Josef Perktold

ENH: outlier detection, skewed, zero-inflated data

1

The Templ et al article uses several methods for outlier detection. A bit ad-hoc. uses Box-Cox transformation of data. Methods are not distribution specific. Might be good to get an...

(random find sorting through articles downloaded for robust in 2016) Aelst, Stefan Van, and Gert Willems. “Fast and Robust Bootstrap for Multivariate Inference: The R Package FRB.” Journal of Statistical...

once we have robust covariance computation as in #8129, we can compute linear regression models including IV based on those cov. I don't find right now the reference for using...

ENH: OLS/WLS based on summary statistic, (cov_data)

4

(I don't find an issue, but the idea is old.) This is similar to #3570 but for the specific case of adding a model class for linear regression based on...

ENH: S-estimator and CM-estimator for outlier robust estimation

5

just parking references to read more carefully for implementation I don't find an issue specifically for S-estimator. related #8997 MM-estimator Do we base the estimation on an objective function or...

vague idea I have robust or non-robust to outliers (squared) mahalanobis distance. For elliptical distributions this distance, i.e. radial function, has a given distribution. In the normal case it is...

DOC: PoissonResults is not in the docs

2

PoissonResults is missing in discretemode.rst no SeeAlso link from the Poisson class, e.g. fit strange, PoissonResults existed forever also discretemode.rst base class list at end of file is missing some...

Reference: scaling of scatter matrix to get covariance

9

(parking a reference to computational detail) how do we normalize a scatter matrix so that it is consistent for specific distribution, commonly the normal cov = sigma = c scatter...

lm-test that are based on auxiliary regression can use cov_type to make the specification test robust to (co)variance and other misspecification. see https://github.com/statsmodels/statsmodels/issues/9121#issuecomment-1892389971 first type of cases is where we...

diagnostic: link test - nonlinearity test for many model

10

see Stata 'r.pdf' manual estimate y with `exog_test = (fittedvalues, fittedvalues**2)` reject correct specification if fittedvalues**2 is significant there might be something similar for OLS already, but Stata's link test...