How do i get a scalar value so i can get rid of this error using fmincon function?

trying to find optimal weights for portfolio and want to use fmincon to maximise sharp ratio

Maximize Sharpe Ratio If Put a negative sign - on the compute_sharpe objective function: maximization problem into a minimization one:

1 %%Maximize Sharpe ratio
2 Aeq = ones(1,N);
3 beq = 1;
4
5 w0 = ones(N,1)*(1/N);
6 % NOTE: use - to convert max optimization to a min ...
optimization problem
7 w4 = fmincon(@(w)-compute_sharpe(w, avg_ret, ...
cov_mat), w0, [], [], Aeq, beq, [], [], [], ...
options);
8 [w4,fval4] = fmincon(@(w)-compute_sharpe(w, avg_ret, ...
cov_mat), w0, [], [], Aeq, beq, [], [], [], ...
options);
9 % fval4 = -sr1 as the fun = -compute_sharp
10 sr = - fval4;
11 disp(['The maximised portfolio Sharpe ratio is ', ...
num2str(sr)]);

i keep getting this error in the screen shot how do i make it a scalar value so i can get rid of it? thanks

[Cam Salzberger]

That error indicates that the objective function, and therefore the "compute_sharpe" function, returns a non-scalar value. Can you post the definition to that?

Also, why do you call fmincon twice, once with one output, and once with two? Why not just make the second call?

[LM]

In my code in the screenshot i used fmincon once the first one in the preview i have posted that your referring to but i still get the error how do i get rid of the error or turn my value into a scalar?

[Cam Salzberger]

As I said, we can't really tell you why your objective function is giving a non-scalar result without seeing your objective function. Can you edit your question and add the definition for "compute_sharpe"?

[LM]

This line here is giving me the error

w4 = fmincon(@(w)-compute_sharpe(w,avg_daily_ret,cov_mat),w0, [], [],Aeq ,beq)

Answer 1

[Walter Roberson]

According to the comments in that code, w is a column vector, which would be N x 1 . w' would then be 1 x N, and so if the w' * cov_mat * w is not going to fail on inconsistent size basis, pvar is going to have to end up a scalar.

You have pret = w * avg_daily_ret; with both defined as column vectors. That would have to fail. Perhaps avg_daily_ret is a row vector or scalar? If the * is to work, then the N x 1 * something is going to have to end up with an N x whatever result, not a scalar.

You then have p_sharpe = (pret - 3) / sqrt(pvar); . We have determined that pret cannot be a scalar (unless w was a scalar, I guess). Therefore p_sharpe cannot be a scalar.

[LM]

What if i try useing avg_daily_ret’ as input, transposed. As it is a row vector now, but if it's a col vector would the calculation go ahead?

[Walter Roberson]

With w on the left side of w * avg_daily_ret then you are going to get N x 1 size for pret, and then you have a problem. Your pret calculation needs to calculate a scalar. Possibly

pret = w' * avg_daily_ret

with avg_daily_ret still a column vector.