Hi everyone,
I'm encountering issues with running panel CFA/EFA. I have a dataset of ~150.000 observations of S&P 1500 companies from 2002-2019 (COMPUSTAT data).
From what I've seen, there are several ways to do this.
1) When I ran GSEM using group(fyear) as a variable, it gets stuck on refining starting values and just keeps running with no effect.
2) The other option I found was running glamm, however I cannot figure out if this is even viable and if yes, how to specify which variables are latent and how to specify the using panel data (I assumed using group(fyear) would work).
To be more precise, I have 11 variables (like ROE, ROA etc) and I'm interesting on how many factors they load.
I ran normal factor varlist, pcf and found 4 factors (5th is marginal), but I am interested in seeing how they load while taking into account time component (which from my understanding this function does not do).
3) I found a function called xtdpml, but again I cannot figure out how to specify latent variables.
Thus, I have two questions:
1) Which method is the best taking into account the panel structure and number of observations?
2) In the best model, how to specify the model to include latent variables?
Thank you!
I'm encountering issues with running panel CFA/EFA. I have a dataset of ~150.000 observations of S&P 1500 companies from 2002-2019 (COMPUSTAT data).
From what I've seen, there are several ways to do this.
1) When I ran GSEM using group(fyear) as a variable, it gets stuck on refining starting values and just keeps running with no effect.
2) The other option I found was running glamm, however I cannot figure out if this is even viable and if yes, how to specify which variables are latent and how to specify the using panel data (I assumed using group(fyear) would work).
To be more precise, I have 11 variables (like ROE, ROA etc) and I'm interesting on how many factors they load.
I ran normal factor varlist, pcf and found 4 factors (5th is marginal), but I am interested in seeing how they load while taking into account time component (which from my understanding this function does not do).
3) I found a function called xtdpml, but again I cannot figure out how to specify latent variables.
Thus, I have two questions:
1) Which method is the best taking into account the panel structure and number of observations?
2) In the best model, how to specify the model to include latent variables?
Thank you!
Comment