Evaluate and compare the predictive performance of different GARCH-type models in forecasting weekly volatility, using out-of-sample evaluation techniques.
- 📈 ARCH/GARCH Modeling – Testing for ARCH effects and fitting GARCH, GARCHX (with exogenous variables), and EGARCH models.
- ⚖ Model Selection – Using the Akaike Information Criterion (AIC) to identify the best in-sample fit.
- 🔄 Forecasting Approach – Implementing a rolling window framework to generate dynamic volatility predictions.
- 📊 Forecast Evaluation:
- 📌 Realized Variance Approach – Weekly volatility forecasts were evaluated using realized variance, computed from daily returns.
- 📉 Mean Absolute Error (MAE) for both variance and volatility forecasts.
- 📊 Mincer-Zarnowitz regression to assess forecast optimality.
- 🔍 Diebold-Mariano tests to compare predictive accuracy between models.
📡 Financial time series data (DAX, Bitcoin, EUR/RUB) sourced from Yahoo Finance, with additional market indicators such as the Volatility Index (VIX) used as an exogenous regressor.