Exchange Rate Volatility Spillovers to Gold Prices In Sri Lanka: Evidence from a Hybrid Econometric–Machine Learning Framework

This study examines spillovers of means and volatilities between the exchange rates and the gold price in Sri Lanka. Both linear and nonlinear spillover channels are explored under a hybrid econometric-machine learning framework. The data consists of daily spot price of local gold and USD/LKR exchange rate over the period 2018-2025. In terms of methodology, the empirical strategy includes Granger causality test, HAC regression of volatility, various GARCH models, Random Forest estimator, and time series cross-validation. It was found that exchange rates strongly influence gold prices and provide predictive signals for future movements. The volatility regression revealed a positive relationship between the exchange-rate volatility and the gold-market volatility. Nonetheless, the GARCH model showed that the relationship is not significant in dynamic and conditional settings, even after controlling for volatility persistence, suggesting little evidence of structural volatility transmission. Volatility persistence in the Sri Lankan gold market is quite high (α + β = 0.947). Random Forest estimator has a very good fit, explaining 95.89% of variation in volatility of gold market. Lagged gold and exchange-rate volatility proved the most informative predictors in the regression. The forecasting model was validated using time-series cross-validation. This study emphasises the importance of distinguishing between structural spillover channel and predictability of financial markets in risk transmission process. The study provides original evidence based on combining econometric and machine learning methods in an emerging economy.