bootstrap for panel data

[River Huang]

Dear All, I modify a code to test the significance of product of two coefficients from two panel data models. The code is

webuse grunfeld, clear
xtset, clear

capture program drop xt_prod
program define xt_prod, rclass
    xtset company     
    xtreg invest mvalue, fe robust
    local a = _b[mvalue]
    xtreg invest kstock, re robust
    local b = _b[kstock]
    return scalar prod = `a'*`b'
    exit
end

bootstrap prod = r(prod), reps(100) cluster(company) idcluster(ID): xt_prod

1. Just for illustration, the first one is a fixed effect model, while the second one is a random effect model.
2. The end is to test whether the product of the coefficient of mvalue in the first model and that of kstock in the second model is statistically significant.

Any suggestions or comments are highly appreciated.

[Clyde Schechter]

The output of -bootstrap- gives you the estimated value of the product (in the Observed Coef. column of the table) as well as a z-statistic and a P-value. Assuming you want to do a significance test of the null hypothesis that the observed coefficient is zero, the z-statistic gives you the test and the P>|z| value gives you the p-value of that test. When I ran your code I got p = 0.004. (You might get a slightly different result because the code did not set the random number seed.)

[River Huang]

Dear Clyde, Thanks for the confirmation. In fact, I modified the code following your prior suggestion.