Herding in the foreign exchange market

Using a recent and comprehensive data set covering nine of the most actively traded currencies on a monthly basis from 1995 to 2024, this paper explores the presence and potential drivers of herding behaviour in foreign exchange rate forecasts. The dataset features an average of 40–50 forecasters per currency, representing a broader range of currencies, a longer time frame, and a larger cross section of forecasters than is commonly found in the FX herding literature. Our results provide mixed evidence on herding, where the balance tends towards anti-herding conclusions.While some revision-based tests suggest herding when current consensus forecasts are used, this evidence weakens considerably when lagged information is employed. In contrast, forecast-error based tests, Bernhardt et al. statistics, and over-reaction regressions more often point to anti-herding, particularly at longer horizons. Overall, we interpret the findings as suggesting thatdifferences among forecasters are largely attributable to heterogeneous views, noise, or idiosyncratic error rather than systematic convergence toward the consensus. When alternative explanations for expectation formation or revisions are considered, the main findings remain unchanged across a wide range of measures, including different types of uncertainty and FX predictors such as the forward premium, the real exchange rate, and the depreciation rate.