Dear all,
I am currently writing my master thesis and am unsure if the regression model that I currently employ is suitable for my analysis. I am hoping to get some feedback on this matter.
I analyze changes in several accounting variables (e.g. assets or debt levels) following a specific event. My dataset is an unbalanced panel set with several observations per firm and a period spanning quarters from 1985 to 2011. I am only interested in the change in dependent variable one year after the event, but as I often have several events per company that I work with I want to account for firm-fixed and probably time-fixed effects. My dataset spans roughly 430 observations.
So far I worked with a time- and firm-fixed regression with two-way clustering (cluster 2 command in stata as taken from Petersen, 2009), but I do realize that I have way too many dummies with that method than generally accepted (1 for less than 10 observations, more like 1 for 5 observations). Since I run several regressions with different dependent variables and also different control variables, typical tests such as the Hausman test or the testparm test proved inconclusive (gave me different results depending on the regression). I would like to use the same method for all regressions.
I was wondering if there is another regression method that would be more suitable? Due to the different test results for the Hausman test I am hesitant to employ a random effects / mixed linear effects model. Also, I tested my model using only firm-fixed effects, which, however, did not lead to higher predictor significance (and at times, the testparm test in Stata suggested that time-fixed effects are important). Even then, I would only be close to the "1 dummy for every 10th observation" threshold. Or am I taking this threshold too seriously?
I am looking forward to your response and highly appreciate any help.
Kind regards,
Lara
I am currently writing my master thesis and am unsure if the regression model that I currently employ is suitable for my analysis. I am hoping to get some feedback on this matter.
I analyze changes in several accounting variables (e.g. assets or debt levels) following a specific event. My dataset is an unbalanced panel set with several observations per firm and a period spanning quarters from 1985 to 2011. I am only interested in the change in dependent variable one year after the event, but as I often have several events per company that I work with I want to account for firm-fixed and probably time-fixed effects. My dataset spans roughly 430 observations.
So far I worked with a time- and firm-fixed regression with two-way clustering (cluster 2 command in stata as taken from Petersen, 2009), but I do realize that I have way too many dummies with that method than generally accepted (1 for less than 10 observations, more like 1 for 5 observations). Since I run several regressions with different dependent variables and also different control variables, typical tests such as the Hausman test or the testparm test proved inconclusive (gave me different results depending on the regression). I would like to use the same method for all regressions.
I was wondering if there is another regression method that would be more suitable? Due to the different test results for the Hausman test I am hesitant to employ a random effects / mixed linear effects model. Also, I tested my model using only firm-fixed effects, which, however, did not lead to higher predictor significance (and at times, the testparm test in Stata suggested that time-fixed effects are important). Even then, I would only be close to the "1 dummy for every 10th observation" threshold. Or am I taking this threshold too seriously?
I am looking forward to your response and highly appreciate any help.
Kind regards,
Lara
Comment