经济学中的决策树集成模型:研究综述与方法选择

Decision Tree Ensemble Models in Economics:A Review and Method Selection

导出

摘要机器学习算法正以独特的方式渗透到经济学研究领域。在本文中,我们将聚焦于广泛应用于经济学领域的决策树集成模型(DTEM),并以理论基础为出发点,结合相关文献综述,从多个应用场景如预测能力、特征工程和因果推断等方面,系统地探讨了DTEM在经济学中的应用。我们也将深入研究DTEM相对于一般机器学习方法的优势之处,如决策树模型在建模逻辑上的可解释性、对复杂特征关系的捕捉能力、不容易过拟合和估计结果稳健。此外,我们采用蒙特卡洛模拟方法,对比了双重机器学习和因果森林方法在不同模型设定下因果效应识别的性能。模拟结果表明,因果森林算法在估计精度上表现更出色,而且更适用于各种模型设定和高维数据环境。这一发现为经济学研究者提供了有关选择适当因果推断方法的有益指导,同时也为了解机器学习在经济学中的潜在应用提供了有力的支持。本文旨在帮助经济学领域的研究者利用机器学习工具来解决复杂的经济问题,推动机器学习与经济学研究更好结合。 Machine learning algorithms are making notable inroads into the realm of economic research.This study is dedicated to an extensive exploration of the application of the Decision Tree Ensemble Model(DTEM),a widely embraced tool in the field of economics.We systematically investigate the utility of DTEM within economics across various scenarios,encompassing predictive accuracy,feature engineering,and causal inference,grounded in theoretical underpinnings and substantiated by an extensive review of pertinent literature.Furthermore,we delve into the inherent advantages of DTEM over conventional machine learning methods.These advantages encompass the interpretability of decision tree models within logical modeling,their proficiency in capturing intricate feature relationships,and their capacity to mitigate overfitting during the learning process,yielding robust estimation outcomes.To enhance our analysis,we employ the Monte Carlo simulation methodology to compare the efficacy of double/debiased machine learning and the causal forest approach in recognizing causal effects under diverse model configurations.Our simulation results underscore the superior accuracy of the causal forest algorithm and its suitability for a spectrum of model configurations,including high-dimensional data environments.This discovery offers valuable insights for economists in their selection of appropriate causal inference methods and lends substantial support to the broader prospect of integrating machine learning into the discipline of economics.Ultimately,the objective of this paper is to empower economists with the tools of machine learning to address intricate economic challenges and foster a more seamless integration of machine learning techniques with economic research.

作者方顺超朱平芳 FANG Shunchao;ZHU Ping fang

机构地区上海社会科学院数量经济研究中心

出处《金融发展》 2024年第1期99-114,共16页 Financial Development

关键词机器学习决策树集成学习因果推断蒙特卡洛模拟 Machine Learning Decision Tree Integrated Learning Causal Inference Monte Carlo Simulation

分类号 F224 [经济管理—国民经济]

作者简介方顺超,上海社会科学院数量经济研究中心,博士研究生;朱平芳,上海社会科学院数量经济研究中心,二级研究员,博士研究生导师。