Chow test in panel data

[Francis Czarnowski]

Hi all,

My estimated equation takes form of:
ln(I)=B1ln(w1)+B2ln(w2)+B3ln(w3)+(control variables)+error

The regressions run is a fixed effect regression for years 2010 to 2015.

The test statistic is defined as:
H=B1+B2+B3

Let's say that H=1 for the analysed period. I would like to see whether the value of H statistically changes in value after year 2010.

Do you have any idea how can I do it? I presume that the Chow test is appropriate, but I'm not sure how to apply it when my test statistic is a sum of three coefficients.

Thank you for your help.

[Fei Wang]

Francis, you may define an indicator p = 1 after 2010, = 0 otherwise. Then add interactions of p with regressors to the original regression, sum up the coefficients of p*lnw1, p*lnw2, and p*lnw3 and you can get the change in H after 2010.

xtreg lnI lnw1 lnw2 lnw3 controlvar c.q#c.(lnw1 lnw2 lnw3 controlvar) i.year, fe vce(cluster panelvar)
lincom c.q#c.lnw1 + c.q#c.lnw2 + c.q#c.lnw3    // change in H

[Francis Czarnowski]

Dear Fei,

Thank you for your comments. It is very helpful.

I wanted to confirm whether I understood this correctly. So basically, I re-run the regression adding the interaction terms (multiplying p or q with the variables). If the F test indicates a statistically significant results of the join hypothesis on the three parameters, I can conclude that the parameters changed over time. Is that correct?

Please find the result of my regression below.

Thank you for your help.

[Fei Wang]

Francis, after running the regression in #2, do not look at the joint significance of the three coefficients, but a t-test for the summation of the three coefficients, according to your definition of H -- This is done by using -lincom-, the second line of the code.

[Francis Czarnowski]

Hi Fei,

Yes, this is what I meant. Thank you so much !