When ARIMA Is the Wrong Baseline

Baseline models are not trophies; they are guardrails. If a naive seasonal pattern explains most of the variance, your audience deserves to see that first.

Diagnostics matter because models can be polite liars. Residual plots do not replace judgment, but they discipline it. When autocorrelation lingers after differencing, say so — hiding the smell does not help downstream decisions.

We teach learners to document the benchmark alongside the model, not beneath it. Transparency about what simpler tools achieve protects trust when complexity finally earns its keep.