Cumperayot P.Gulati V.Kouwenberg R.Mahidol University2025-12-072025-12-072025-01-01Journal of Economic Surveys (2025)09500804https://repository.li.mahidol.ac.th/handle/123456789/113421Economic fundamental models often perform poorly in exchange rate forecasting. Theory and empirical evidence suggest that one likely reason is that the relation between the exchange rate and fundamentals is highly unstable, with frequent changes in the best predictors. Since 1987, researchers have applied new statistical methods to better capture the time-varying relations, applying models with time-varying parameters, forecast combinations with dynamic weights, Bayesian model averaging, automated model selection criteria, and machine learning techniques. In this review, we synthesize the results of these studies, highlight the forecasting techniques with the most promising performance, and provide recommendations for researchers and market participants.Economics, Econometrics and FinanceArticleSCOPUS10.1111/joes.700432-s2.0-10502339733214676419