Structural Change

The issue is how to estimate a model, when we think that some of the structural parameters have changed over the sample period. An important case is when policy has changed.

The methodology will be very different if one think that the change is surprise that isn't anticipated by the agents (simpler to deal with) or whether the change is anticipated.

When the change is a surprise, we could deal with it in the following way:

• In estimated_params, you specify priors for the different values of a parameter and for which periods it applies (we must come up with a new syntax)

• in DsgeLikelihood.m and DsgeSmoother.m, we call dynare_resolve for each different sets of parameters and store T R and Z, Q and H in three dimension arrays (the third dimension being time)

• we must write a set of Kalman filter and smoother routines with variable T, Z, Q and H coefficient matrices.