Abstract We present incentivised panel data measuring risk preferences of subsistence farmers from across Ethiopia and pair them with rainfall data. We use these data to test the hypothesis that risk preferences may adapt to the environment of the decision maker. We find that rainfall shocks decrease risk tolerance for the same individuals over time. We also find that historical rainfall characteristics and geographical features can explain 40% of the variation in preferences across individuals. The time-changing effects are perfectly aligned with the geographical effects we document, painting a unified and highly coherent picture. This provides the first real-world evidence that preferences may systematically adapt to the environment of the decision maker.