Joan Alegre Canton

Structural models often fix some parameters and estimate the rest, but the calibration-estimation partition is usually chosen by convention. This paper treats that choice as an econometric partition-selection problem. For each admissible partition, I construct a scalar sensitivity statistic measuring the local response of a target object to perturbations of the calibrated parameters. The selected partition minimizes this statistic and therefore minimizes worst-case local bias from calibration errors. The procedure requires only local derivatives, avoids repeated re-estimation, and applies to a broad class of structural models.

Structural models often calibrate some parameters and estimate others, yet this partition is typically chosen informally. We propose a partition-selection criterion that selects, among admissible partitions, the one minimizing the worst-case local bias of the posterior mean induced by calibration errors. We derive an exact finite-sample sensitivity formula, avoiding Bernstein-von Mises approximations. The Bayesian rule is asymptotically equivalent to its minimum-distance counterpart, but Monte Carlo and Smets-Wouters evidence shows substantial finite-sample differences. The method is straightforward to implement and available for MATLAB/Dynare.

Testing whether a macroeconomic model can match its empirical targets is difficult when those targets come from multiple databases, structural parameters are not point identified, and the covariance matrix of the target estimators may be singular. We develop a minimum distance correct specification test that is robust to all three complications. The test uses the worst-case critical value over all positive-semidefinite covariance completions consistent with the observable within-database blocks. We illustrate the test in an impulse-response-matching exercise based on a New Keynesian model.