Re: fminsearch: Assignment has more non-singleton rhs dimensions than

On 3/16/2016 11:31 AM, RvdH wrote:
> Hi all,
>
> While running our script to estimate some parameters we bumped into a
> problem. We are using fminsearch and Matlab returns the following error:
>
> Assignment has more non-singleton rhs dimensions than non-singleton
> subscripts
>
>
> Error in fminsearch (line 189)
> fv(:,1) = funfcn(x,varargin{:});
>
> Error in MLE (line 6)
> [x,fval,exitflag,output] = fminsearch(@(Guess)MLEfunc( Guess, V, r
> ),Guess);
>
>
> We went through our code multiple times but we can't find the problem.
> We used the debugger (DBSTOP as well) to check what goes wrong in line
> 189 of fminsearch and apparently fv(:,1) is 1x3 and
> funfcn(x,varargin{:}); is 1x177, which is exactly what the error describes.
>
> Our code is as follows:
>
> Script:
> meanV = mean(V);
> varV = var(V);
>
> Guess = [meanV,varV]; % The starting guess
>
> [x,fval,exitflag,output] = fminsearch(@(Guess)MLEfunc( Guess, V, r
> ),Guess);
>
> Function:
> % pre-define: % n = # of observations (number)
> % h = length of time between two obs measured in years (number)
> % r = riskfree rate (vector)
> % F = face value of zero coupon debt (vector)
> % V = estimates of the market value of assets (vector)
>
> % assumptions:
> % A year has 252 trading days
>
> function [ LLFS ] = MLEfunc( X, V, r )
>
> n = size(V);
> h = 1/252;
> F = 100;
> T = size(V);
>
> mu = X(1);
> sigma = X(2);
>
> % Compute Rk = log(Vkh/Vk-1h)^2
> R = zeros(n,1);
> for k = 2:n
> R(k-1) = (log(V(k)/V(k-1)))^2;
> end
>
> % Compute Rrk
> Rr = zeros(length(R),1);
> for k = 1:length(R)
> Rr(k) = (R(k)-(mu-(sigma^2/2)*h))/(sigma^2*h);
> end
>
> % LLF when V observed
> LLFV1 = -(n/2)*log(2*pi*sigma^2*h);
> LLFV2 = -(1/2)*sum(Rr);
> LLFV3 = sum(log(V));
>
> LLFV = LLFV1 + LLFV2 + LLFV3;
>
>
> % Compute dkh(sigma)
> d_sigma = zeros(n,1);
> norm_d_sigma = zeros(n,1);
> for k = 1:n
> d_sigma(k) =
> (log(V(k)/F)+(r(k)+(sigma^2/2))*(T-k*h))/(sigma*sqrt(T-k*h));
> norm_d_sigma(k) = normcdf(d_sigma(k));
> end
>
> % LLF when V unobserved
> LLFS1 = LLFV;
> LLFS2 = -sum(norm_d_sigma);
>
> LLFS = LLFS1 + LLFS2;
> LLFS = -LLFS;
> end
>
>
> V is a 177x1 vector, r is a 177x1 vector. The parameters that have to be
> estimated are mu and sigma, which means x (output in fminsearch) is
> supposed to be 1x2.
> We have no clue how to solve this problem, so any kind of help would be
> greatly appreciated!

A few lines in your code look suspicious to me:

n = size(V);
h = 1/252;
F = 100;
T = size(V);
...
for k = 2:n
...

What is the size of n? It is a length-2 vector, not a scalar. I think
you want

n = length(V);
T = n;

Alan Weiss
MATLAB mathematical toolbox documentation