I am running a nonparametric kernel regression where the dependent variable is transformed using the Inverse Hyperbolic Sine:

Code:
 npregress kernel asinh_spending age asinh_income education i.year i.FIPS, kernel(guassian)
I would like to convert the coefficients back to values before running the margins command and then plotting the margins and bootstrapped standard error:
Code:
matrix A = e(B)
forvalues i = 1/52 {
     matrix A[1,`i'] = sinh(A[1,`i']
}
ereturn post A
The coefficient matrix changes, but the margins command no longer works because i get the error:
Code:
e(return) does not identify the estimation sample r(322)
Is it possible to do this in stata via an alternative method?

What I would like is the marginsplot graphs to show income in values instead of the inverse hyperbolic sine.