Model Selection in Time Series Analysis

This lecture covers the process of model selection in time series analysis, focusing on deciding the number of AR and MA terms based on autocorrelation and partial autocorrelation functions. It explains how to estimate candidate models, compare them using information criteria, and run diagnostic checks. The lecture also discusses the challenges of finding the right model order solely based on ACF and PACF. Additionally, it explores diagnostics for assessing the adequacy of ARMA candidate models and the importance of information criteria in comparing multiple models. Practical illustrations and examples are provided, including forecasting with ARMA models and evaluating forecast accuracy. The lecture concludes with a discussion on the limitations of ARIMA modeling and extensions like SARIMA, ARFIMA, and ARCH models.