Determining ARMA Model Order Based on AIC Criterion

When performing time series analysis in MATLAB, we typically need to determine the appropriate order for ARMA models. This can be achieved using the Akaike Information Criterion (AIC), which serves as a standard for model selection. The AIC criterion penalizes model complexity while maintaining minimal fitting deviation from the data. By comparing AIC values across ARMA models with different orders, we can identify the optimal model order that effectively describes the underlying data patterns. In implementation, MATLAB provides several key functions for this process. The `aic` function calculates the AIC value directly from model estimation results, while `armax` or `arima` functions can be used to estimate ARMA models with specified orders. A typical approach involves iterating through possible combinations of AR and MA orders (p and q values), computing AIC for each combination, and selecting the model with the minimum AIC value. The algorithm essentially balances model fit against complexity, where lower AIC values indicate better trade-offs between accuracy and parsimony. For practical application, one might implement a grid search across reasonable order ranges (e.g., p=0:5, q=0:5), using nested loops to estimate each ARMA(p,q) model and store corresponding AIC values. The optimal order corresponds to the minimum AIC value in the resulting matrix, ensuring the selected model captures essential patterns without overfitting.