This paper proposes a sign-narrative VAR approach to identifying financial shocks in currency markets. The approach imposes minimal sign restrictions shared across canonical exchange rate models, leveraging their key insights while remaining robust to misspecification relative to structural models typically used in the literature. To sharpen the identification, sign restrictions are complemented with narrative restrictions anchored on episodes of well documented FX market dysfunction. Focusing on two emerging economies (Brazil and Chile), our estimates suggest that financial shocks account for about one third of UIP fluctuations, and contribute less than 10% to the variance of macro variables including output and inflation. While infrequent, when they do materialize, financial shocks trigger sharp declines in output, suggesting economically meaningful spillovers from frictions in currency markets to the real economy.