A broad simulation-based inference approach for poorly defined models is built from a study. Using auxiliary statistics matched to simulated counterparts, this approach does not require parameter identification or a one-to-one binding function. Beyond simulator parameter calibration, the (pseudo-)simulators’ asymptotic validity and bootstraps are characterized. In a modeled dynamic stochastic equilibrium model simulation and two empirical applications on the New Keynesian Phillips curve and Industrial Production index, impulse-response (IR) matching illustrates the approach. In addition to Wald-type statistics, local projections IRs are examined using a factor-analytic distance measure without a weighting matrix.

Image courtesy of interviewee