New techniques like deep learning and large language models (LLMs) are increasingly applied in everyday life and policy contexts. Quantifying the uncertainty surrounding predictions based on these models remains an underexplored field. The research project by André Lucas, Professor of Financial Econometrics, pushes the frontier on three fronts.
First of all, Lucas will construct hybrid models exploiting the best side of econometric and machine learning models and apply them to study the spread of fine particulate matter. Furthermore, he will construct bands of uncertainty for these models. And finally, Lucas will construct frames of uncertainty for models for unstructured data like text and picture predictions to study the quality of bank's climate risk reporting.
The research has societal relevance. Lucas: “Most parts of society are amazed by the power of new machine learning techniques and the predictions they come up with. At the same time, most of us have also experienced situations where the machine fails miserably at making a correct estimate of what we are after. This underlines a need for some measure of accuracy of the predictions formed by these new tools, such that we know better when to lean on a prediction, or when to be more skeptical.”
The grant will allow Lucas to recruit a PhD student and several Research Assistants to implement this research agenda, develop software, and distribute that in an accessible way.
About the NWO Open Competition – Social Sciences and Humanities (SSH) grant
With this grant, NWO Social Sciences and Humanities wants to offer researchers the opportunity to carry out research into a subject of their own choosing without any thematic constraints. The funding instrument aims to serve a broader group of researchers in different stages of their academic careers.