Abstract: Big data related research and machine learning methods provide a new way to find the key factors of performance differentiation in China's manufacturing industry and explore the transmission path of key factors to performance differentiation. Based on the ridge regression and LASSO algorithm, we screen out the important influencing factors of Chinese manufacturing performance differentiation, and then construct Bayesian networks to determine the dependence of manufacturing performance differentiation on key factors. The results show that the most important factor affecting the performance differentiation of Chinese manufacturing enterprise is capital productivity. The manufacturing industry in China follows a performance growth model of “capital efficiency-driven” rather than “capital quantity-driven”. The leverage ratio is relatively less important among the factors influencing the performance differentiation of enterprises in the raw materials industry. The capital-intensive characteristics of equipment manufacturing industry determine the important role of capital cost and asset liquidity in the performance differentiation of enterprises. As to consumer goods industrial enterprises, whether they can achieve economies of scale is also a significant factor affecting their performance level.

Key words: manufacturing industry, performance differentiation, order of importance, conduction path, machine learning

摘要： 大数据相关研究和机器学习方法为寻找中国制造业绩效分化的关键因素,以及探索关键因素对绩效分化的传导路径提供了新途径。本文基于岭回归和套索算法(LASSO)筛选出中国制造业绩效分化的重要影响因素,构建了贝叶斯网络确定制造业绩效分化对关键因素的依赖程度,结果显示:中国制造业企业绩效分化的最重要影响因素为资本生产率,制造业遵循“资本效率驱动型”而非“资本数量驱动型”的绩效增长模式。原材料工业企业绩效分化影响因素中杠杆率的重要度相对较低,装备制造业资本密集型的特点决定了企业绩效分化中资金成本和资产流动性的主要作用,消费品工业企业能否实现规模经济也是影响其绩效水平高低的重要因素。

关键词: 制造业, 绩效分化, 重要性排序, 传导路径, 机器学习