Dear Statalist,
I am currently writing my master's thesis and am undecided between various possibilities of panel data regressions that are possible, I'm hoping you could give some pointers for this.
I am researching the effect of ETF expense ratios on the ETF's financial performance and ESG performance. The dataset I have is an unbalanced panelset where the panelvariables are the ETFid and the date as measured in yearmonth. I have currently 80 ESG ETFs that I am using for the analysis, but since a lot of them have only been around for a couple of years, the majority of my observations are from 2018 onwards. The set goes back all the way until 2005 though as the first ESG ETFs were created then.
For my analysis I plan on regressing the financial / ESG performance variables on the expense ratio to find out how these variables interact. Given the nature of the dataset an -xtreg- makes most sense to me, but I'm struggling with the options.
1. the independent variable, expense ratio, doesn't change within the same ETF over time (except for a small number of funds). In other words, the independent variable is non-variying. I have read other threads here which mention how you can use -fe- to counter this but I'm unsure if this applies her because my non-varying independent variable is not the classic gender variable that is often used in those examples. Moreover, I have also added a Expense ratio x date interaction variable to account for the fact that the expense ratios don't change over time.
2. Using xttest0 I have found that there is heteroskedasticity in my data, hence I am using -robust for standard errors. However, there is also some literature present which encourages the use of clustered standard errors. Again, I am unsure which one is more fitting here.
Long story short, if you could give some advice / pointers on the following concepts that would be much appreciated
- Using -re
- Including the Expense Ratio x Date interaction variable
- Using robust SE's vs. clustered SE's
Many thanks in advance,
Jasper
I am currently writing my master's thesis and am undecided between various possibilities of panel data regressions that are possible, I'm hoping you could give some pointers for this.
I am researching the effect of ETF expense ratios on the ETF's financial performance and ESG performance. The dataset I have is an unbalanced panelset where the panelvariables are the ETFid and the date as measured in yearmonth. I have currently 80 ESG ETFs that I am using for the analysis, but since a lot of them have only been around for a couple of years, the majority of my observations are from 2018 onwards. The set goes back all the way until 2005 though as the first ESG ETFs were created then.
For my analysis I plan on regressing the financial / ESG performance variables on the expense ratio to find out how these variables interact. Given the nature of the dataset an -xtreg- makes most sense to me, but I'm struggling with the options.
1. the independent variable, expense ratio, doesn't change within the same ETF over time (except for a small number of funds). In other words, the independent variable is non-variying. I have read other threads here which mention how you can use -fe- to counter this but I'm unsure if this applies her because my non-varying independent variable is not the classic gender variable that is often used in those examples. Moreover, I have also added a Expense ratio x date interaction variable to account for the fact that the expense ratios don't change over time.
2. Using xttest0 I have found that there is heteroskedasticity in my data, hence I am using -robust for standard errors. However, there is also some literature present which encourages the use of clustered standard errors. Again, I am unsure which one is more fitting here.
Long story short, if you could give some advice / pointers on the following concepts that would be much appreciated
- Using -re
- Including the Expense Ratio x Date interaction variable
- Using robust SE's vs. clustered SE's
Many thanks in advance,
Jasper
