In this paper, we investigate a long-standing issue in the international finance literature: namely, how to capture the behaviour of central banks when deciding foreign exchange policies. Essentially, the main empirical problem is that a researcher observes numerous large-scale purchases of foreign currency but a general absence of sales. This asymmetry has motivated the use of heavily dependent parametric models. We take a fresh look at this problem by allowing for a more flexible estimation, robust to various model specifications. Our results indicate that our method outperforms some of the standard models used to date. Hence, our main contribution is to provide policymakers with an improved and readily accessible toolkit to evaluate their actions. To shed some light on this, we estimate policy functions for the cases of Turkey and Colombia and highlight marked differences with the related literature.