How to include confiedence interval of estimation, other than OLS, while plotting quantile regression coefficients using -qregplot- ?

[pankaj swain]

Respected STATA Users,

I am aware that, -qregplot- command allows inclusion of "ols" option for plotting OLS confidence intervals. This helps in visualisation and drawing inference about whether there is statistically significant difference between quantile regression estimates using, say, -qreg2- and the corresponding OLS estimates. For example, in my dataset the following codes works fine.

qregplot Pub_Dummy CRAR RiskLev GNPA PCR NIM CorpLoan ContLiab OpEff Size ROA GDPGr GsecYld CMR EPUInd CPInfl ExcUSD, ols

My question, though is, regarding incluson of confidence intervals from OLS estimation with vce(cluster clusterid) or with vce(robust) options, for better comparison. Is there any code which enables me to do so? I being the novice, attempted with the following code, which yielded errors.

. quietly qreg2 $ylist $xlist, cluster(BankID)
qregplot Pub_Dummy CRAR RiskLev GNPA PCR NIM CorpLoan ContLiab OpEff Size ROA GDPGr GsecYld CMR EPUInd CPInfl ExcUSD, ols cluster(BankID)
option cluster() not allowed
If the above message is not useful, please contact me at [email protected]

I would really appreciate any suggestions regarding the correct way to proceed. Especially, I want to include confidence intervals from OLS estimation carried out with clustered standrad error, i.e., vce(cluster clusterid) option in conjunction with my quantile plots. Please help.

Thanks and regards
pankaj

[FernandoRios]

Qregplot is not an estimation command
do not include cluster() option to the line
see helpfile for details on options

[pankaj swain]

Thank you for your suggestion. I came across an option olsopt(regress options) in the helpfile for -qregplot-, as below.

olsopt(regress options) Used to provide additional information to estimate the model via regress. For example, olsopt(robust) request OLS regression to estimate the model using robust standard errors.

As it is under -qregplot- helpfile, I assumed that it can be used in place of ols option, for example:

qregplot Pub_Dummy CRAR RiskLev GNPA PCR NIM CorpLoan ContLiab OpEff Size ROA GDPGr GsecYld CMR EPUInd CPInfl ExcUSD
> , olsopt(cluster)

or

qregplot Pub_Dummy CRAR RiskLev GNPA PCR NIM CorpLoan ContLiab OpEff Size ROA GDPGr GsecYld CMR EPUInd CPInfl ExcUSD
> , olsopt(robust)

However, both of these options produce quantile estimation plots only, ignoring the ols option altogether (i.e., without OLS coefficients and confidence intervals). So, can anyone please throw light on the utility of olsopt with -qregplot- command ?

Many thanks
pankaj

[FernandoRios]

You still need to include -ols- as the main option

[pankaj swain]

Sorry, but your answer is not clear to me. Do you mean to say, the above command would work if used along with -ols- option? For example, would the following command work?

 
 qregplot Pub_Dummy CRAR RiskLev GNPA PCR NIM CorpLoan ContLiab OpEff Size ROA GDPGr GsecYld CMR EPUInd CPInfl ExcUSD, ols olsopt(cluster)

If that is not the case (I have already tried), I don't get the utility of olsopt option? That means olsopt is redundant or of little use, as far as -qregplot- command is concerned.

Sorry, if I am missing something out there.

Thanks
pankaj

[pankaj swain]

In continuation of my original question,

Can I plot the confidence intervals from my OLS estimates with clustered or robust SE, along with quantile regression coefficients for different regressors, using -qregplot- command or any other command for that matter?

My empirical model includes a measure of risk as its dependent variable, which is known to exhibit extreme volatility at tail distributions. This makes a case for estimation using quantile regression. Although theoretically sound, I need to substantiate by showing statistically significant difference between OLS estimates and QR estimates at different distributions of dependent variable. The -qreg2- estimate with clustered option executes Parente- Santos Silva test for intra-cluster correlation and rejects the null of no intra-cluster correlation. Therefore, to have a meaningful comparison, I am interested in comparing the QR confidence intervals with the confidence intervals obtained form OLS with vce(cluster clusterid). However, the vanila ols option after -qregplot- produces confidence intervals pertaining to simple OLS regression (without clustered standrad errors).

I have tried my best with the very limited understanding I have got, to state my concerns. I would really welcome any input to solve my problem.

Thanks
pankaj