摘要

Increasingly, there are calls to audit models used in public policy for how accurately they portray the reality facing policy makers. Although models cannot be logically verified as accurate representations of reality, past work concludes that modelers can reasonably demonstrate the degree of correspondence. How to establish real-world correspondence remains an open question. This paper outlines a framework of pre-modeling data diagnostics to provide evidence of correspondence. The scheme-based on Nonlinear Time Series Analysis-provides a rigorous benchmark for specification and testing of policy models that reduces the likelihood that a misspecified model is discredited by an unanticipated crisis.

  • 出版日期2015-12