treated as a classical hypothesis test, which we label inferential expectations, can explain the effect of uncertainty on exchange rates. In a non-stochastic environment, exchange rates closely follow standard predictions. In our stochastic environment, inferential expectations with a low test size α (conservative inferential expectations) predict exchange rates better than rational expectations in ten sessions out of twelve. Belief conservatism appears magnified rather than diminished at the market level, and the degree of belief conservatism seems connected to the failure of uncovered interest rate parity regressions.