Akaike Information Criterion (AIC) compares model fit while penalizing estimated parameters. It favors models that explain observed data well without unnecessary complexity; lower values are preferred. section 7.3.10
Use AIC to support model-comparison decisions:
- Compare competing models fitted to the same observations. section 7.3.10
- Interpret log-likelihood together with parameter count rather than treating fit alone as sufficient. section 7.3.10
- Use BIC or HQIC alongside AIC when a stronger or sample-size-sensitive complexity penalty matters. section 7.3.10
Comparing Holt–Winters and ARIMA
A Holt–Winters model with additive trend and seasonality produced AIC = 198.23435477953498. Its AIC was lower than the corresponding ARIMA value, supporting Holt–Winters as the better-fitting model for this comparison. section 7.4.7
>>> print(f"AIC: {hw_model.aic}")
AIC: 198.23435477953498
>>> print(f"BIC: {hw_model.bic}")
BIC: 223.54401397788672
Inspect the fitted Holt–Winters model's AIC and BIC values directly. section 7.4.7
