Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

newey-west different than R? #1655

Closed
jseabold opened this issue May 3, 2014 · 4 comments

Comments

Projects
None yet
3 participants
@jseabold
Copy link
Member

commented May 3, 2014

I haven't looked at (or thought about) anything but comparing the answers.

https://stackoverflow.com/questions/23420454/newey-west-standard-errors-for-ols-in-python

@josef-pkt

This comment has been minimized.

Copy link
Member

commented May 4, 2014

I didn't compare with R, only with Stata for the unit tests

new = results.get_robustcov_results(cov_type='HAC',maxlags=1, use_correction=True, use_t=True)
replicates the pandas results (except we have alpha=0.1 by default for confint)

I don't have any idea about the difference to R, I never checked it. IIRC, I never tried to figure out how to get sandwiches in R. Difference to R in HAC bse of the constant is relatively large.
Also I've never seen a special treatment of the constant in any HAC paper.

Unit tests are mostly against Stata.

Unit tests for hac against Gretl use
cov_hac = sw.cov_hac_simple(res, nlags=4, use_correction=False)
at decimal=5

@kdiether

This comment has been minimized.

Copy link

commented May 4, 2014

It's the same; I went and checked (not that I know R at all); R has what I would consider a strange default here. It looks like you need to pass the following to get the same as statsmodels: prewhite=FALSE

library(sandwich)
library(lmtest)

a <- matrix(c(1,3,5,7,4,5,6,4,7,8,9))
b <- matrix(c(3,5,6,2,4,6,7,8,7,8,9))

temp.lm = lm(a ~ b)

temp.summ <- summary(temp.lm)
temp.summ$coefficients <- unclass(coeftest(temp.lm, 
               vcov. = NeweyWest(temp.lm,lag=1,prewhite=FALSE)))

print (temp.summ$coefficients)

Output:

             Estimate Std. Error   t value  Pr(>|t|)
(Intercept) 2.0576208  2.6605060 0.7733945 0.4591196
b           0.5594796  0.4007965 1.3959193 0.1962142
attr(,"method")
[1] "t test of coefficients"
@josef-pkt

This comment has been minimized.

Copy link
Member

commented May 4, 2014

@kdiether Thanks for finding this

Good to know, prewhitening and recoloring is not yet available in statsmodels
Also I haven't looked closely enough at optimal window length selection to start implementing it.

What the defaults for the options for various sandwich robust standard errors should be is still a bit open.

@josef-pkt

This comment has been minimized.

Copy link
Member

commented May 22, 2014

closing this, since there is no difference given the right options.

Defaults for the various sandwiches, robust covariances are still a bit open for rethinking.
But even Stata across functions and across user supplied functions is not always consistent. Additionally, the best choice often depends on the application and the data.

@josef-pkt josef-pkt closed this May 22, 2014

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
You can’t perform that action at this time.