我国科技企业融资的决定因素研究——基于科创板企业的机器学习分析

本文以手工收集的2019年7月至2020年12月所有申报科创板上市的企业为研究对象,研究了企业层面五大维度数十个因素对我国科技企业融资的预测效果,通过传统的OLS和机器学习降维排序方法,为科创板增量改革的实践效果提供了直接证据。研究发现:企业能否在科创板上市方面,企业研发水平、成长性和公司治理水平这三类反映企业未来发展潜力的特征占据主导作用,且研发水平最为关键,盈利的重要性最低;在众多研发变量中,企业研发人员人数占总人数比最为重要;企业能否在科创板上市和上市后二级市场表现的变量重要性排序各不相同,甚至相反;公司治理方面,国有股份占比的重要性强于其他治理机制。结合研究结论,本文从重视企业未来可持续发展、构建科研人才队伍培育和激励机制以及统筹考虑上市后的市场表现等角度,为科创企业发展提供了政策建议。

Determinants of Financing for High-tech Enterprises in China:Machine Learning Analysis Based on the STAR Market

China's economy has entered a new era, transforming from high-speed growth to high-quality growth. In this new era, scientific and technological innovations are a matter not only of development but also of survival. The construction of an effective multi-level capital market to enable an innovative and developed real economy is currently an important topic. The establishment of the STAR board and a pilot registration system for high-quality capital market transformation and reforms are warranted to integrate science and technology with the capital market.
The study sample includes all of the enterprises that applied for listing on China's STAR Market from July 2019 to December 2020. The financial data of the samples that went public successfully are obtained from the CSMAR and Wind databases, and the data of the samples that failed to go public are manually collected from their preliminary prospectuses. We construct a high-dimensional enterprise characteristic research index with dozens of factors related to enterprise research and development (R&D), corporate governance, growth, profitability, and risk levels. Based on global stakeholders' perspectives, we explore all of these factors' predictive effects on the listing performance of technology enterprises (e.g., whether to be listed, duration from declaration to listing, amount of funds raised through the listing) and their market performance after listing (stock return and liquidity 3 months after listing and return on assets 1 year after listing). This paper provides comprehensive and direct empirical evidence regarding the practical effects of the incremental reforms of the Science and Technology Innovation Board obtained using the traditional OLS approach and a machine learning dimensionality reduction algorithm.
We use the Boosting regression tree model. The basic idea of this model is that we first set a regression tree from the initial training and then train a new basis regression tree to achieve a loss function that gradually decreases with an increase in iterations. Finally, the weighted regression function is obtained by combining multiple basis regression trees. Moreover, the importance ranking of the enterprise characteristics obtained from the predicted variables is determined based on the training model analysis. Accordingly, using the dimensionality reduction method of principal component analysis, this paper divides the five indexes into several principal components to study the predictive ability of several enterprise characteristics on the financing of science and technology enterprises at two levels. In the first level, we integrate the five segmentation variables into the machine learning model to obtain the R&D, governance, growth, profit, and risk models to study the importance of the various variables within the research groups. In the second level, we integrate the principal components obtained through the dimensionality reduction of the five indexes into the machine learning model to obtain the overall model to study the importance of the various variables without research groups.
Through the methods of OLS and machine learning, this study finds that corporate governance, rather than R&D, is the most important factor in predicting whether an enterprise can be listed on China's STAR Market. R&D investment is the most important of all the R&D variables to predict whether an enterprise can be listed on China's STAR Market. The factors that predict whether an enterprise can be listed on the STAR Market and how an enterprise performs on the secondary market after listing are quite different. The proportion of state-owned shares is more important than the other variables of corporate governance.
The policy implications of these findings are as follows. First, we should balance the attributes of science and innovation with enterprise growth, corporate governance, and other factors. Second, we should consider scientific research personnel important in the development of enterprise innovation. It is necessary to fully support innovation-driven talent. Third, we should pay more attention to the indicators of frontier innovation and establish a multi-angle science and innovation attribute measurement system. Fourth, we should introduce relevant policies to support the sustainable development of enterprises after listing to protect the interests of the majority of investors.
This study makes the following contributions to the literature. First, to the best of our knowledge, this paper is the first to study the factors predicting high-tech enterprises' financing by the STAR board. Few studies report on the financing of high-tech enterprises. Second, this paper uses machine learning to assess the predictive ability of enterprises' multi-dimensional characteristics on financing and adopts the dimensionality reduction algorithm to further study the important relationships between dimensions. Third, the literature on corporate financing focuses on companies' financing motivation to issue stocks or bonds, whereas this study further determines the factors that predict the success of a company's listing on the basis of the company's financing motivation.