Dynamic Correlation Analysis between Bitcoin and Platinum Group Metals (Platinum and Rhodium)

Abstract

The extreme price volatility of Bitcoin frequently prevents its widespread adoption. The persistent "Digital Gold" narrative often dominates its price analysis, largely ignoring the predictive value of strategic industrial commodities like Platinum Group Metals. This study aims to investigate whether integrating industrial metals specifically platinum and rhodium enhances the short-term forecasting accuracy of Bitcoin prices. Utilizing high-frequency 5-minute interval data over 729 days, this research applies a comparative quantitative approach using univariate and multivariate Long Short-Term Memory (LSTM) deep learning architectures. Results demonstrate the multivariate LSTM model achieves highly accurate forecasting, recording a Mean Absolute Percentage Error (MAPE) of 3.95% and a Root Mean Squared Error (RMSE) of 0.0598. Compared to the univariate baseline model (MAPE of 5.14%, RMSE of 0.0725), the multivariate approach demonstrates a notable decrease in error rates. This improvement suggests platinum and rhodium price movements contain useful informational value for Bitcoin forecasting, rather than mere random noise. Specifically, rhodium demonstrates strong predictive relevance for Bitcoin market movements. In conclusion, while not strictly proving causal structural integration, these findings highlight Bitcoin's sensitivity to the global real-sector economic cycle. Practically, these findings suggest investors can refine short-horizon forecasting and mitigate risk by monitoring industrial commodity prices. Given persistent nominal offset deviations, future research should prioritize explicit connectedness testing (e.g., lead-lag analysis) and develop a hybrid model incorporating Natural Language Processing (NLP) for news sentiment analysis.