Herding behavior and loss functions of exchange rate forecasters over interventions and financial crises

•This study examines herding behaviors and loss functions for yen/dollar forecasts.•Interventions amplified herding for a one-month horizon.•The recent financial crises pronounced anti-herding for one-year horizon.•Asymmetry in loss functions differs among forecasters.•Herding behavior and loss functions are time-varying.

This study examines herding behaviors of yen/dollar exchange rate forecasters, focusing on whether their behaviors are time-varying. The empirical results show that interventions amplified herding for a one-month horizon. During the recent financial crises, herding was pronounced for the one-month forecast horizon, and anti-herding, for a one-year horizon. Quantitative Easing 1 (QE1) also saw pronounced anti-herding for one-month and one-year horizons. Their loss functions are also evaluated under a possibly asymmetric loss function. The degree and direction of asymmetry are time-varying, and their variations differ among forecasters. The results are broadly consistent with (anti-) herding and its time-varying feature.