The European Central Bank has published a working paper examining herding in foreign exchange forecasts and finding no consistent or robust evidence that professional forecasters systematically converge on the consensus view. Using monthly individual forecasts for nine major currencies against the USD from 1995 to 2024, with an average of 40 to 50 forecasters per currency, the paper concludes that the balance of evidence points instead to anti-herding, especially at longer horizons, with differences across forecasters better explained by heterogeneous views, noise and idiosyncratic error. The paper notes that the views expressed are those of the authors and do not necessarily reflect the ECB. Revision-based tests suggested herding when current consensus forecasts were used, with 86% of cases indicating herding, but this fell to 44% when lagged consensus information was used. By contrast, forecast-error tests found neither herding nor anti-herding in 51% of cases, and where the null was rejected the results more often indicated anti-herding. Over-reaction regressions also pointed in that direction, with mostly negative coefficients becoming significant for one of nine currencies at the 1-month horizon, five of nine at the 3-month horizon and all nine at the 6-month horizon. Adding a broad range of uncertainty measures and FX predictors, including the forward premium, real exchange rate and depreciation rate, did not materially change the findings.
European Central Bank2026-06-03
European Central Bank working paper finds no robust herding in foreign exchange forecasts and stronger evidence of anti herding
The European Central Bank published a working paper finding no consistent evidence of herding in professional foreign exchange forecasts for nine major currencies against the USD from 1995 to 2024, and instead identifying stronger evidence of anti-herding, especially at longer horizons. Revision-based tests showed apparent herding when using current consensus forecasts, but this weakened when lagged consensus was used, while forecast-error and over-reaction tests more often indicated anti-herding.