新质生产力下重点工业产品产量预测研究——基于动态分层面板与可解释机器学习

Research on Forecasting the Output of Key Industrial Products under New-Quality Productive Forces—Based on a Dynamic Hierarchical Panel and Interpretable Machine Learning Approach

摘要: 在高质量发展与新质生产力培育背景下，重点工业产品月度产量变化能够反映技术扩散、产业链景气与结构替代。本文以国家统计局公开发布的规模以上工业生产资料为基础，构造重点工业产品的产品–月份短面板，并在原有滞后变量基础上补充增长产品占比、产品销售率、出口交货值增速、高技术制造业溢价等月度外部特征。研究采用动态固定效应、类别异质性检验、正则化回归和受约束集成学习相结合的建模路径，重点比较不同产品类别的增长分化、预测误差结构和模型稳定性。研究表明，新质生产力相关产品具有较高增长水平，但部分产品同时表现出更强波动和更高预测风险；短期预测中，产品自身滞后增速仍是最稳定的基础信号，外部景气变量主要用于刻画月度背景和修正模型偏差。本文的价值不在于简单堆叠算法，而在于以产品层级数据、月度外部变量和误差分解共同支撑产业监测模型。

Abstract: Under the background of high-quality development and the cultivation of new quality productive forces, monthly output fluctuations of key industrial products can reflect technological diffusion, industrial-chain dynamics and structural substitution. Based on public monthly industrial production statistics by the National Bureau of Statistics, this paper constructs a product-month short panel of key industrial products and enriches the feature space with external monthly indicators, including the breadth of products with positive growth, product sales ratio, export delivery growth and the high-tech manufacturing premium. A combined framework of dynamic fixed effects, category heterogeneity tests, regularized regression and constrained ensemble learning is used to examine growth differentiation, predictive errors and model robustness across product groups. The results indicate that products related to new-quality productive forces usually show stronger growth, while some of them also exhibit higher volatility and larger forecasting risk. For short-term prediction, product-specific lagged growth remains the most stable signal, whereas external monthly indicators help describe macro-industrial conditions and reduce interpretive ambiguity. The contribution of this paper lies in strengthening industrial monitoring with product-level data, external monthly variables and structured error decomposition rather than merely adding algorithms.