Correcting for heteroskedasticity in a panel dataset

[El Mkwara]

Hello everyone,

I am trying to find the impact of financial risk on financial performance of a panel of Zimbabwe commercial banks. I used Pagan - Hall test approach as well as White test and Groupwise test as statistical approaches to investigate the homoscedasticity of the model itself. The p value I obtained from Pagan - Hall, White test and Groupwise test (Modified wald test) is less than 0.05

. ivhettest
OLS heteroskedasticity test(s) using	levels of IVs	only
Ho: Disturbance is homoskedastic
White/Koenker nR2 test statistic	: 20.259	Chi-sq(8)	P	value	=	0.0094

Given the above. How do l correct for heteroskedasticity ? What commands should l use?

[Andrew Musau]

If you have \(N>>T\), then cluster robust standard errors are robust to the presence of heteroscedasticity and arbitrary forms of serial correlation. See Stock and Watson (2008).

xtset panelvar year
xtreg yvar xvars i.year, fe cluster(panelvar)

where "panelvar" is your panel identifier.


Reference:
Stock, J. H., and M. W. Watson. 2008. Heteroskedasticity-robust standard errors for fixed effects panel data regression. Econometrica 76: 155–174. https://doi.org/10.1111/j.0012-9682.2008.00821.x.

[El Mkwara]

Hi Andrew,

l am new to Stata. Can you take me through what the code should be like give l have 9 variables (N) and time 7 years (T). My dependent variable is ROE and independent variables are LIQ, CRED, CAR, CAPR, EXCH, FINS, INFL and OPRIS

[Maxence Morlet]

Just to be precise, N is not the number of variables you have but the number of units (I presume in your case firms).

xtreg [depvar] [indep. variables], fe cluster(firmID)

[Carlo Lazzaro]

El:
like Andrew, I would also add -i.year- in the right-hand side of your regression equation.
Just a word of advice: get yourself familiar with the building blocks of panel data regression via any decent textbook on this topic.
In addition -xtreg- entry, Stata .pdf manual is a valuable source to start with.

[Andrew Musau]

Hi Andrew,

l am new to Stata. Can you take me through what the code should be like give l have 9 variables (N) and time 7 years (T). My dependent variable is ROE and independent variables are LIQ, CRED, CAR, CAPR, EXCH, FINS, INFL and OPRIS

As the previous replies suggest

xtset firm year
xtreg ROE LIQ CRED CAR CAPR EXCH FINS INFL OPRIS i.year, fe cluster(firm)

This assumes that your number of firms \(N> 30\). If you have less than 30 firms, then the cluster robust standard errors may be biased due to a small number of clusters.