Explainable Ensemble Forecasting of Corporate Green Innovation for Managerial Decision Support
Year: 2026
Type: preprint
Abstract: Combining heterogeneous firm information can improve prediction without necessarily changing managerial review priorities. We develop an explainable ensemble framework integrating innovation history, financial resources, artificial intelligence (AI) assets and inventions, and climate-related disclosures to forecast green invention applications among Chinese nonfinancial listed firms. Expanding historical windows separate training outcomes from later forecast cohorts, and historical out-of-fold predictions train a nonnegative ridge stacking layer. Across 10,447 test observations with outcome years 2021–2023, stacking achieves a coefficient of determination of 0.635 and root mean squared error of 0.7297, compared with 0.612 and 0.7525 for a financial ridge benchmark. Matched ensembles achieve root mean squared errors of 0.7422 with a common information core, 0.7381 after adding climate language, and 0.7356 after adding AI information. Each block contributes conditional predictive information. A ridge model using all 16 predictors attains root mean squared error of 0.7527. Feature attribution identifies innovation history and AI inventions as prominent predictors, while accumulated local effects reveal a rising and flattening software response. XGBoost slightly improves on stacking under squared error, whereas persistence achieves the lowest mean absolute error. Continuous forecasting gains produce only small changes in fixed-capacity review rankings. The study provides an empirical computational framework distinguishing information value, functional flexibility, explanation, and review objectives in managerial forecasting. Mathematics Subject Classification (2020): 62M20; 62P20; 68T05
Source: Research Square
Author: Huan Chen
Institution: Guangdong University of Technology
Language: English
Subfield: Marketing
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7236673589
