Structural break in Bayesian estimation and DYNARE syntaxis

Dear Michel

Perhaps this question is not well thought out, but…

I want to use Bayesian estimation to deal with model with structural breaks. Namely, I used to have the following model (simplified version):


y = gammay(-1) + deltax + e;

x = alphax(-1) + betay + u;


y and x are observable

Now, I believe that x has a structural break at time t0, I add one extra data series which is zero untill t0 and one ater that, call it It. In description I list it among exogenous observable variables

the model becomes


y = gammay(-1) + (delta+z_dIt)*x + e;

x = (alpha+z_a*It)x(-1) + (beta+z_bIt)*y + u;


I obviously add three new parameters and do not call model linear any more.

If I run dynare, it checks steady state, plots priors and then says that “inner matrix dimensions must agree”. Is there anything wrong with this definition? How can I fix it?

PS. I think I know: the problem is with definition of a product. Clearly, I need a “dot product”. The dynare does not seem to recognise it, if I simply write


y = gammay(-1) + deltax+z_d*It.*x+ e;

x = alphax(-1)+z_aIt.x(-1) + betay+z_b*It.*y + u;


I also tried to define separately xt = x.*It; but it did not help. I am afraid I cannot make x.*It as a new observable variable as I need to multiply It on unobservable variables too. Is the any way to explain DYNARE what it is?

Dear Tanya,

the linearization will destroy your smart set up. Use instead a linear scheme.

Model X and the extra variable as two different AR processes. This will work as you have observation on the extra variable.

Then have X and the extra variable acting additively in the y equation