I am very confused about the output of my regressions.

  1. First of all, I am not even sure if I could divide my sample as I did, meaning that by subsampling as I did the variable ESG score is both criteria for the division as well as an explanatory variable.
  2. Then, I am not sure about the best way to interpret the coefficients and their probabilities. For example, the variable "Size", "Blockholding", "Forecast bias" and "Forecast dispersion" are not significant in the regression that includes the Full Sample neither in the subsample that only includes observations with low ESG scores. Nevertheless, those variables become statistically significant in the subsample containing the observations with a high ESG score. How can I interpret those results? Do they even have interpretation possible?

Note: I already have performed the Hausman test to check if the fixed-effects model was the most appropriate method, which it seems to be. But, I did not test yet for endogeneity concerns.

  • So, this confusing output may be related to endogeneity ? Or, it may be a sign that my data lacks of some quality?

In any case, thank you for your attention and for any help you can offer!

Output of the regressions: full sample and 2 subsamples (low ESG score and high ESG score)


For your first question, if you split your sample as you did, you would be treating it as two samples used to fit models for each of two separate populations: those with low ESG scores and those with high ESG scores. (Since you are using ESG score in your model, though, you could also consider an interaction between ESG and your other predictors if you want to try to keep this to one combined model.)

If you consider your low and high ESG models as two separate models, then the coefficients and probabilities should have their standard interpretation with respect to the low and high ESG populations. For example, holding all other things equal, a one unit increase in blockholding decreases the predicted cost of equity by 0.00333 for the low ESG population. The high p-value of 0.501 means that that variation is likely to have just occurred by chance. It's not unreasonable that this variable is significant for those with high ESG, though, because that's a separate model for a separate population.

| cite | improve this answer | |
  • $\begingroup$ Thank you very much for your insightful comment and suggestion, your insights were definitely very helpful! $\endgroup$ – Patrícia Apr 8 at 19:16
  • $\begingroup$ I have been thinking about your suggestion of using a combined model, and also based on the literature that I have reviewed, I am considering to execute this one: !Here it is the model Do you think this one may be appropriate or it can raise some problems in terms of econometrics? $\endgroup$ – Patrícia Apr 10 at 15:14
  • $\begingroup$ Unfortunately the page you linked is blocked on my work computer, so I'm not sure what type of model you're referring to. I'm also not really well-versed in econometrics, so someone else may be able to better speak to potential issues relating to that field. Sorry I couldn't be more help here. $\endgroup$ – llottmanhill Apr 10 at 19:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.