If you're setting a transition matrix with fixed values (like 0.5), “MATLAB” interprets this as a fixed transition matrix which won’t be estimated.
So, to estimate the transition probabilities, you need to set the transition matrix in “dtmc” to “NaN” values, like how you're doing for the “AR” and constant terms.
Here's how you can modify your code:
% Initial values to start the estimation
p0 = NaN(2); % Transition matrix elements to be estimated
mc0 = dtmc(p0);
c3 = [0; 0; 0];
AR3 = repmat({eye(3)}, 1, 5); % 5 lags, each with identity matrix
Sigma1 = 3*eye(3);
mdl01 = varm(Constant=c3, AR=AR3,
Covariance=Sigma1);
mdl02 = varm(Constant=c3, AR=AR3,
Covariance=Sigma1);
mdl0 = msVAR(mc0, [mdl01; mdl02]);
% Final estimation of MSVAR
modest = estimate(mdl, mdl0, msMat);
summarize(modest)
Feel free to learn more about “msVAR” here:
