A new approach to estimation of actively managed component of foreign exchange reserves

Changes in foreign exchange (FX) reserves are difficult to measure in an economically meaningful way because central banks do not decompose reported data into passive and active components. Only the latter should be used when the usefulness of FX reserves in crisis management is assessed or symptoms of currency manipulations are looked for. The applicability of the existing approach to identification of active component of FX reserves is highly limited as it relies on data that are available for a relatively short timespan. To overcome these problems the new approach to estimation of active component of FX reserves is laid out. It makes use of a time-varying coefficient model estimated with Bayesian techniques. The empirical results are obtained for 20 countries over 1995-2017 period. The main finding is that the estimates from the new approach are highly correlated with those from the existing approach, but the timespan of the former is substantially larger than that of the latter. The estimates based on the new approach are cross-checked against the data on FX market interventions of the Czech National Bank. It is demonstrated that the estimates are in general superior to plain changes in FX reserves as a measure of FX interventions and are not worse than those from the existing approach.