Ramsey policy fails Blanchard-Kahn condition

Hi all,

I’ve estimated my model with a Taylor-type rule and now try to do utility-based welfare cost comparisons. But when Ramsey policy is implemented in the model, the FOCs fail the BK condition (no stable equilibrium). I conduct the Ramsey policy exercise as follows:

  1. Estimate the model with a Taylor-type rule, and treat the estimated structural parameters w.r.t. the private sector as the “true” parameters;
  2. Derive Ramsey policy FOCs using Andrew Levin’s codes;
  3. Replace the Taylor-type rule with the Ramsey policy FOCs and try to run stochastic simulations using “stoch_simul”
  4. Get error messages…

I’ve checked the steady states and also tried global sensitivity analysis. Steady states are good, but even if I’ve specified a large enough parameter space “dynare_sensitivity” still declares that no parameter region is acceptable!!!

As I am new to Ramsey policy implementations, I’m comfused why Ramsey policy can’t be done in my model setting, whereas a simple Taylor-type rule is fine. I guess Levin’s codes should be right, then it seems an inherent failure of my model. But if it was, then why my model goes well with a Taylor rule?

Your insights are greatly appreciated. Thank you.


I have never worked with Levin’s code, so I cannot say anything about this. Moreover, I don’t know whether it is compativle with recent Dynare versions.
Why can you not use Dynare’s Ramsey commands?

More fundamentally, why does Levin’s code allow for parameterizations that are not BK stable when just put them into the model for stoch_simul?

First, many thanks for the reply.

As to Dynare’s Ramsey command, I think this command only loglinearizes the FOCs including the recursive form of discounted lifetime household utility (the welfare function in my interest). That command is right to do estimation, impulse response and variance decomposition. However, it’s not good for welfare analysis which should be based on the second-order approximation of the welfare function instead of 1st-order approximation.

In regard to Levin’s codes, it does not involve any numerical computation. The only purpose of the codes is to derive explicitly FOCs w.r.t. Ramsey policy problem symbolically. Thus, in fact, I do not need to check parameter values to get the symbolic FOC generator run.

Determinacy on one hand is a mathematical issue, however, on the other hand, I just find it really bizarre in an economics sense that the planner is not able to optimally allocate resources subject to standard sticky prices and rigid real wages in my model conditional on a large enough (I think) parameter space. Nonetheless, my model goes well with a standard Taylor rule implying that my model is not wrong in a visible way.

My conjectures:

  1. something wrong with the codes, I mean, MY codes (checked, not found at the moment)
  2. the economy operates in an even larger parameter space that I can hardly imagine?