[bashtage]

Target 9.1: merge after v9.0.0 is tagged. Closes #887.

Adds bandwidth_method="newey-west" | "andrews" to every kernel covariance estimator
(including the Andrews, Gallant, NeweyWest and ZeroLag aliases) and to
PreWhitenedRecolored, which passes it to its kernel.

• The value is lower case and case-insensitive. It is exposed as a read-only
  bandwidth_method property.
• "newey-west" is the default and is unchanged: the bandwidths and long-run sums are
  bit-identical to main on 12 kernels.
• "andrews" implements Andrews (1991, Section 6, eqs. 6.3-6.4).
  • An AR(1) is fitted to each column: OLS with an intercept, and sigma^2 = RSS/(T-1),
    as R's ar(method="ols") does.
  • The bandwidth is c_k (alpha(q) T)^(1/(2q+1)). arch's existing kernel constants are
    already Andrews' constants (Bartlett 1.1447, Parzen 2.6614, QS 1.3221, Tukey-Hanning
    1.7462), and the derivation reproduces them.
  • Weights are respected. Constant and zero-weight columns are skipped.
  • Complex data are handled as separate real and imaginary columns.
  • An explicit bandwidth overrides the method. force_int and the nobs - 1 cap work
    as before.

Validation: R 4.6.1 with sandwich 3.1.3, bwAndrews(x, kernel=..., approx="AR(1)", weights=w, prewhite=FALSE).

• 6 seeded datasets (univariate and multivariate, T = 50 to 500, nonzero means,
  non-uniform and zero weights) × 4 kernels × center.
• Largest relative difference 2.2e-15. The tests use rtol 1e-10, with the R values stored
  as constants.
• Also checked: one value by hand from the formulas, and all 12 kernels against an
  independent least-squares reference.

Tests:

• arch/tests/covariance/test_andrews_bandwidth.py, 153 tests.
• Validation rows and case-insensitivity checks for every estimator.
• Fast suite 27,325 passed. The new tests pass in random orders and without matplotlib.
  numpydoc lint 0, and the docs build has no new warnings.

For review:

• The TukeyParzen kernel_const is 1.8576, but the formula gives 1.8587 (0.06%). It is
  left unchanged, since changing it would alter the default bandwidth; the test tolerates
  the difference.
• PhillipsPerron could use bandwidth_method="andrews" for its long-run variance. That
  belongs in a separate issue.

Thanks to @ZacKienzle2 for the proposal.

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 99.31034% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.95%. Comparing base (d2c90b4) to head (60e97ab).
⚠️ Report is 6 commits behind head on main.

Files with missing lines	Patch %	Lines
arch/tests/covariance/test_andrews_bandwidth.py	99.14%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##             main     #964    +/-   ##
========================================
  Coverage   98.95%   98.95%            
========================================
  Files         136      137     +1     
  Lines       30043    30334   +291     
  Branches     2331     2345    +14     
========================================
+ Hits        29729    30018   +289     
- Misses        256      257     +1     
- Partials       58       59     +1     

Flag	Coverage Δ
adder	98.90% <99.31%> (+<0.01%)	⬆️
subtractor	98.90% <99.31%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.