Independent variables Fama MacBeth regression drop for no reason

Hi all,

I'm performing a Fama MacBeth analysis with the xtfmb function. The regression looks like this: excessreturn = Rm-Rf SMB HML Mom and eight dummy variables.

For some reason the xtfmb function drops the first four factors (Rm-Rf, SMB, HML and Mom). The dummy variables are still included and the results look normal.

Anyone has an idea what could be the problem? The four variables are all "float" whereas the dummies are "byte".

This has cost me a lot of time so any help is highly appreciated!

Re: Independent variables Fama MacBeth regression drop for no reason

When I used "tsset tickersymbol fiscalyear,y" the situation was as described above.

I just found out that when I use "tsset fiscalyear tickersymbol" the function does not drop the variables. Does this make sense to anyone? The output is not as desired right now so I would like to run the test as it should (tsset tickersymbol fiscalyear).

Re: Independent variables Fama MacBeth regression drop for no reason

I have similiar results. The command just works with tsset timeseries firm_id, but with tsset firm_id timeseries (as desired) it doesnt work. Did you find a solution yeT?
Thank in advance..

Re: Independent variables Fama MacBeth regression drop for no reason

Suggest:
1. Post your commands and output, verbatim, so that we can have a proper look for clues as to what is happening
2. Post the result of -xtsum- after your -tsset- command

Given that -xtfmb- is a user-written package, if we can't get it to work as desired then it would also be worth contacting the author for help.

Re: Independent variables Fama MacBeth regression drop for no reason

It might have something to do with the timevariable. I've generated a new variable that starts at one for each firm and increases with every observation of that specific firm. When I use this as time variable instead of years, the variables dont drop. This causes a bias though, and I'd like to run the regression as it should be.

Re: Independent variables Fama MacBeth regression drop for no reason

It looks like your minimum year is 2002.5 and maximum is 2009.5 - is this right? If so, that might be the problem - if the time unit is years then they all need to be integers (if you want to use monthly data you should store the dates in monthly format).

Re: Independent variables Fama MacBeth regression drop for no reason

The minimum is 2001 and the maximum is 2011, and the fiscalyear variable is integer. Some variables are %9.0g and some are %8.0g, does that make a difference?

Re: Independent variables Fama MacBeth regression drop for no reason

Ah yes sorry, I got mixed up with the next line. The %9.0g and %8.0g should not make a difference - this simply controls the way the values are displayed (see -help format-).

I'm really not sure what the problem is. Since it works when you rescale the time variable it seems that there must be some kind of problem with structure of the data, but I'm finding it a bit hard without looking at it. Feel free to send me your data (send me a private message for email address), otherwise suggest check with the author of -xtfmb-

Re: Independent variables Fama MacBeth regression drop for no reason

Okay - thanks for sending me the data because now the problem is clear.

The Fama MacBeth model (at least as implemented in -xtfmb-) works by running a cross-sectional regression for each year, and then averaging the coefficients from the first step. The four variables that are being omitted do not vary within each year, eg in 2001 rmrf never changes. Therefore the first stage of the model (the cross-sectional regression) omits them simply because it's impossible to include them in the cross-sectional regression since they don't vary.

So my take is that omitting those variables is correct, and if you don't want to omit them then you need to use a different model (ie not Fama MacBeth).