Problem Statement

The project aims to analyze the price predictability of key commodities—Gold, Crude Oil, and Coffee—by addressing challenges in time series data, such as non-stationarity, seasonality, and volatility clustering. It seeks to understand interdependencies and enhance forecasting accuracy for global commodity markets.

Approach Used

The study employs statistical models like VAR, ARIMA, SARIMA, and linear regression to capture temporal dependencies, seasonal trends, and macroeconomic influences. Each model's performance is evaluated using metrics like R², RMSE, and F-statistics, with additional focus on understanding the complex relationships among the commodities.

Results

• VAR Model: Explored interdependencies, showing Gold's high sensitivity to Crude Oil and Coffee.
• ARIMA Model: Captured short-term dependencies, with Crude Oil achieving the most accurate predictions.
• SARIMA Model: Addressed seasonality but struggled with Gold's non-linear dynamics.
• Linear Regression: Achieved moderate to high explanatory power (R²: 0.492-0.741) by incorporating macroeconomic factors.

Technology Used

• Models: VAR, ARIMA, SARIMA, and Linear Regression.
• Evaluation Metrics: R², RMSE, F-statistics, and IoU for performance evaluation.
• Techniques: Seasonal adjustment, lag order optimization (AIC), and data preprocessing for stationarity.