[Veeckman Art]

Dear Sano

Thank you for your recommendations!
My promotor suggested to examine whether Region could be a mediator. This would mean that the effect of environmental performance on financial performance could result in a different outcome since the different Regions could value environmental practices differently.
(Is this logic right?)

So the model build in #11 looks at the effect of different regions on financial performance (it does not examine if there is possible difference between environmental performance valuation on financial performance)

If I would want to examine the effect of the different regions on the valuation of environmental performance on financial performance.

What should i do then?
- Region separated Regression (how should i do this?)
- Interaction variable between environmental performance and region dummy (how should i do this?)

Thank you for your suggestion

[Veeckman Art]

Dear Sano

Sorry within #16 I wrote mediator, however, this is wrong I meant to say moderator
Thus, I would like to check if the Region in which a firm operates has a moderating effect in the relationship between environmental performance and financial performance.

Should I therefore add the interaction term? (while leaving the environmental performance CEP variable and the Region dummy variables in the model)?
OR should I do this by Region separated regression (how should i do this?…possible solution by if region==1, region==2, region==3, region==4)?

Kind regards!

[Jacob Sano]

Veeckman,
I definitely follow your logic in #16. The difference between the separated regressions and interacted variable would be that in the separated regression, the regional effect would be applied to all of your estimates. Given the small z-score on CEP in the initial regression, I'm not thinking that you'll find significant regional differences. I think the thing to do would be to run a separate regression for each region (regression if region == regionnumber, options like you said in #17) and see if your CEP estimate comes in significant for any region. Remember to remove your region dummies from the regression and that if you don't have any time-invariant regressors you can use the -xtreg, fe- estimation model.

[Veeckman Art]

Dear Sano

Thank you again for commenting!

Thus, you recommend, running 4 new regressions while adding "if region==1;2;3;4" at the end of each regression.
And if for one or more regions (thus different regressions) the variable of environmental performance turns out significant, it is possible to say that these regions value (positive or negative) environmental performance since it has impact on financial performance (is this correctly interpreted?)

I'm sorry i'm a bit confused but am I using the region dummy variable then truly as a mediator between the relationship of environmental performance or financial performance? (Since my hypothesis states that I expect different valuation for environmental performance on financial performance across different European regions)

I'm just wondering, since my course book uses an interaction term of environmental performance and region(dummy) when using a moderator.

Kind regards!

[Jacob Sano]

Both methods will allow you to see how the effect of environmental performance on financial performance varies from region to region. I think you're correctly interpreting what I was suggesting regarding separated regressions. If your course book recommends using interaction terms, that works too! Again, there are slight differences between the methods in that using interactions in a single regression won't apply regional differences for the rest of your regressors.

[Veeckman Art]

Dear Sano

Okay I'll try it by using the code hereunder, thus I removed the region dummies from the regression and added if Region==1;2;3;4 at the end of each model.
However, I still opt for the - xthybrid - method since I have another time invariate variable (namely, the Sector in which the firm operates, I created S1-S9 for each sector, however leave one sector out of the model to avoid dummy trap).
Furthermore I keep the different Years in the model Y13-Y18, but also leave out one Year to avoid dummy trap

Could you confirm that I correctly interpreted what you said in #18 and executed the right method hereunder?

xthybrid ROA_w CEP Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 if Region==1, clusterid(id) full
xthybrid ROA_w CEP Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 if Region==2, clusterid(id) full
xthybrid ROA_w CEP Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 if Region==3, clusterid(id) full
xthybrid ROA_w CEP Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 if Region==4, clusterid(id) full

xthybrid TobinsQ_w CEP Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 if Region==1, clusterid(id) full
xthybrid TobinsQ_w CEP Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 if Region==2, clusterid(id) full
xthybrid TobinsQ_w CEP Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 if Region==3, clusterid(id) full
xthybrid TobinsQ_w CEP Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 if Region==4, clusterid(id) full

[Veeckman Art]

Dear

Currently the codes from #21 appear to be working, however, when applying region separated regressions, the results for environmental performance remain insignificant for all 4 regions environmental performance had p-value higher than 0,10
Which indicates that environmental performance has no significant effect on financial performance, while taking into consideration the different European regions.

Is it correct that - xthybrid - can not take interactions term up in the regression such as;
xthybrid ROA_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8, clustered(id) full
xthybrid TobinsQ_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8, clustered(id) full

However when using FE this would be possible >> only problem here is that the different sectors would be omitted (time invariate)
xtreg ROA_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 S9, fe cluster(id)
xtreg TobinsQ_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w Y14 Y15 Y16 Y17 Y18 S2 S3 S4 S5 S6 S7 S8 S9, fe cluster(id)

Would an possible solution here be using reghdfe or ivreg2?
ivreg2 ROA_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w i.Year i.Sector, cluster(id)
ivreg2 TobinsQ_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w i.Year i.Sector, cluster(id)

reghdfe ROA_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w i.Sector i.Year, noabsorb cluster(id)
reghdfe TobinsQ_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w i.Sector i.Year, noabsorb cluster(id

Just to be sure what would be the difference between clustering on id but insert i.Year and clustering by both id and Year
reghdfe ROA_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w i.Sector i.Year, noabsorb cluster(id)
reghdfe ROA_w c.CEP##i.Region Size Inno_w Leverage_w Growth_w Mshare_w Capital_w CurrentRatio_w Cashflow_w i.Sector, noabsorb cluster(id Year)


Kind regards!

[Jacob Sano]

Yes, the code from #21 does run region separated regressions. You can use the coefficient estimates from these regressions to see how the effects of your regressors vary from region to region.
It looks like -xthybrid- doesn't allow interactions to be made within the command. Having run the separated regressions and finding no significant effect for your CEP variable, running a joint regression with interactions will at this point be redundant.

[Veeckman Art]

Dear Sano

Okay, I understand. That since there is no significant effect of CEP by applying if region==1;2;3;4, it is not longer necessary to do quite the same thing by creating an interaction variable between CEP and i.Region, since this will also result in an insignificant effect.

Thus, since environmental performance is not significant within any of the region separated regressions, it is possible to conclude that the region in which a firm operates does not impact the relation between environmental performance and financial performance. This would be hypothesis 2

Also, if I just want to control whether a firm increases environmental performance with the aim of increasing financial performance, should i really add the regional dummy to this code?

I ask this since hypothesis 1 states, that European firms adopt environmental practices with the aim to increase their financial performance.
Within hypothesis 1 I'm not yet making a distinction between different European regions.
It is only in Hypothesis 2 that I state that there might be a difference between the regions in valuation of environmental performance.

If I would not use the different EU regions to examine hypothesis 1, I would still use - xthybrid - model since I've got different sectors which are time invariate, and not FE model because FE would omit this sector variable

Thank you for all the help and clear information

Kind regards

[Jacob Sano]

Veeckman,
Your conclusions make sense to me. As you seem to be getting at here, you don't need to include to include variables that are time-invariant within panels if you don't need the coefficient estimates, as the effects of these variables will be picked up by the panel fixed effects. Thus, the region variable can be excluded in your joint regression for hypothesis one.

[Veeckman Art]

Dear Sano

Thank you for your comment!

I understand most of the models and which codes I should apply thanks to you and Carlo.

However, I have one last question concerning data treatment.

I'm not sure what is allowed to do with panel data with concern to data treatment. (and what should be the best option to treat the data)
In many studies, researchers winsorize all their variables at both sides by applying (1 99) cut (1% lowest and 99% highest)
I'm therefore wondering if:
- Is winsorizing data allowed and to what extend.
- Is it for example allowed to winsorize the variable ROA on both sides thus winsorizing by (5 95) cuts (5% lowest and 95% highest)
- While only winsorizing the variable TobinsQ on one side with (0 95) cuts (no winsorizing on the low side, but only 95% on the high side)
- And winsorize the variable Mshare on one side with (0 90) cuts

To what extend should you treat data, is for example looking at a box plot per variable enough to control that there are no longer outliers present in the dataset (indicated by a individual dot)?

For me this is not really clear, since the data I collected comes from an individual database and is thus not calculated by myself.
Shouldn't I just work with the data I collected from these database, without treating the data, since this if correct data?

Kind regards!

[Carlo Lazzaro]

Veeckman:
exception made when you are 100% sure that "weird" data come from erroneous data entry procedure, data should be analysed as they are (and not maked-up).
Panel data regression is no exception to this rule.

[Veeckman Art]

Dear Carlo

I agree with you that I should not modify data to make it 'better'
However I'm not sure why people would than consider to do data treatment to remove outliers?

Hereunder, I show some box plots of variables which I use in my model
Just to be sure, you would recommend not winsorizing these variables right?

Thank you for your help!
Kind regards

[Veeckman Art]

Dear
I have constructed a FE effect regression with a interaction effect between a dummy variable for European Regions (North, South, East, West) (i.Region), and a continuous variable, environmental performance (CEP), in order to measure the moderating effect of European regions on the valuation of corporate environmental performance on financial performance.

However, when looking at the output for model 3.1 and 4.1 (see picture), the interaction effect is significant but the main effect is insignificant.
Since my reference dummy is North Europe, should i say that there is no moderating effect for CEP (if north Europe, since you only need to look at CEP here)
And should I add the the sum of CEP and interaction effect of Eastern Europe to calculate the total effect of CEP on CFP for European regions? (however since main effect of CEP is insignifaect, is it just 0 and thus the sum would be >> 0 + 0,143?)

I hope someone can provide me a clear answer to this?

Kind regards

[Carlo Lazzaro]

Veeckman:
you can say that, since your reference dummy is Northern Europe, there is no evidence of moderating effect for CEP (however, was the coefficient for Northen Europe when CEP=0 omitted?).
I'm not clear with omitting Southern and Western Europe coefficients from your calculation (provided that it makes methodological sense in your reserach field).