ARMA Time-Series Modeling with AIC, BIC, and Ljung–Box Diagnostics

Summary

This article introduces the autoregressive moving-average model as a combination of AR terms, which use past observations, and MA terms, which represent past shocks. It describes choosing the orders p and q with autocorrelation and partial autocorrelation plots, then comparing candidate models with AIC or BIC. Both criteria balance fit against model complexity; BIC penalizes added parameters more strongly and may favor a simpler specification. The article also explains the Ljung–Box test as a check for residual autocorrelation, comparing a statistic based on lagged correlations with a chi-squared reference distribution.

The discussion relates ARMA to financial and economic series, including prices, rates, and exchange rates, and notes its ability to represent serial dependence and random shocks. It does not provide an empirical example or fitted model. The text cautions that information criteria do not guarantee an optimal model and recommends considering stability and residual dependence. ARMA also does not capture volatility clustering, for which ARCH or GARCH models may be more appropriate. The Ljung–Box test is presented as a diagnostic, not a substitute for model selection.

Key ideas

• ARMA combines autoregressive terms based on past observations with moving-average terms based on past errors.
• AIC and BIC compare candidate models by balancing fit and complexity, with BIC applying a stronger complexity penalty.
• Autocorrelation and partial autocorrelation plots can help guide choices of the AR and MA orders.
• The Ljung–Box test checks for remaining serial correlation, including in model residuals.
• ARMA does not model volatility clustering, and information criteria alone do not establish that a model is adequate.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.