Version 6

Release 6.3

  • Performance enhancement for long-run covariance estimators when numba is installed (GH687)

  • Python 3.12 support

  • Compatability with NumPy 2

  • Fixes for future changes in pandas

Release 6.2

  • Fixed a bug that affected forecasting from FIGARCH models (GH606).

  • Added a performance warning when testing for unit roots in large series using a lag-length search with no-max-lag specified.

  • Fixed a bug that affected forecasting from arch.univariate.volatility.FIGARCH models (GH606).

  • Changed the default value of reindex to False so that forecasts will not match the input by default. Set reindex to True if this is required.

  • Made from __future__ import reindex a no-op.

  • Updated notebooks to reflect best practices

Release 6.1

  • Pushed back the adoption of Cython 3 until a later date

  • Fixed a bug that occurred when:

    • Using a AR, HAR or other model with lagged dependent variables; and

    • rescale=True or with data that was automatically rescaled.

Release 6.0

  • Minimum supported Python is 3.9

  • Bumped minimum NumPy, SciPy, pandas, statsmodels and Cython

  • Removed dependence on property-cached

  • Added compatability with Cython 3