Forecasting commodity prices: A quantile factor approach

This thesis investigates the application of Quantile Factor Models to improve the accuracy of commodity return forecasts, addressing the challenge of non-normal distributions often seen in commodity prices. Commodities typically exhibit skewness and kurtosis, complicating forecasting with traditional models that assume normality. By adopting Quantile Factor Models, this study aimed to capture the distributional nuances of commodity returns across quantiles. Through a comparative analysis with other models, including Autoregressive (AR) models and AR models augmented with Principal Component Analysis (AR+PCA), the findings reveal a performance hierarchy in which the AR+PCA model consistently provides the most accurate forecasts, followed by the AR model alone. Surprisingly, the Quantile Factor Model underperforms, even compared to a basic random walk model, suggesting that while theoretically aligned with skewed distributions, quantile-based approaches face practical limitations in forecasting commodities. The AR+PCA model’s success reflects its ability to capture essential variance patterns without added complexity, enhancing forecast accuracy by isolating the principal drivers of returns. Conversely, the AR model’s strong baseline performance demonstrates the value of simplicity in financial forecasting, especially within the complex data landscape of commodity markets. The Quantile Factor Model’s lower-than-expected performance indicates that while theoretically promising, its practical application may require refinement in quantile selection and parameter tuning to be effective in financial contexts. Overall, this thesis underscores the potential of PCA-augmented models in commodity forecasting while also highlighting limitations in applying Quantile Factor Models as standalone tools. By contrasting advanced quantile models with more conventional approaches, this research offers insights into model selection for commodity price forecasting and provides a basis for future work to enhance the empirical robustness of quantile-based methods in finance.