Quantifying the Restrictiveness of Theories
Quantifying the Restrictiveness of Theories
复制标题
量化理论的限制性
DOI：
--
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
Annie Liang
中科院分区：
文献类型：
--
作者：
D. Fudenberg;Wayne Gao;Annie Liang
我们提出了一个算法量化的经济模型的限制。我们的限制性度量是在模拟的、假设的数据集上进行评估的，这些数据集是从满足一些依赖于应用程序的内容限制的分布中随机抽取的，例如人们应该喜欢更多的钱而不是更少的钱。我们衡量模型的最佳版本（即，给出最低交叉验证预测误差的参数）与最佳可实现的改进相比，改进了诸如随机猜测的朴素预测规则。能够很好地拟合几乎所有数据的模型是没有限制的。我们说明了所提出的方法与两个应用程序：使用累积前景理论预测的确定性等价物彩票，并使用泊松认知层次模型来预测分布的初始游戏。麻省理工学院经济系（Department of Economics，MIT）美国宾夕法尼亚大学经济系。宾夕法尼亚
We propose an algorithm for quantifying the restrictiveness of economic models. Our restrictiveness measure is evaluated on simulated, hypothetical data sets that are drawn at random from a distribution that satisfies some application-dependent content restrictions, such as that people should prefer more money to less. For each such data set, we measure the extent to which the best version of the model (i.e the parameters that give the lowest crossvalidated prediction error) improves on a naive prediction rule such as guessing at random, compared to the best achievable improvement. Models that can fit almost all data well are not restrictive. We illustrate the proposed approach with two applications: using Cumulative Prospect Theory to predict certainty equivalents for lotteries, and using the Poisson Cognitive Hierarchy Model to predict the distribution of initial play in games. ∗Department of Economics, MIT †Department of Economics, U. Pennsylvania ‡Department of Economics, U. Pennsylvania