This paper examines whether country-level macroeconomic, valuation, momentum, positioning, volatility, and financial-market signals can predict future monthly exchange rate returns and generate economic value in a long-short currency strategy. This paper constructs monthly returns for 46 currencies and matches them with country-level predictive signals. After stationarity tests and signal transformation, this paper then uses univariate regression models as an initial screening step and builds country-level multivariate forecasting regression models under two model selection approaches: a strict statistical selection method and a category-protected selection method. The selected signals are then used in an expanding-window forecasting framework to construct long-short portfolios. Results show substantial cross-country heterogeneity: no single signal or common signal set consistently predicts all currencies. However, the selected signals can generate meaningful portfolio economic value. Among all the specifications, the Top/Bottom 2 portfolio under the category-protected approach provides the best overall balance between return, volatility, and drawdown. It also outperforms the passive Equal-Weighted FX benchmark and has low correlation with the global equal-weighted local equity benchmark. Overall, the findings suggest that exchange rate predictability should be evaluated not only through regression statistics, but also through portfolio performance, risk control, and diversification value.
Research Article
Open Access