Integrated assessment models are economists' principal tool for analyzing the optimal carbon tax. The "damage function," which links temperature to economic impacts, has come under fire from economists because of its strong assumptions that may produce significant, and ex-ante unknowable misspecifications. I find that learning about damage function parameters can backfire and reduce ex-post welfare over the next 100 years when the policymaker has misspecified the damage function in her model. Combining robust control with learning to guard against misspecifications actually worsens the welfare backfire. This accentuates shortcomings in simultaneously accounting for model uncertainty and Bayesian learning in dynamic models.

中文翻译：

---

损害未知时的最佳气候政策

综合评估模型是经济学家分析最优碳税的主要工具。将温度与经济影响联系起来的“损害函数”受到经济学家的抨击，因为其强大的假设可能会产生重大的、事前不可知的错误说明。我发现，当决策者在她的模型中错误指定了损害函数时，学习损害函数参数会适得其反，并在未来 100 年减少事后福利。将稳健控制与学习防止错误指定相结合实际上会加剧福利适得其反。这突出了在动态模型中同时考虑模型不确定性和贝叶斯学习的缺点。