We provide an innovative application of explainable artificial intelligence to economic panel data. We apply boosted trees in combination with Shapley values to achieve post-model explanations. As a benchmark, we assess a pooled regression approach to discuss the economic information content of interpretable machine learning.

The investigated data set comprises daily Value-at-Risk figures of 594 American companies as well as innovative explanatory variables. Our results provide evidence that the applied machine learning setup describes a fruitful alternative to interpretable regression approaches. Also, innovative economic insights are generated via Shapley values.

我们为经济面板数据提供可解释的人工智能的创新应用。我们将提升树与 Shapley 值结合应用以实现模型后解释。作为基准，我们评估了一种汇总回归方法来讨论可解释机器学习的经济信息内容。

调查数据集包括 594 家美国公司的每日风险价值数据以及创新的解释变量。我们的结果证明，所应用的机器学习设置描述了可解释回归方法的富有成效的替代方案。此外，创新的经济见解是通过沙普利价值观产生的。