Machine Learning in Asset Pricing: Support Vector Regression and the Role of Exchange Rate Risk

Exchange rate fluctuations can influence stock returns through multiple channels, including corporate profitability, capital flow, and investor sentiment. However, traditional asset pricing models, such as the Capital Asset Pricing Model (CAPM), Fama-French three-factor (FF3), and Carhart four-factor (FF4) models, typically overlook exchange rate risk as a priced factor. Furthermore, the relationship between exchange rates and stock returns is often nonlinear and dynamic, highlighting a gap in the asset pricing literature. This study utilizes monthly data from January 2005 to December 2023, covering 844 stocks listed on the HOSE and HNX. Using Support Vector Regression (SVR), we forecast stock returns and construct long-short portfolios to evaluate the role of exchange rate risk and the predictive power of SVR relative to linear models. The novelty of our approach lies in exploiting SVR’s ability to capture nonlinear relationships, enabling us to uncover exchange rate–related anomalies that traditional linear models fail to detect. The empirical findings reveal that the SVR-based long-short portfolio incorporating exchange rate risk yields a statistically significant and positive return. Moreover, regression analyses confirm the presence of an exchange rate- related anomaly, with SVR outperforming OLS in terms of both forecasting accuracy and anomaly detection. These results emphasize the relevance of exchange rate risk in asset pricing and demonstrate the potential of machine learning approaches to uncover complex non-linear relationships and improve return predictability.