In this paper, we revisit the predictive content of interest rates for daily exchange rate returns. The novelty of our approach is to take into account dependencies of higher orders by allowing for a time-varying asymmetry in the distribution of exchange rates. Using data on USD/EUR currency pair over the period 1999-2019, we find the dynamic asymmetry component to be significant and driven by interest rate differentials, but also by general uncertainty and past unexpected shocks. In line with recent currency crash theories, our study suggests that the larger the difference between interest rates, the more likely the high yield currency is to appreciate but also to experience currency crashes. To assess the economic significance of our results, we introduce a directional forecasting approach derived from our model. We show that a trading rule based on these forecasts provides better in-sample and out-of-sample economic performance compared to benchmark models.