[ENH] Add exogenous variable support to ARIMA

[satwiksps]

[ENH] Add exogenous variable support to ARIMA

Reference Issues

• Implements #3077.

What does this implement ? Explain your changes.

This PR adds full support for exogenous variables (exog) to ARIMA and AutoARIMA.

Summary of enhancements:

• Added "capability:exogenous": True for both ARIMA and AutoARIMA.
• Modified _fit, _predict, and iterative_forecast in ARIMA to:
  • accept exogenous variables,
  • perform an OLS regression of y on exog,
  • run ARIMA on the residual series,
  • include regression contribution back into predictions,
  • support multi-step forecasting with future exog.
• Updated AutoARIMA so that if exog is provided, it is correctly forwarded to the wrapped ARIMA model.

Tests added

• test_arima_with_exog_basic_fit_predict
  Verifies that ARIMA fits and predicts correctly when exogenous variables are supplied.
• test_arima_exog_shape_mismatch_raises
  Ensures shape mismatch errors are raised for incorrect exog dimensions in fit and predict.
• test_arima_iterative_forecast_with_exog
  Tests multi-step forecasting with future exogenous values.
• test_arima_no_exog_backward_compatibility
  Ensures ARIMA without exog behaves exactly as before (backwards compatibility).

All existing ARIMA tests continue to pass, and the new tests confirm correctness and robustness of the exogenous variable support.

Does your contribution introduce a new dependency? If yes, which one?

No new dependencies.

Any other comments?

• All changes were validated through both the existing ARIMA test suite and dedicated exog tests.
• Closes - #3077

PR checklist

For all contributions

• [ ] I've added myself to the list of contributors.
• [x] The PR title starts with [ENH].

For new estimators and functions

• [ ] Not applicable

For developers with write access

• [ ] Not applicable.

[aeon-actions-bot[bot]]

Thank you for contributing to aeon

I have added the following labels to this PR based on the title: [ enhancement ]. I have added the following labels to this PR based on the changes made: [ forecasting ]. Feel free to change these if they do not properly represent the PR.

The Checks tab will show the status of our automated tests. You can click on individual test runs in the tab or "Details" in the panel below to see more information if there is a failure.

If our pre-commit code quality check fails, any trivial fixes will automatically be pushed to your PR unless it is a draft.

Don't hesitate to ask questions on the aeon Slack channel if you have any.

PR CI actions

These checkboxes will add labels to enable/disable CI functionality for this PR. This may not take effect immediately, and a new commit may be required to run the new configuration.

• [ ] Run pre-commit checks for all files
• [ ] Run mypy typecheck tests
• [ ] Run all pytest tests and configurations
• [ ] Run all notebook example tests
• [ ] Run numba-disabled codecov tests
• [ ] Stop automatic pre-commit fixes (always disabled for drafts)
• [ ] Disable numba cache loading
• [ ] Regenerate expected results for testing
• [ ] Push an empty commit to re-run CI checks

[JoaquinAmatRodrigo]

Hi @satwiksps,

Thanks for working on this issue. I'm not involved in the Aeon development, so I'm not very familiar with the codebase. After looking at your modifications, I have a question regarding input alignment: the input data (y and exog) are assumed to be aligned. If the ARIMA model includes differencing, it’s necessary to drop from exog the same number of initial observations as the differencing applied to y; otherwise, the alignment is lost.

Is this already handled in your implementation?

[satwiksps]

Is this already handled in your implementation?

Hello @JoaquinAmatRodrigo I believe the alignment is handled correctly in the current implementation.

• Exogenous variables are used only once, during the initial OLS regression step y - (X @ beta). This produces a residual series that is the same length as y, so alignment is preserved at this stage.

• Differencing is applied after residualization np.diff(residual_series, n=d). At this point, the model works entirely on the differenced residuals and does not use exog anymore, so no trimming or shifting of exog is required after differencing.

• For forecasting (one-step or multi-step), future exog rows are used only to compute the regression contribution [1, future_exog_row] @ beta. This does not depend on differencing and does not require dropping the first d rows.

So I think this implementation preserves correct alignment without needing to manually drop the first d rows of exog. I might be wrong. I would appreciate if you could suggest me if any changes I need to do in my implementation if I am missing something. Thank you.