Forevalues and invalid syntax(198)

[Lara Amanda]

Hello everyone

I was wondering if I could some get help to figure out the problem below:

I'm trying to run this code available on Feenstra (2014, In Search of the Armington Elasticity) documentation

But stata issued the following error message: invalid syntax(198)

*2SLS IV
forvalues i=1/`indnum' {
    preserve
    keep if loopindustry==`i'
    *for Sato-Vartia price indicator
    qui tab ccode, gen(c_I_) 
    foreach var of varlist x1sv_ij x2sv_ij {
        gen `var'hat = `var'
        sum `var'hat if loopind == `i'
        *replace `var'hat = `var'hat - r(mean) if loopind == `i'
        }
    *Run 2sls regression for sigma
    ivregress 2sls  ysv_ij (x1sv_ijhat x2sv_ijhat = c_I_*) ,  small 

    gen conssv=_b[_cons] 
    gen thetasv1=_b[x1sv_ijhat]
    gen thetasv2=_b[x2sv_ijhat]
    estat vce            /* VAR-COV TO BE USED TO CALCULATE CI'S FOR SIGMA */
    gen rho1sv = .5 + sqrt(.25 - 1/(4+thetasv2^2/thetasv1)) if thetasv2>0
    replace rho1sv = .5 - sqrt(.25 - 1/(4+thetasv2^2/thetasv1)) if thetasv2<0
    gen sigmasv = 1 + (2*rho1sv-1)/((1-rho1sv)*thetasv2)

    ******************************************************
    *             Get standard error of Sigma
    *  using Taylor series expansion aka Delta method
    *              (not used in published paper)
    ******************************************************
    
    gen sigmasvse=0
    cap if   thetasv2>0 {
        testnl 1 + (2* (.5 + sqrt(.25 - 1/(4+_b[x2sv_ijhat]^2/_b[x1sv_ijhat]))) -1)/((1- (.5 + sqrt(.25 - 1/(4+_b[x2sv_ijhat]^2/_b[x1sv_ijhat]))) )* _b[x2sv_ijhat] )=0
        replace sigmasvse=sigmasv/sqrt(r(F))
    }
    cap if  thetasv2<0 {
        testnl 1 + (2* (.5 - sqrt(.25 - 1/(4+_b[x2sv_ijhat]^2/_b[x1sv_ijhat]))) -1)/((1- (.5 - sqrt(.25 - 1/(4+_b[x2sv_ijhat]^2/_b[x1sv_ijhat]))) )* _b[x2sv_ijhat] )=0
        replace sigmasvse=sigmasv/sqrt(r(F))
    }
    mkmat  groupcom industrycountry thetasv1 thetasv2 rho1sv sigmasv sigmasvse , matrix(SV`i')
    matrix bSV`i'=nullmat(bSV`i')\SV`i'

    * Generate optimal weights
    gen uhatsv = ysv_ij - conssv - thetasv1 * x1sv_ij - thetasv2 * x2sv_ij
    
    * Generate squared error term: we subtract the mean of error by country and good. 
    egen uhatsvmean = mean(uhatsv)
    replace uhatsv = uhatsv-uhatsvmean
        
    gen uhatsv2=uhatsv^2
    drop if uhatsv2==.
    qui regress uhatsv2 c_I_*, noc
    predict uhatsv2hat
    gen ssvhat=sqrt(uhatsv2hat)

    * Weight data and reestimate 
    gen ones=1
    foreach var of varlist ysv_ij  x1sv_ijhat x2sv_ijhat ones {
        gen `var'svstar = `var'/ssvhat
        }
        
    ivregress 2sls ysv_ijsvstar onessvstar (x1sv_ijhatsvstar x2sv_ijhatsvstar = c_I_*) , small noc
    gen thetasv1w=_b[x1sv_ijhatsvstar ]
    gen thetasv2w=_b[x2sv_ijhatsvstar ]
    estat vce            
    gen rho1svw = .5 + sqrt(.25 - 1/(4+thetasv2w^2/thetasv1w)) if thetasv2w>0
    replace rho1svw = .5 - sqrt(.25 - 1/(4+thetasv2w^2/thetasv1w)) if thetasv2w<0
    gen sigmasvw = 1 + (2*rho1svw-1)/((1-rho1svw)*thetasv2w)
    sum groupcom num thetasv1w thetasv2w rho1svw sigmasvw

    *standard error of sigma using delta method (not used in published version of the paper)
    cap gen sigmasvwse=0
    cap if  thetasv2w >0 {
        testnl 1 + (2* (.5 + sqrt(.25 - 1/(4+_b[x2sv_ijhatsvstar ]^2/_b[x1sv_ijhatsvstar ]))) -1)/((1- (.5 + sqrt(.25 - 1/(4+_b[x2sv_ijhatsvstar ]^2/_b[x1sv_ijhatsvstar ]))) )* _b[x2sv_ijhatsvstar ] )=0
        replace sigmasvwse=sigmasvw/sqrt(r(F))
    }
    cap if   thetasv2w<0 {
        testnl 1 + (2* (.5 + sqrt(.25 - 1/(4+_b[x2sv_ijhatsvstar ]^2/_b[x1sv_ijhatsvstar ]))) -1)/((1- (.5 + sqrt(.25 - 1/(4+_b[x2sv_ijhatsvstar ]^2/_b[x1sv_ijhatsvstar ]))) )* _b[x2sv_ijhatsvstar ] )=0
        replace sigmasvwse=sigmasvw/sqrt(r(F))
    }
    
    mkmat thetasv1w thetasv2w rho1svw sigmasvw sigmasvwse ssvhat , matrix(SVw`i')
    matrix bSVw`i'=nullmat(bSVw`i')\SVw`i'
    restore
    
}

[Nick Cox]

Where is the code failing? For example, it could fail right at the start if indnum is not defined, as it is not in what you show us.

[William Lisowski]

Let me add that the copy of "Feenstra (2014, In Search of the Armington Elasticity)"

which I used a web search to locate at

https://www.nber.org/system/files/wo...063/w20063.pdf

contains no Stata code. Perhaps there is another version available elsewhere?

From the Statalist FAQ linked to from the top of the page, it tells us

13. How should I give literature references?

Please give precise literature references. The literature familiar to you will be not be familiar to all members of Statalist. Do not refer to publications with just author and date, as in Sue, Grabbit, and Runne (1989).

References should be in a form that you would expect in an academic publication or technical document. Good practice is to give a web link accessible to all or alternatively full author name(s), date, paper title, journal title, and volume and page numbers in the case of a journal article.