Cointegration and Long-Horizon Forecasting (2025)
2 days ago
- #multivariate analysis
- #cointegration
- #forecast accuracy
- At long horizons, ignoring cointegration does not harm forecast accuracy under standard multivariate measures.
- Univariate Box-Jenkins forecasts perform equally well as cointegration-based methods in this context.
- Standard forecast accuracy measures fail to account for preserving cointegrating relationships among variables.
- The authors propose alternative measures that explicitly value the maintenance of cointegrating relationships.