White's heteroscedasticity consistent standard errors

[Andreas Mueller]

Dear Stata Community

I want to calculate heteroscedasticity consistent standard errors in order to perform a t-test that adjusts for potential heteroscedasticity. I know that -regress- in combination with vce(robust) will produce respective standard errors. Unfortunately, my data which I want to assess is not obtained by using a regression analysis. I have data of abnormal returns (so called monthly CTAR) over approximately 200 months and for these monthly CTARs, I want to calculate the White standard errors in order to assess the statistical significance of the mean monthly CTAR. Is there a simple solution to obtain such robust standard errors outside of a regression environment?

Thank you and kind regards
Andreas


Below you fill find an exceprt from my data. I want to calculate robust standard errors for both the monthly equally-weighted and value-weighted CTAR.

CTAR_t_ew    CTAR_t_vw
-.1137243    -.1925612
.2465935    .0349583
.1006278    -7.23e-06
.0062178    -.0639477
.0475239    .1765106
-.0061777    .0092571
-.0184741    -.0950309
.0105682    -.0597086
-.0253923    .0125138
-.0125468    -.0286284
-.0430542    -.0763519
.0192854    .0950896
.0019887    -.03727
-.0218045    -.0063499
-.0324663    -.0247297
-.014546    -.0537305
-.0016332    .0341355
-.0343921    -.0340582
-.0321075    .0083132
-.041963    .0038711
-.025144    -.0127
-.0194596    .017901
-.0379724    -.0053466
-.0044999    .00965
.0027177    .0801118

[Rich Goldstein]

a t-test is a regression so I am unclear what you are requesting; please clarify

[Andreas Mueller]

Basically, I want to obtain the robust standard errors of both CTAR_t_ew and CTAR_t_vw calculated over all my monthly CTARs similar to using the -egen- command with sd() in order to determine whether they are significantly different from zero when using robust standard errors rather than normal standard errors.

[Joao Santos Silva]

Dear Andreas,

Heteroskedascity means that the variance of the regression errors depends on the regressors; if you do not have a regression and regressors the concept does not apply. So, in your case, you do not have to worry about heteroskedasticity.

One way to confirm this is to run the following two regressions:

reg CTAR_t_ew
reg CTAR_t_ew, robust

As you'll see, the estimated intercepts in these regressions are the sample mean, and the regressions will also give you the corresponding standard error. You will also see that because there are no covariates in these models, the robust and regular standard errors are the same.

All the best,

Joao