I am currently using Eviews for my study project. I want to analyse determinants of CASH holdings during the period 2018-2020 for a group of 265 companies. My dependent variables is CASH and my independent variables are CF, CAP, G, LEV LIQ, SIZE, TANG, VOL and a DUMMY variable for the year 2020. I want to perform a panel regression to interpret the different results (coefficient, significance, …).
To deal with any issue of heteroscedasticity, I did a 90% winsorization (top 5% and bottom 5%) on all my variables, excepted for the dummy variable, to exclude the outliers values. I also converted my data into log form (the variables CF, CAP, G, LIQ had negative values so I applied a function log(X + rounded minimum value) for these variables).
I performed the Hausman test and results indicate that 'fixed effects' is a more appropriate model than 'random effects' or 'pooled OLS'. I tried to perform GLS (with cross-section weights) method but I still find heteroskedasticty in my model…
I don’t have so much experience in econometrics and I really have no idea to how to face this issue. Could the community helps me to find some solution to get an homoscedastic model ?
You might try this reference in the attached screenshot. David Booth
Hello Hugo Jln. I have never used Eviews, so cannot offer any technical advice. But I do see that there is an Eviews discussion forum:
- http://forums.eviews.com/
It seems to me you might have better luck posting your question there. And if I was you, I would ask if it is possible to model the heteroscedasticity rather than attempt to remove it. HTH.
If you used GLS and see heteroscedasticity, aren't the GLS weights modeling that?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression