Hi, I would like to to quantify the bias that occurs while using the loglinear form of a deterministic model. So far, I have used the original nonlinear model, since there is no need to loglinearize in a deterministic setup, but I know there should be a small gap between loglinear and nonlinear solutions and it would be great if I could compare the two solutions. Since I impose the ZLB on nominal interest rate, is it possible to use the semi-loglinear model, that is, the model in which all equations apart from the Taylor rule are loglinearized ?
I am not sure what exercise you are trying to do. Could you please clarify what your question is.
Ok, it’s not very clear. So far, I have solved the nonlinear deterministic model. Now I want to solve the log-linearized model to check if there is a bias in the log-linearized solution. There should be a small bias, at least this what the authors of the paper have found (in the stochastic model). I could simply use model(linear) to do so, even if there is no need to loglinearize a deterministic model. It’s an exercise I am interested in. However, there is a zero lower bound on nominal interest rates, so I cannot linearize the interest rate rule. Can I solve this problem in Dynare ? Thanks
I still don’t get it. The true equilibrium conditions are nonlinear. The solution to the linearly approximated equilibrium condition obviously do not exactly solve the original nonlinear ones. But if you are doing deterministic simulations, there is no point is using a loglinear version, because the nonlinear model can be solved exactly. Why would you want to introduce an approximation error?