Plot constant over time

[Roeland de Bruin]

Hi,

I would like to plot my constant term (which I obtained from a regression) over time, to see whether there is a pattern. Is that possible in stata?

My second question, how can I see which percentage funds has a negative constant? (short explanation: the constant indicate a possible outperformance of mutual funds, I would like to know which percentage of funds under/outperformed)

Thank for reading my questions!

[Roeland de Bruin]

Any ideas?

[Rich Goldstein]

sorry, but the first question is completely unclear to me - how can the constant change?

re: your second question - you need to provide more information about what you are doing; at the very least, show us your command exactly as entered

[Roeland de Bruin]

Thanks for your response. Oh off course the constant can't change over time, sorry! For the second question; I use a Carhart four-factor model to investigate mutual fund performance. I use the following command: 'reg Ret MKT SMB HML MOM, robust'. The constant (alpha) of the regression output indicates an average out/underperformance. But furthermore I would like to which percentage funds underperformed, so which percentage of my funds has a negative alpha. Thanks!

[Oded Mcdossi]

Your question is still unclear, to me at least. As to your first question, I think you can test changes in the constant term over time, it depends on the type of your analysis. This is one way of doing that.

​use http://www.stata-press.com/data/r12/union.dta, clear    
statsby, by(year): reg grade south union black
tw line _b_cons year, sort

[Roeland de Bruin]

Thanks Oded, it worked! And for the other question, I don't really know how to make it clearer. Attached a histogram of my constant term. I'm investigating funds, and each fund has a constant, as shown in the histogram. I'd like to know which percentage funds has a negative alpha. Thanks!

[Oded Mcdossi]

Dear Roeland, I'm not familiar with funds research, but to get the proportion of negative alpha you can type:

gen neg_alpha=alpha<0 if !mi(alpha)
tab ​neg_alpha

The percentage of 1 is the proportion of all the negative alpha in your sample

[Roeland de Bruin]

Thanks for the help, Oded!

[Roeland de Bruin]

Oh Oded, one last question, a little bit in the same direction as the previous. I'd like to compare the constants of two different regression with each other (to see which sample performed better). Can I do this with a two sample t test, of a Wilcoxon rank sum test? I can only compare variables with each other using these tests, but I can save my estimates (constants) as variables and than compare them, am I right? Or, are there any other possibilities? Great thanks for the help again..!

[Rich Goldstein]

"help suest"

[Roeland de Bruin]

Thanks. I've tested the following (compared a constant of -0.0157 with a constant of 0.00233)

test [EU1_mean]_cons=[EU2_mean]_cons

( 1) [EU1_mean]_cons - [EU2_mean]_cons = 0

chi2( 1) = 241.09
Prob > chi2 = 0.0000

So the coefficients differ (at a 5 % signifcance level?). I'm not sure whether I now can conclude that the sample with the constant of 0.00233 performed better, given this test result.