AR(1) version of State-Space Model (SSM) framework

Hi guys,

I am using the SSM framework to estimate the yield curve model of Diebold-Li, similar as in https://nl.mathworks.com/help/econ/examples/using-the-kalman-filter-to-estimate-and-forecast-the-diebold-li-model.html

However, the link above uses a VAR(1) model for the transition matrix A, whereas I want to restrict A to be diagonal (three individual AR(1) processes).

Is there a clever way to incorporate this restriction in the provided code:

%% State-Space Estimation of the DNS-model with Kalman filter 

Mdl = ssm(@(params)Example_DieboldLi(params,yields,maturities));

% Initialisation 
A0 = EstMdlVAR.AR{1};      % Get the VAR(1) matrix (stored as a cell array)
A0 = A0(:);                % Stack it columnwise

options = optimoptions('fminunc','MaxFunEvals',25000,'algorithm','quasi-newton', ...
    'TolFun' ,1e-8,'TolX',1e-8,'MaxIter',1000,'Display','off');

[EstMdlSSM,params] = estimate(Mdl,yields,param0,'Display','off', ...
    'options',options,'Univariate',true);

I suspect I have to change the optimization procedure from 'fminunc' to 'fmincon', but I dont know how to add the constraints given the way the code is written.

I have tried to initialise the off-diagonal elements of A to 0, and kept the optimization procedure to 'fminunc', but I ended up with a full VAR(1) matrix.

A0 = zeros(3);               % AR(1) version
A0(1,1) = EstMdlAR_beta1.AR{1,1};
A0(2,2) = EstMdlAR_beta2.AR{1,1};
A0(3,3) = EstMdlAR_beta3.AR{1,1};
A0_column = A0(:);           % Stack it columnwise

Thanks in advance.