Log Message:
py-statsmodels: updated to 0.14.0

Release 0.14.0

The Highlights
==============

New cross-sectional models and extensions to models
---------------------------------------------------

Treatment Effect
~~~~~~~~~~~~~~~~
:class:`~statsmodels.treatment.TreatmentEffect` estimates treatment effect
for a binary treatment and potential outcome for a continuous outcome variable
using 5 different methods, ipw, ra, aipw, aipw-wls, ipw-ra.
Standard errors and inference are based on the joint GMM representation of
selection or treatment model, outcome model and effect functions.

Hurdle and Truncated Count Regression
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
:class:`statsmodels.discrete.truncated_model.HurdleCountModel` implements
hurdle models for count data with either Poisson or NegativeBinomialP as
submodels.
Three left truncated models used for zero truncation are available,
:class:`statsmodels.discrete.truncated_model.TruncatedLFPoisson`,
:class:`statsmodels.discrete.truncated_model.TruncatedLFNegativeBinomialP`
and
:class:`statsmodels.discrete.truncated_model.TruncatedLFGeneralizedPoisson`.
Models for right censoring at one are implemented but only as support for
the hurdle models.

Extended postestimation methods for models
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Results methods for post-estimation have been added or extended.

``get_distribution`` returns a scipy or scipy compatible distribution instance
with parameters based on the estimated model. This is available for
GLM, discrete models and BetaModel.

``get_prediction`` returns predicted statistics including inferential
statistics, standard errors and confidence intervals. The ``which`` keyword
selects which statistic is predicted. Inference for statistics that are
nonlinear in the estimated parameters are based on the delta-method for
standard errors.

``get_diagnostic`` returns a Diagnostic class with additional specification
statistics, tests and plots. Currently only available for count models.

``get_influence`` returns a class with outlier and influence diagnostics.
(This was mostly added in previous releases.)

``score_test`` makes score (LM) test available as alternative to Wald tests.
This is currently available for GLM and some discrete models. The score tests
can optionally be robust to misspecification similar to ``cov_type`` for wald
tests.

Stats
~~~~~

Hypothesis tests, confidence intervals and other inferential statistics are
now available for one and two sample Poisson rates.

Distributions
~~~~~~~~~~~~~

Methods of Archimedean copulas have been extended to multivariate copulas with
dimension larger than 2. The ``pdf`` method of Frank and Gumbel has been
extended only to dimensions 3 and 4.

New class ECDFDiscrete for empirical distribution function when observations
are not unique as in discrete distributions.

Multiseason STL decomposition (MSTL)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The existing :class:`~statsmodels.tsa.seasonal.STL` class has been extended to \ 
handle multiple seasonal
components in :class:`~statsmodels.tsa.seasonal.MSTL`.