We demonstrate how the EVT‐based signalling approach for currency crises can be applied to an individual country with a small sample size. Using Thai historical data, first, we study the tail characteristics of the distributions of two Thai baht instability measures and 21 economic fundamentals. Then, we test asymptotic dependence between the currency instability measures and lagged economic fundamentals. Empirically, we find that the distributions of both currency instability measures and economic variables are heavy tailed. Assuming a normal distribution for the variables tends to underestimate the probability of extreme events. Furthermore, most of the economic variables which are usually used as signalling indicators for currency crises are asymptotically independent of the currency instability measures. Signals issued by these variables are thus not reliable. Nevertheless, the non‐parametric EVT approach facilitates the selection of economic indicators with credible signals and high crisis prediction success.