Let say we have 5 deep/structural parameters (\delta, \beta, \alpha, \rho, \sigma) and 3 variables (c_t, r_t, y_t) in our model.
Does calibration targeting mean manipulating a subset of the parameters, say (\delta, \beta, \alpha) to get specific values for the rest of the parameters (i.e., \rho, \sigma), as well as specific values for a subset of the steady-state variables (c_{ss}, r_{ss}, y_{ss})?. All explanations are welcomed…many thanks!!
NB: parameters above have no specific meaning.
No, calibration usually means fixing the parameter value to reach particular observed targets in your variables observed empirically.
Hi, Usually calibration is done by finding values for the parameters such that the ratios of some trended variables (e.g. consumption and output) in the model match the average ratios we observe in the data, or such that the levels of some stationary variables (e.g. the real interest rate or the unemployment rate) match the average levels in the data. For instance, the discount factor \beta will be chosen so that the steady state level of real interest rate r^{\star} in the model match the average of the real interest rate in the data.
Best,
Stéphane
FYI:
There are ample materials on calibration in first few chapters.
Hi, many thanks for the info. I have this book, so I will definitely check it out. My problem, however, was to understand the term ‘calibration targets’ which is not defined in the book. Not sure if it is a standard term in DSGE modeling, but I came across it while reading this paper. The authors, for example, estimate the regression model, log z_t = \rho log z_{t-1} + \epsilon_t separately outside the DSGE model, and use the estimated parameter \hat{\rho} to calibrate the persistent parameter in the TFP (shock) equation in the DSGE model. They refer to \hat{\rho} as a calibration target, and they have a table for such targets.
Also on Prof. Pfeifer’s github, he uses the term ‘calibration targets’ in these sentences, “…It demonstrates how in a linearized model a steady_state-file can be used to set the deep parameters of the model to satisfy calibration targets on the non-linear model. The steady_state-file takes the calibration targets and calls a numerical solver on some of the nonlinear steady state equations to get the corresponding parameters that make the steady state satisfy the targets.”
While I don’t quite yet fully understand prof. Pfeifer’s statements here, is he talking about something like \hat{\rho} when he uses the term “calibration target”?
Think about calibration as an extreme form of moment matching. Whenever you have a parameter affecting an object, you can set a target for that object and then choose the parameter to match the target for that objective. What exactly that target is, is up to you, the model builder.