Accurate forecasting of commodity prices remains a crucial challenge due to inherent market volatility and regime-dependent behaviour. This study examines the predictive performance of two nonlinear time series models, the Threshold Autoregressive (TAR) model and the Markov Switching Model (MSM), in modeling and forecasting the prices of gold and cocoa. These commodities exhibit complex dynamics characterized by abrupt structural breaks and asymmetric responses to economic shocks, features that are inadequately captured by linear models. The TAR model is employed to detect endogenous threshold effects, while the MSM accounts for unobservable regime shifts through a probabilistic framework. Monthly average prices of International Cocoa (US /tonne) and International Gold (US /fine ounce) spanning the period from January 2003 to December 2022 (a 20-year window) were subjected to unit root testing, transformation, and differencing to ensure stationarity prior to modeling. The models’ forecasting accuracy was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Results indicate that both TAR and MSM significantly improve out-of-sample forecasts by capturing both abrupt and smooth nonlinear transitions. Notably, the gold market showed stronger regime-switching dynamics, while cocoa prices exhibited clearer threshold-based behaviour. MSM model outperforms the TAR model in forecasting gold prices, as it records lower values for both MAE and RMSE, indicating higher predictive accuracy. For Cocoa, TAR slightly outperforms MSM in both MAE and RMSE, though the difference is minimal. Thus, both models perform comparably for Cocoa, with a marginal edge for TAR. Model suitability is observed to be commodity-specific. These findings underscore the utility of regime-sensitive models in commodity price forecasting and offer valuable insights for market participants and policy decision-makers operating in volatile economic environments.
Cudjoe et al. (Mon,) studied this question.