Forecasting Currency Volatility in Global Markets Using Transformer Models: Implications for U.S. Trade and Investment Strategies

Abstract

Currency volatility has become one of the most significant challenges for global financial stability and U.S. trade competitiveness. Traditional econometric approaches, such as GARCH and VAR models, often fail to capture the nonlinear dependencies and structural breaks inherent in currency markets. This paper proposes the use of transformer-based deep learning models for forecasting short- and medium-term exchange rate volatility across major and emerging market currencies. By leveraging self-attention mechanisms, transformers can model long-range dependencies in high-frequency financial data, capturing hidden structures often overlooked by conventional models. Empirical analysis demonstrates that transformer models outperform GARCH, LSTM, and GRU baselines in predictive accuracy and volatility clustering detection. Furthermore, the study evaluates the strategic implications of currency volatility forecasts for U.S. trade policy, hedging strategies, and foreign investment decisions. Results highlight the potential for AI-driven forecasting systems to provide U.S. firms, investors, and policymakers with actionable insights for risk management, portfolio allocation, and international trade strategy.