Multiverse robustness tests in macroeconomic calibration models

It is important to evaluate the reproducibility and replicability of published findings. Such work in economics has largely been limited to non-structural empirical studies, but we suggest extending it also to macroeconomic calibration models. Building on previous work about parameter uncertainty in macroeconomic models, we propose evaluating the robustness reproducibility of calibration results by running the calibration for all combinations of reasonable parameter values and expressing the uncertainty in terms of a 95% multiverse interval. We apply this approach to two macroeconomic calibration models from the American Economic Review. Our multiverse calibration results suggest that parameter uncertainty can strongly limit the informativeness of calibration results. For example, the 95% multiverse interval of the welfare gain of an optimal tax/subsidy of operating costs of incumbents is 0% to 10% in one model and the 95% multiverse interval of the annualized TFP growth rate in China is -3% to +7% in the other model.