Sensitivity: An inquiry into the robustness of model calibration as used in modern macroeconomic research

Abstract

Calibration is a much used but problematic method for achieving quantitative predictions from modern macroeconomic models. As with other quantitative work in economics, there is reason to believe that findings on the basis of calibration can be subject to measurement error, specification error, and biases. To investigate the robustness, replicability, and credibility of findings based in calibration, I replicate and extend a model of the recent Chinese growth experience. The extensions are based on the principles of sensitivity analysis as suggested by Edward Leamer (1983) with the explicit goal of illustrating the possible uncertainty about the model results. In the case of the particular model investigated here, the results are reasonably robust to alternative calibrations as long as one excludes any alternative calibrations that violate the theoretical assumptions. There are calibrated parameters for which the model is much more sensitive to alternative specifications.