Prediction Standard Errors #554

Closed
jseabold opened this Issue Oct 30, 2012 · 6 comments

Projects

None yet

2 participants

@jseabold
Member

Move wls_prediction_std somewhere so it can be attached to results instance. Perhaps a get_prediction method similar to get_influence.

@josef-pkt
Member

bump

They are also part of the influence summary_table and also checked against SAS, but only for OLS.
tests for WLS case is still missing

@josef-pkt
Member

see https://groups.google.com/forum/?fromgroups#!searchin/pystatsmodels/get_predict/pystatsmodels/6obOiz-llO4/UaxrSp-TKiEJ

for discussion and related tickets

get_predictresults

#719 linear_model
#554
#59 GLS
#459 discrete
related for fittedvalues
#411

@josef-pkt
Member

see http://stackoverflow.com/questions/17559408/confidence-and-prediction-intervals-with-statsmodels/17560456

we have mean prediction confidence intervals for OLS hidden in outlier influence

@josef-pkt
Member

I finally have a Stata WLS example, but Stata bails out for forecast and residual standard error

. predict forecast_std, stdf
not possible with weighted data
r(135);

. predict resid_std, stdr
not possible with weighted data
r(135);

It has standard errors for the prediction of the mean predict predict_std, stdp

@josef-pkt
Member

finally a reference

"Prediction and Prediction Intervals with Heteroskedasticity"
Wooldridge Introductory Econometrics p 292
use variance of residual is correct, but is not exact if the variance function is estimated. It ignores the variance coming from the estimated variance function parameters.

@jseabold
Member

Closing in favor of #719.

@jseabold jseabold closed this Sep 26, 2014
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment