Dynare BCA Estimation/Implementation

Estimation ML/Bayesian estimation
1

In business cycle accounting, one must estimate the parameters of the Markov process s_t = P_0 + Ps_{t-1} + \epsilon_t,\;\; var(\epsilon_t) = V = QQ', using, say, MLE.

For replication purposes, prof. Johannes Pfeifer uses values for P_0, P, Q from the original paper in both the parameters; block and estimated_params; block (i.e., as priors) in dynare.

  1. For other countries, we need to change the initial values of the parameter in parameters; block and priors in estimated_params; block, for P_0, P, Q, yeah? Or no major issues on the results for using similar values in the original paper?
  2. Also, in the replication mod file, the log of the data is linearly detrended. But one could have also used the HP filter, right?
  3. And also in the mod file, it says government spending enters model in logs. So, for example, in resource constraint, g is in logs and other variables are in levels? But you cannot add levels and logs data, right…I know my understanding here is wrong, but I cannot understand that statement.
%resource constraint (A.2.1)
exp(c)+exp(g)+(1+gamma_z)*(1+gamma_n)*exp(k(+1))-(1-delta)*exp(k)=exp(y);
2
  1. You can either try that or work with a “more standard prior”. For example, and AR(1) with 0.9 on the diagonal. Ideally, you verify that the starting values do not make a difference.
  2. No, the HP filter is a two-sided, i.e. non-causal filter. It contradicts the state space representation of the model.
  3. That is a mistake in the comment. Thanks for pointing it out. I fixed it.