GoldenHour is Data Science's Course project focused on forecasting gold and silver prices using advanced time series models, including ARIMA, SARIMAX, ARIMA-GARCH, and LSTMs. The project investigates the "safe-haven" hypothesis, analyzing how precious metals respond to geopolitical risk and uncertainty.
data/
gold_silver.csv # Main dataset: daily gold/silver prices + GPR index
headers/ # Dataset metadata and documentation
models/
arima-baseline/ # ARIMA baseline results
arima-garch-hybrid/ # ARIMA-GARCH hybrid results
lstm-deep-learning/ # LSTM model and results
sarimax-exogenous/ # SARIMAX with exogenous variables
notebooks/
01_data_exploration.ipynb
02_arima_baseline.ipynb
03_sarimax_exogenous.ipynb
04_arima_garch_hybrid.ipynb
05_lstm_deep_learning.ipynb
06_lstm_walk_forward.ipynb
utility/ # Helper functions (empty)
docs/ # Methodology and documentation (Italian)
- Time Series Forecasting: ARIMA, SARIMAX, ARIMA-GARCH, and LSTM models
- Geopolitical Risk Index: Integrated as exogenous variable (Caldara & Iacoviello GPR)
- Walk-Forward Validation: Realistic backtesting for all models
- Volatility Modeling: ARIMA-GARCH hybrid for financial volatility
- Deep Learning: LSTM with expanding window and multi-step forecasting
- All models use log returns (not raw prices) to ensure stationarity
- Exogenous variables are lagged to prevent data leakage
- Business day frequency is enforced for all time series
- Model selection via AIC/BIC and walk-forward validation
- Metrics: RMSE and MAE on reconstructed prices
- Clone the repository
- Install required Python packages (see notebooks for details)
- Run analysis notebooks in order for full workflow
- Review methodology in
docs/Progetto Serie Temporali Finanziarie_ ARIMA_SARIMAX.md
- Gold/Silver prices: [source documented in headers/]
- Geopolitical Risk Index: policyuncertainty.com
- DataScience-Golddiggers (UnivPM)
MIT License