Computational Reproducibility in Finance: Evidence from 1,000 Tests.

This article examines the computational reproducibility of 1,008 empirical results addressing six finance research questions, analyzed by 168 research teams in the Finance Crowd Analysis Project (#fincap). It finds that exact reproduction of results by running original code on the same data occurs only 52% of the time, with reproducibility positively associated with researchers’ coding skills and effort, and negatively associated with technical complexity, code complexity, and outlier results. Researchers tend to overestimate the reproducibility and underestimate the difficulty of reproducing their own work. The study identifies common barriers to reproducibility—including bugs, insufficient documentation, and computational constraints—and offers practical guidelines for researchers and discusses policy options for academic journals to improve reproducibility standards in finance research.