Recent empirical studies have established that deviations from the Uncovered Interest Parity (UIP) condition may be different across macroeconomic regimes. We extend this work to account for possible nonlinearities and endogeneity by estimating a Threshold Vector Autoregression (TVAR) model. Using carry trade proxies as in Brunnermeier et al. (2009) alongside a measure of realized exchange rate volatility, we endogenously identify two volatility regimes: low and high. Simulating an incentive to open a carry-trade position through an orthogonal shock to the interest rate differential, we find that carry trade performance varies across different regimes. This suggests that UIP deviations are more pronounced in the low volatility state and non-linearities play a role in explaining the forward bias.