Probit estimation using MATA Optimize 'initial values not feasible'

[Roy Wirapati]

Hi,

I am quite new with MATA. I am currently working on an exercise of performing probit estimation using MATA's optimize() but I keep on getting the 'initial values not feasible' error message.I have tried searching for other discussions and applied the solution from other discussions to no avail.

Basically, I'm simulating a binary choice model

and then I have to perform logit estimation of y on x similar to standard STATA probit command:

 probit y x

.

This is the code that I have come up so far.. Thank you very much in advance for your help!

mata
mata clear

// a. Simulate Y

    // generate random variables
    X = rnormal(1000,1,0,1)        // X variable
    u = rnormal(1000,1,0,1)        // error term

    // setting up coefficients
    alpha = J(1000,1,0)        // constants
    beta  = 2                
    
    // Simulate Y*
    Y_star = alpha + beta*X + u
    
    // Simulate Y
    Y = Y_star :> J(1000,1,0)
    
    // check if the proportion of 1's is between 0.4 and 0.6
    mean_Y = mean(Y)
    mean_Y
    
// b. estimate probit estimator
    
    // preparing probit link function
    x  = X,J(1000,1,1)
    b  = (beta,1)
    xb = x*b'
    mu = normal(xb)
    
    // define function
    void PROBIT(todo,Y,mu,f,g,H) {
        f = Y :* ln(mu) + (1 :- Y) :* ln(1 :- mu)
    }
    
    // optimization
    p = J(1,cols(X),beta)        // initial value
    S = optimize_init()    
    optimize_init_evaluator(S,&PROBIT())
    optimize_init_evaluatortype(S,"v2")
    optimize_init_technique(S,"bfgs")
    optimize_init_argument(S,1,Y)
    optimize_init_argument(S,1,X)
    optimize_init_params(S,p)
    optimize(S)
    beta_probit = optimize_result_params(S)'
    beta_probit

end