robustvcv matlab function error message

Good morning to everyone, I had an error when I try to estimate a robust variance covariance matrix, using robustvcv. In particular, I want to evaluate the variance covariance matrix of a matrix of data (a matrix with 4683 x 20 observations), using the prameters estimated from fmincon.

I use the following syntax:

[vcv]=robustvcv('f', pars, 0, data), where data is the original matrix of input data of the f function, 'f' is a functiont that returns, as first output a scalar(sum of loglikelihood) and a second input a vector Tx1 of loglikelihood values, as required by function robustvcv. 0 is the usual value that refers to the lag.

pars is a 1x82 vector of parameters estimated from fmin* ,but I get the following error message:

Index in position 1 exceeds array bounds (must not exceed 1).

Could someone tell me what could cause the error message?

I use the same function robustvcv on a vector 1x4 parameters, but using a vector Tx1 of observations, and it gives me the right answer, without error.

[Alessandra Costa]

Has everyone suggestions?

[Geoff Hayes]

Alessandra - where did you get the robustvcv function? Is it included in a MATLAB toolbox or is this third party code or something that you have written?

[Alessandra Costa]

It's a function of Mfe toolbox of Sheppard

[Alessandra Costa]

https://github.com/bashtage/mfe-toolbox/blob/master/utility/robustvcv.m Here there's the code regarding robustvcv function

[Geoff Hayes]

Alessandra - please copy and paste the full error message to this question so that we can get an idea of which line is throwing the error. You may also want to include your f function (just attach it).

[Alessandra Costa]

Now, it seems it goes to infinite loop. The function is in the file attached below.

[Alessandra Costa]

I think there could be some errors in the code; even if the feval function gives me the exact result of likelihood, when I debugged the hessian_2sided function, which is called by robustvc, I noted that gm and gp take the same values, and the same happens for Hm and Hp. All these variables have the final value of loglikehood, thus in H I get only NAN.

n = size(x,1);

if size(x,2)~=1

error('X must be a column vector.')

end

try

feval(f,x,varargin{:});

catch FE

errtxt=['There was an error evaluating the function. Please check the arguements. The specific error was:' FE.message];

error(errtxt);

end

fx = feval(f,x,varargin{:});

% Compute the stepsize (h)

h = eps.^(1/3)*max(abs(x),1e-8);

xh = x+h;

h = xh-x;

ee = sparse(1:n,1:n,h,n,n);

% Compute forward and backward steps

gp = zeros(n,1);

gm = zeros(n,1);

for i=1:n

gp(i) = feval(f,x+ee(:,i),varargin{:});

gm(i) = feval(f,x-ee(:,i),varargin{:});

end

hh=h*h';

Hm=NaN*ones(n);

Hp=NaN*ones(n);

% Compute "double" forward and backward steps

for i=1:n

for j=i:n

Hp(i,j) = feval(f,x+ee(:,i)+ee(:,j),varargin{:});

Hp(j,i)=Hp(i,j);

Hm(i,j) = feval(f,x-ee(:,i)-ee(:,j),varargin{:});

Hm(j,i)=Hm(i,j);

end

end

%Compute the hessian

H = zeros(n);

for i=1:n

for j=i:n

H(i,j) = (Hp(i,j)-gp(i)-gp(j)+fx+fx-gm(i)-gm(j)+Hm(i,j))/hh(i,j)/2;

H(j,i) = H(i,j);

end

end