企业数字化转型的测度难题:基于大语言模型的新方法与新发现

Measurement Problem of Enterprise Digital Transformation: New Methods and Findings Based on Large Language Models

社会各界关于企业数字化转型的重要性已经基本达成共识,但对企业数字化转型的效果存在严重分歧。产生这一现象的主要原因是现有研究对企业数字化转型的测度存在问题:一是测度对象不够统一明确,二是测度方法不够科学准确。这导致很多研究结论不可比较、难以复制和相互冲突。为了更好地处理上述难题,本文运用机器学习和大语言模型构造一套新的企业数字化转型指标。本文首先对2006—2020年上市公司年报中的句子进行人工打标签,然后用标记结果训练和微调包括大语言模型在内的多个机器学习模型,选择其中分类效果最好的ERNIE模型作为句子分类模型来预测全部文本中句子的标签,最终构造了企业数字化转型指标。理论分析和数据交叉验证均表明,本文构建的指标相对已有方法更准确。在此基础上,本文实证检验企业数字化转型对财务绩效的影响。研究发现:第一,企业数字化转型能够显著提高财务绩效,其中,大数据、人工智能、移动互联、云计算和物联网均有明显作用,但区块链并没有明显的作用;第二,只有在财务绩效较差的企业中,数字化转型才能够显著提高财务绩效;第三,企业数字化转型提高财务绩效的主要渠道包括改善效率和降低成本。本文研究对于推动企业数字化转型和实现经济高质量发展具有一定现实意义。

he rapidly growing digital economy in China has become the second-largest digital economy in the world.The importance of digital transformation for enterprises is widely acknowledged,but there are serious disagreements about the effects of enterprise digital transformation.The main reason for this phenomenon is that existing research has problems in measuring enterprise digital transformation,which is reflected in two aspects:Firstly,the measurement objects are neither unified nor clearly defined;secondly,the measurement methods are not scientific and accurate enough.This results in many research conclusions being incomparable,difficult to replicate,and often contradictory.Therefore,to ensure that theory effectively informs practice,academia needs to reach a consensus on the measurement of enterprise digital transformation and work hard to alleviate the problem of inaccurate measurement methods.Only in this way can the confusion be clarified,and the way for theoretical insights be paved.This paper uses an advanced machine learning method,the large language models,to construct a set of digital transformation indicators based on the annual report texts of Chinese listed companies from 2006 to 2020.Specifically,the indicators are measured in five steps.The first step is to sort out the annual reports of listed companies,and use the two parts of the annual report,“Management Discussion and Analysis”and“Table of Contents,Definitions,and Key Risk Indicators”as relevant texts for enterprise digital transformation.The second step is to divide all texts into sentences to form a predictive sentence pool.In the third step,a set of sentences is randomly selected and extracted based on the presence of keywords to constitute a pending tagged sentence library.This library is manually annotated to determine whether companies are using digital technologies including big data,artificial intelligence(AI),mobile Internet,Internet of Things,blockchain,and cloud computing.The fourth step uses supervised machine learning methods with models such as ERNIE and BERT for the training of sentence classifiers.The fifth step is to use the trained ERNIE model to predict sentence by sentence in the predictive sentence pool,to assess whether and which digital technologies are utilized by the listed companies,thereby constructing a new set of digital transformation indicators for enterprises.To verify the effectiveness of the new indicators,this paper conducts comparisons with patent data,regional data and international literature in six aspects,and finds that the digital transformation indicators constructed are highly consistent with reality.With these indicators,this paper empirically tests the relationship between enterprise digital transformation and corporate financial performance,and obtains some new findings.Digital transformation generally improves financial performance(measured with ROA and ROE).Big data,AI,mobile Internet,cloud computing,and Internet of Things improve ROA and ROE,while blockchain has no such effect.For companies with poor financial performance,digital transformation can significantly improve financial performance,while for companies with good financial performance,digital transformation has no significant effect on financial performance.There are two main channels for enterprise digital transformation to im-prove financial performance,namely,improving efficiency and reducing costs.The contributions of this paper can be summarized in three aspects.Firstly,this paper provides a new method for measuring enterprise digital transformation.It proposes a novel approach to construct digital transformation indicators based on the annual reports of listed companies in China.The new indicators promote in-depth research on enterprise digital transformation in terms of research methods,and provide empirical evidence from China for the general digital economics literature.Secondly,this paper reveals impacts of different digital technologies on corporate financial performance and identifies different channels.Thirdly,this paper enriches the application of large language models in economic literature,for there are limited studies using large language models.This article is of great significance for promoting enterprise digital transformation and achieving high-quality economic development.