The commercial risk in time-series modelling is not fitting an ARMA model; it is mistaking a noisy in-sample pattern for a reliable forecast. Box–Jenkins is a loop that forces every candidate to earn its place: identify a parsimonious model, estimate it, then ask whether residual dependence remains.
Learning objectives. By the end of this note you should be able to:
- execute the identify–estimate–diagnose loop;
- compare models using likelihood, AIC, and BIC;
- use Ljung–Box residual diagnostics;
- separate an adequate statistical model from a useful trading forecast.
This lesson combines the ACF and PACF, the ARMA model classes, and their stationarity and invertibility restrictions into an auditable workflow.
In this lesson
- The modelling loop
- Identification begins with the data transformation
- Parsimony is a risk control
- Example 1 (short-rate model selection)
- The model acceptance record
Before you start
- Stationarity, White Noise, and the ACF/PACF
- AR, MA, and ARMA Models: Representation and the Wold Decomposition
- Derivation — Stationarity and Invertibility Conditions for ARMA
The full lesson, with worked examples and exercise solutions, is for members.
Read the full lessonNew to SelfMadeQuant.com? See membership options or browse the Econometrics and Financial Time Series syllabus.