Fill gaps from assistant conversations: ARIMA exogenous and simulation docs

[mintlify]

Summary

Fills two StatsForecast documentation gaps that caused repeated unanswered assistant conversations: ARIMA exogenous-regressor mechanics/errors and the missing "Handling large simulations" section.

Changes

• statsforecast/docs/tutorials/simulation.html.mdx: added the "Handling Large Simulations" section that the page outline promised but the body lacked. Covers the 100,000-point warning, why AutoARIMA simulations are slow (models are refit on each StatsForecast.simulate call, including the order search), how to reuse a fitted model via the model-level API, that n_jobs parallelizes across series (not paths), and seed reproducibility.
• statsforecast/docs/how-to-guides/exogenous.html.mdx: added "Access the fitted coefficients" (regression-with-ARIMA-errors formulation, sf.fitted_[i, j].model_['coef'] dict with ex_1, ex_2, intercept/drift keys) and "Troubleshooting" covering xreg is rank deficient (training-only check, includes the implicit trend column, no check on future X_df), the Expected X to have shape error, and the missing-X_df error.

Context

Recurring assistant queries asked about rank-deficient exogenous columns in the forecast horizon, extracting exogenous coefficient contributions, X_df shape errors, and slow AutoARIMA simulations. All statements were verified against the statsforecast source (arima.py, models.py, core.py).

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
nixtlaverse_nixtla_io	🟢 Ready	View Preview	Aug 31, 2026, 4:21 PM

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}