Estimation and calibration

Dear Dynare,

I have two agents DSGE model and use five observables: Y, G, L, Pi, Rb.
The model is nonlinear and I use exp form such as exp(Y), exp(G) … in the model block and use separate steady state file with ys=log(ys).

  1. For the data of Y, G obtained real data per capita, did seasonal adjustment, took a log, detrended using one-sided HP filter in Eviews (and also in Matlab for comparison - result are same) and demeaned which means I obtained percentage deviations from steady state/trend (as I understood correctly the guide).
  2. For inflation data of Pi, I first got raw CPI data, and then use LN(1+LN(CPI_t/CPI_t-1))
  3. For the data of Rb (nominal int rate), I got raw annual net interest rate, and then use LN(1+Rbobs_t/400).
  4. For L (per capita hours worked), I seasonally adjusted, took a log and demeaned by taking out long-run mean from each observation.

So, my observation equations are:

My questions are:

  1. My estimation does not work without estimated_params_init (use_calibration). It shows
    dynare_estimation_init:: The steady state at the initial parameters cannot be computed.
    Should I consider it bad? But with estimated_params_init (use_calibration) it works okay.

  2. Another question on government expenditure share (gamma_g=Y_ss/G_ss). My raw data shows that in the long run the GovExp/GDP share should be equal to 0.15-0.2. However, detrending and demeaned gov expenditure is near zero. How should I calibrate government expenditure share n order to find steady state taking into account I enter detrended and demeaned observables into Dynare?

  1. Yes, this is bad, because it implies that your estimation will reject parameter draws just because the steady state cannot be computed.
  2. When you work with detrended data, you need to manually set the government spending to GDP share to its long-run value. Often, one simply takes the data mean.

Thank you for your comments. I will try to fix it.