企业的慷慨因何而来？——基于机器学习的证据

How does the Generosity of Enterprises Come? Evidence from Machine Learning

《财经研究》2023年49卷第6期 页码：153 - 168 online：2023年6月3日

作者

中：陈运森1 , 周金泳1 , 黄健峤2

英：Chen Yunsen1, Zhou Jinyong1, Huang Jianqiao2

作者单位：1. 中央财经大学 会计学院，北京 100081; 2. 浙江财经大学 会计学院，浙江 杭州 310018

摘要及关键词

摘要：慈善捐赠是企业参与第三次分配和实现共同富裕的重要渠道，而现有研究仅关注单一特征与捐赠行为之间的联系，缺乏对不同捐赠动机的比较分析，企业捐赠的主导因素仍有待深入探究。文章基于机器学习中的集成学习方法，综合讨论了多维度捐赠动机特征对捐赠行为预测能力的差异，从而识别出影响企业参与捐赠的主要动机，并找出预测能力最强的特征。研究发现：（1）与战略性动机、政治动机和外部监督压力动机相比，企业慈善捐赠行为主要受内部治理动机驱使；（2）集成学习方法对慈善捐赠行为的预测能力优于传统线性研究方法，其中渐进梯度回归树具有最强的解释能力和最高的预测精度；（3）在多维度动机特征中，管理层薪酬激励、大股东资金占用行为、销售费用率、分析师关注程度、实际业绩表现和商业信用融资对慈善捐赠行为的预测效果最佳。文章不仅运用机器学习方法有效识别了企业慈善捐赠的关键因素，而且对完善三次分配制度、推进共同富裕具有重要启示。

关键词：捐赠；共同富裕；机器学习；集成学习；社会责任

Summary: Charitable donation is an important channel for enterprises to participate in the third distribution. As a micro way to achieve the goal of common prosperity, charitable donations have become increasingly important. Benefiting from the government attention and the widespread concern on ESG, Chinese enterprises have actively participated in charitable donations in recent years. However, the real motivation of corporate charitable donations in China has always been one of the most controversial topics in practical and academic levels. Existing literature mainly focuses on the relationship between single characteristic and donation behavior, and makes predictions only within the sample, lacking a comprehensive consideration of different charitable donation motivations. Therefore, the intrinsic driving factor of corporate charitable donations is still a fundamental issue and remains to be further explored. According to previous literature, this paper divides the charitable donation motivation into four categories, namely, strategic motivation, political motivation, internal governance motivation, and external supervision pressure motivation. By using XGBoost, gradient boosting regression tree and random forest in the integrated learning method, this paper finds that: (1) Compared with strategic motivation, political motivation, and external supervision pressure motivation, corporate charitable donations are mainly driven by internal governance motivation, which indicates that corporate donations are mainly affected by internal governance factors. (2) The predictive ability of ensemble learning method for corporate charitable donations is better than that of traditional linear research methods, and the gradient boosting regression tree has the strongest explanatory ability and the highest prediction accuracy. (3) Among the multidimensional motivational characteristics, management compensation incentive, large shareholders’ fund occupation behavior, sales expense ratio, analyst attention, real performance, and trade credit financing have the best prediction effect on corporate charitable donations. The possible contributions of this paper are as follows: At the theoretical level, by systematically testing corporate charitable donation motivation, it evaluates and compares the predictive ability of different dimensions of motivational characteristics for corporate charitable donations, and answers the basic question of the internal factors driving corporate donations, which enriches the research in the field of corporate charitable donation motivation. At the methodological level, it applies the ensemble learning in machine learning to the study of charitable donation motivation for the first time, and constructs a model with stronger explanatory ability and higher prediction accuracy for corporate donations, which enriches the application scope of machine learning in the field of accounting and finance, and lays a good foundation for further prediction analysis. At the practical level, it shows that in order to support more enterprises to participate in charitable donations, policymakers can guide listed companies to strengthen internal governance, and increase the emphasis on the dominant factors of charitable donations.

Key words: donations; common prosperity; machine learning; ensemble learning; social responsibility

其他信息

DOI：10.16538/j.cnki.jfe.20221217.102

收稿日期：2022-07-06

基金项目：国家自然科学基金项目（72272168，72002189）；浙江省自然科学基金项目（LQ21G020008）

点

这里“阅读原文”，查看更多