T-test Brown Warner 1985

[Anna Papadopoulos]

Hello everyone,

Since I am doing an event study with multiple events per firm I want to account for cross-sectional dependence according to Brown and Warner (1985) who estimated the test statistics as follows:

Now I want to conduct the t-test. However, I do not know how to do it in STATA while accounting for cross-sectional dependence. I applied the following code:

summarize mean_abnormal_return_daily if dif==-1

And then I applied the ttest:

However, this would not account for cross-sectional dependence. I also calculated the t-statistics by creating a new variable but then I would not know how to aggregate it into one t-statistics. Thus, does anyone of you has an idea how to properly conduct a t-test according to Brown and Warner (1985) in STATA?

Thank you very much in advance.

Best

Anna

[Carlo Lazzaro]

Anna:
why not using -regress-?

[Anna Papadopoulos]

Regress would not account for cross-sectional dependence I think and also I need to report the results with a two-tailed test.

[Carlo Lazzaro]

Anna:
take a look at main.pdf (uni.lu)

[Karaca ali]

Hi, I have a cross-sectional dependency problem in the models without heteroscedasticity and autocorrelation. I applied "xtgls y1 x1 x2, panels(correlated)", however, it considers heteroscedasticity too. Does anyone have any idea how to solve only the cross-sectional correlation problem? Thank you

[Carlo Lazzaro]

Karaca:
please start a new thread with a subject consistents with your topic;
please provide more details about the issue you're facing (as per FAQ). Thanks.

[Karaca ali]

Thank you, Carlo. I am running a linear regression in panel data analysis by applying the LSDV model. I did run Pesaran 2004 cd test before running the regression and this is what I got.

. asdoc xtcd SMALLLoBMRF ME1BM2RF SMALLHiBMRF BIGLoBMRF ME2BM2RF BIGHiBMRF, replace

Average correlation coefficients & Pesaran (2004) CD test
Variables series tested: P1 P2 P3 P4 P5 P6
Group variable: zon
Number of groups: 6
Average # of observations: 300.00
Panel is: balanced

Dependent Variables	CD-test	p-value	corr	abs(corr)
(P1)SMALLLoBMRF	41.630	0.000	0.680	0.680
(P2)ME1BM2RF	41.120	0.000	0.672	0.672
(P3)SMALLHiBMRF	39.780	0.000	0.650	0.650
(P4)BIGLoBMRF	46.150	0.000	0.754	0.754
(P5)ME2BM2RF	45.860	0.000	0.749	0.749
(P6)BIGHiBMRF	41.330	0.000	0.675	0.675

Notes: Under the null hypothesis of cross-section
independence CD ~ N(0,1)





. asdoc xtcd MktRF SMB HML RMW CMA Mom, replace

Average correlation coefficients & Pesaran (2004) CD test
Variables series tested: MktRF SMB HML RMW CMA Mom
Group variable: zon
Number of groups: 6
Average # of observations: 300.00
Panel is: balanced

Independent Variables	CD-test	p-value	corr	abs(corr)
MktRF	47.080	0.000	0.769	0.769
SMB	15.900	0.000	0.260	0.263
HML	27.920	0.000	0.456	0.456
RMW	9.570	0.000	0.156	0.184
CMA	28.030	0.000	0.458	0.458
Mom	38.110	0.000	0.622	0.622

Notes: Under the null hypothesis of cross-section
independence CD ~ N(0,1)


There is a cross-sectional problem but not sure what is the right approach to solving it.

[Karaca ali]

I implemented Regression with Driscoll-Kraay standard errors: "xtscc P1 MktRF SMB HML Mom D2 D3 D4 D5 D6" where Ds represent dummy variables but not sure if it right approach or not.