We present novel panel data from Ethiopia including incentivized measures of risk- tolerance paired with detailed data on rainfall. In cross-sectional analysis, we find a positive correlation between rainfall shocks and risk-tolerance. Using fixed-effects panel regressions, however, we find risk-tolerance to decrease following shocks. We use these contradictory findings to revisit the literature on shocks and preferences. That literature implicitly assumes that preferences are uniformly distributed across respondents ex ante. We show that this assumption is violated in our data. This has wide-reaching implications for the possibility of detecting the effect of shocks on preferences in cross-sections, which we discuss.