Demonstration of ARMA
Models
By,
Dr. Mehfooza. M, AOP/VIT, Vellore
Conclusion
• Stationarity First: Always verify mean/variance stability.
• ACF/PACF Patterns:
• ARMA(1,1) suggested by:
• ACF decay + PACF cutoff (mixed signatures)
• Least Squares Estimation:
• Minimize RSS via recursive residuals (manual grid search).
• Model Selection:
• Trade-off: Complexity (ARMA) vs. Simplicity (AR/MA).
• Forecasting:
• ARMA(1,1) captures both past values and shocks.
Conclusion