Fundamental predictors of price bubbles in precious metals: a machine learning analysis

In this study, we present a two-step method for predicting price bubbles in precious metals, which combines a widely recognized right-tailed unit root test to detect bubbles with various machine learning algorithms to pinpoint the potential predictors of bubble formation and their relative significance. We utilize logistic regression, support vector machines, CART, random forests, extreme gradient boosting, and neural networks algorithms, which are more precise than traditional methods in making predictions and can handle binary classification and regression issues. Our analysis covers monthly prices for gold, silver, palladium, and platinum from 1990M1-2022M10. The study extends the literature by utilizing the Generalized Supremum Augmented Dickey-Fuller test to identify potential price bubbles and analyzing the effect of macroeconomic, financial, and uncertainty factors on the likelihood of bubbles using machine learning algorithms. The findings indicate that macroeconomic factors play a significant role in the formation of price bubbles in precious metals; specifically the consumer confidence index in the USA was a common factor that had a positive impact on the likelihood of bubbles in gold, platinum, and silver. However, the leading factors in the formation of bubbles in palladium were found to be financial variables and uncertainty variables. As predicting bubbles is crucial for regulators and policymakers to take preventive measures against future crises, identifying the key predictors of bubble formation and forecasting them in the early stages is essential.