Incorporate Heteroscedasticity and Autocorrelation corrections in sureg command

[Parvathi Jayaprakash]

Dear forum members,

I have a balanced panel of 80 countries with 17 years of time series data. I am trying to incorporate seemingly unrelated regressions for three equations.
The xtsur command execution

xtsur ( lnGDP lnICTPassive PD IDV MAS UAI LTO IND ICTPD ICTIDV ICTMAS ICTUAI ICTLTO ICTIND)
( lnHDI lnICTPassive PD IDV MAS UAI LTO IND ICTPD ICTIDV ICTMAS ICTUAI ICTLTO ICTIND)
(lnCO2e lnICTPassive PD IDV MAS UAI LTO IND ICTPD ICTIDV ICTMAS ICTUAI ICTLTO ICTIND)

resulted in an output as follows

classdef _b_stat() in use
(nothing dropped)
(327 lines skipped)
(error occurred while loading xtsur.ado)
r(310);

Further, I tried performing the same execution with pooled sureg command.

 sureg (lnGDP = lnICTPassive PD IDV MAS UAI LTO IND ICTPD ICTIDV ICTMAS ICTUAI ICTLTO ICTIND Region WGI i.Year i.country)
( lnHDI = lnICTPassive PD IDV MAS UAI LTO IND ICTPD ICTIDV ICTMAS ICTUAI ICTLTO ICTIND Region WGI i.Year i.country)
(lnCO2e = lnICTPassive PD IDV MAS UAI LTO IND ICTPD ICTIDV ICTMAS ICTUAI ICTLTO ICTIND Region WGI i.Year i.country), corr

The Breusch Pagan test of heteroscedasticity indicates the model is heteroscedastic

 Breusch-Pagan test of independence: chi2(3) =   791.853, Pr = 0.0000

kindly help me correct the Heteroscedasticity and Autocorrelation in sureg command shown above.


Thanks in advance.

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - in addition to Stata code in code delimiters, provide readable output, and sample data using dataex.

xtsur is user-written. Technical problems with it should be directed to the author.

I'm not sure about sur, but conventional regressions i.panel is the same as xtreg with fixed effects. Also, the test you report is a test of independence, not heteroskedasticity. It tests whether the errors are correlated across equations, not heteroskedasticity or auto correlation. So a significant test is consistent with using sureg. As far as I can see sureg does not have the options you seek. I guess you can handle heteroskedasticity using weights.

As I understand it, sureg only gives you additional efficiency - the equation by equation results are consistent. So you could go with three separate regressions. Alternatively, you could do sureg using sem.