We apply multivariate extreme value theory to test whether extreme exchange rate depreciations and devaluations are associated with extreme movements in lagged macro economic and financial variables. We find that nearly all fundamental variables have no relation with extreme exchange rate returns, except for the real interest rate. The estimated probability of a currency crisis occurring within twelve months of a positive extreme value of the domestic real interest rate is equal to 30%. Our findings can explain why existing early warning systems for currency crises perform poorly out of sample.