Wit+1/Rt+1 = a1 * Wit/Rt + a2 * (Wit/Rt)^2 + a3 * (Wit*Hit)/Rt + a4 * Hit + a5 * Hit^2 + vt +eit
==> Wit+1 = a1 * Wit * Rt+1/Rt + a2 * (Wit)^2 * Rt+1/Rt^2 + a3 * (Wit*Hit) * Rt+1/Rt + a4 * Hit* Rt+1 + a5 * Hit^2 * Rt+1+ vt +eit
estimate: Wit+1 = d1 * Wit + d2 * (Wit)^2 + d3 * (Wit*Hit) + d4 * Hit + d5 * Hit^2 + eit
This is a paper from Shaw (1989), "Life-Cycle Labor Supply and Human Capital". However, I can't quite figure out how she estimated the rental rates jointly with the model and there is very little guidance in the paper.
My thought process was that given a nonlinear model I use GMM with IV. Using time dummy variables as IV I state that Dt = 1 if year = t and Dt = 0 if year != t then I estimate the entire panel data set, then change the target year for the dummy variable and repeat the estimation process. I do this T number of times for all years.
Then for example, the change in the coefficient of hours Hit will be equal to a4 * Rt+1. Assuming a1-a5 are constant then a4*Rt+2/a4*Rt+1 = Rt+2/Rt+1. Normalizing one of the rental rates R=1 then I can observe the rental rate over time.
My estimates are not quite the same as Shaw's. So where am I going wrong? Is it correct to use time dummies in this case to estimate the time specific shock?
My stata code looks like this
gen time_dummy = 0 save mydata2, replace ***************************Estimation************* ************** xtset id year forvalues i = 1967/1980{ replace time_dummy = 1 if year == `i' replace time_dummy = 0 if year != `i' gmm(rwage_f - ({xb: rwage rwage2 rEarnings hrs hrs2}+{b0})), instruments(rwage rwage2 rEarnings hrs age time_dummy) wmatrix(robust) replace d1=_b[rwage] if year == `i' replace d2=_b[rwage2] if year == `i' replace d3=_b[rEarnings] if year == `i' replace d4=_b[hrs] if year == `i' replace d5=_b[hrs2] if year == `i' replace constant = _b[/b0] if year == `i' } replace rental_rate = 1 if year == 1967 sort id year replace rental_rate = (alpha4/alpha4[_n-1])*rental_rate[_n-1] if id==id[_n-1] & year >1967 |
0 Response to How to Estimate Unobservable Time-Specific Shocks on Dependent Variables?
Post a Comment