Fama MacBeth Regression with missing explanatory variables

[Jason Lii]

I am trying to run a Fama-MacBeth regression with multiple independent variables. The dependent variable is the firm bond return this month, the independent variables are the lagged bond returns for each of the last 12 months. However, due to my construction method, the returns of some months cannot be calculated and I set them to missing values. The problem is that when I ran the FM regression with lagged 12 month returns, STATA will automatically drop the observations with even one missing return in the last 12 months. And this more than halved the number of observations I have. As a result, I do not have enough statistical power.

Is there a way in STATA to adjust the missing variables problem in FM regression? Thanks a lot!

[Attaullah Shah]

Stata will always drop rows with missing values when estimating a regression. That is the default in my asreg program (available from SSC) for Fama-Macbeth regression and should be the default in any other program. If you try to replace missing values with some methods of interpolations, your data will not be a real dataset, rather a manipulated one with some statistical techniques.

More on asreg can be found here https://fintechprofessor.com/stata-p...ions-in-stata/

[Phil Bromiley]

There are multiple imputation and maximum likelihood (see the SEM documentation) approaches to missing data. How these fit with FM estimation I have no idea.

[Jason Lii]

Stata will always drop rows with missing values when estimating a regression. That is the default in my asreg program (available from SSC) for Fama-Macbeth regression and should be the default in any other program. If you try to replace missing values with some methods of interpolations, your data will not be a real dataset, rather a manipulated one with some statistical techniques.

More on asreg can be found here https://fintechprofessor.com/stata-p...ions-in-stata/

Hi Attaullah. Thank you for your clarification. Actually, I have been using your asreg extensively in my research. It is very helpful.

[Jason Lii]

There are multiple imputation and maximum likelihood (see the SEM documentation) approaches to missing data. How these fit with FM estimation I have no idea.

Thank you! I'll check it out.