We incorporate structural modelers into the economy they model. Using traditional moment matching, they treat policy changes as zero probability (or exogenous) “counterfactuals.” Bias occurs since real-world agents understand policy changes are positive probability events guided by modelers. Downward, upward, or sign bias occurs. Bias is illustrated by calibrating the Leland model to the 2017 tax cut. The traditional identifying assumption, constant moment partial derivative sign, is incorrect with policy optimization. The correct assumption is constant moment total derivative sign accounting for estimation-policy feedback. Model agent expectations can be updated iteratively until policy advice converges to agent expectations, with bias vanishing.

中文翻译：

---

平衡假药

我们将结构建模者纳入他们建模的经济中。他们使用传统的时刻匹配，将政策变更视为零概率（或外生的）“反事实”。之所以会发生偏差，是因为现实世界中的代理商将政策变更理解为建模者指导的正概率事件。发生向下，向上或符号偏斜。通过将Leland模型校准为2017年的减税来说明偏差。传统的识别假设，即恒定力矩偏导数，在策略优化中是不正确的。正确的假设是用于估计策略反馈的恒定矩总导数符号。可以迭代更新模型代理商的期望值，直到政策建议收敛到代理商期望值且偏差消失为止。