Paper presentation at the RISE8 Workshop (MPI Munich)

On Dec 16, Tom Dudda presented a recent working paper titled „Evaluating Machine Learning in Business and Economics Research: Transparency, Predictive Power, and Practical Costs“ (joint work with Lars Hornuf) at the 8th Research on Innovation, Science and Entrepreneurship Workshop (RISE8) held at the Max Planck Institute for Innovation and Competition in Munich.

In their paper, Tom Dudda and Lars Hornuf screen over 50k academic publications in leading business and economic journals for articles related to machine learning. They find that predictive machine learning studies are often intransparent about the relative predictive performance compared to less-complex and less-costly traditional statistical models. These studies receive fewer citations, arguably due to a less rigorous analysis.

Because of opaque reporting practices, it’s often unclear whether the predictive gains of more complex models justify their increased costs. To better evaluate their true economic value, the paper advocates for standardized and transparent reporting that relates a model’s predictive performance to its costs—financially, environmentally, and in terms of a loss in explainability and interpretability.

You can access the working paper here: https://dx.doi.org/10.2139/ssrn.4981802.* 

Link to the workshop website: https://www.ip.mpg.de/en/research/innovation-and-entrepreneurship-research/rise-workshop.html.

*A previous version of the paper circulated under the title The Perks and Perils of Machine Learning in Business and Economic Research. The revised version of the paper will be available soon on SSRN.