EcoSta 2021: Start Registration
View Submission - EcoSta2021
A0460
Title: Machine learning and big data in econometrics: A machine learning-based specification test Authors:  Gilles Hacheme - Aix Marseille Univ, CNRS, AMSE (France) [presenting]
Abstract: Machine Learning (ML) and Big data become ubiquitous in many scientific fields. But their contribution to social sciences is not yet evident. Indeed, the massive flood of data generated by the growing use of the internet is revolutionizing social sciences and particularly Economics. There is an increasing number of research papers in Economics analyzing different web platforms to understand interactions inside those markets better. While the data sphere has been exponentially growing, ML methods have shown their relevance in extracting information from those massive data. The aim is to show the potential benefit of ML techniques for Econometrics clearly. We suggest some way ML can be used in Econometrics for better model specification. Indeed, the increasing popularity of ML models is related to their ability to give better forecasting/prediction results than structural (parametric) models. ML models are known for their ability to capture very complex interactions and non-linearities. The downside is often the poor explainability of their results. Nevertheless, instead of opposing the structural models to the ML ones, we can use the latter as a benchmark to improve the first ones (the structural models) that are far better in terms of explainability. So, we suggest the use of ML to specify better parametric models often used in Economics.