Using fminunc for probit estimation

I am estimating a simple probit model using fminunc as an optimization tool. The reason why I don't use other optimization strategies is because I want to change code later to estimate ordered probit. I get an error "Input to ROOTS must not contain NaN or Inf." while estimating. Could you please suggest what might be wrong with my code? (see below: w- vector of independent vars, yd-dependent dummy). Many thanks, Yevgeniya

w=M.data(:,3);
yd=M.data(:,5);
k=size(w,2); 

[b,r,stats]=regress(yd,w);
start0=b; 
data=[w yd];
f0=find(yd==0);
f1=find(yd==1);

opts = optimset ('Display','iter','TolX',1e-15,'TolFun',1e-15,'MaxIter',2000,'MaxFunEvals',1000000);
[param,fval] = fminunc(@(start)P_Lik(data,k,indiv,f0,f1,start),start0,opts);

function L = P_Lik(X,k,indiv,f0,f1,start);
bcoef=start(1:k); 
y=X(:,end);
Xb=X(:,1:k)*bcoef;
w=zeros(size(X,1),1);
w(f0)=log(1-normcdf(Xb(f0)));
w(f1)=log(normcdf(Xb(f1)));
L=-sum(w);

[Joshua Scott]

Hello Yevgeniya,

I realize this post was made awhile back. I am curious, what edits did you make in order to make this ordered probit?

Thank you

Best,

Josh

Answer 1

[Walter Roberson]

You should probably be using the FunValCheck option

[Yevgeniya]

Walter, thanks! This was helpful.

Answer 2

[J Ahmed]

Hi...

I am having same problem even when I turn 'FunValCheck' on. After few simulations it stops with the captioned message. Probably, Yevgeniya, you might help if you have already solved the problem.

Thanks!

[Yevgeniya]

Hi Ahmed,

I ended up switching to fmimcon and setting up the constraints to make sure that I don't get numbers out of the acceptable range.

Hope this helps.

Yevgeniya

Answer 3

[J Ahmed]

Thanks Yvegeniya. I'll try that then.