Rescuing Econometrics: From the Probability Approach to Probably Approximately Correct Learning

Haavelmos 1944 monograph, The Probability Approach in Econometrics, is widely acclaimed as the manifesto of econometrics. This book challenges Haavelmos probability approach, shows how its use is delivering defective and inefficient results, and argues for a paradigm shift in econometrics towards a full embrace of machine learning, with its attendant benefits. Machine learning has only come into existence over recent decades, whereas the universally accepted and current form of econometrics has developed over the past century. A comparison between the two is striking, with the practical achievements of machine learning significantly outshining those of econometrics. The relative efficiency of machine learning is based on its theoretical foundation, and particularly on the notion of Probably Approximately Correct (PAC) learning. Econometrics should therefore renounce its outdated foundation and rebuild itself upon PAC learning theory so as to unleash its pent-up research potential. The book is catered for applied economists, econometricians, economists specialising in the history and methodology of economics, advanced students, and philosophers of social sciences.

Format: Hardback
Length: 224 pages
Publication date: 01 December 2023
Publisher: Taylor & Francis Ltd

The Probability Approach in Econometrics, a groundbreaking 1944 monograph by Haavelmos, is widely regarded as the manifesto of econometrics. This influential work challenges Haavelmos probability approach, highlighting its shortcomings and advocating for a paradigm shift in econometrics towards a full embrace of machine learning. While machine learning has emerged relatively recently, compared to the established form of econometrics that has evolved over the past century, a comparison between the two reveals significant disparities. The practical accomplishments of machine learning surpass those of econometrics, substantiating the existence of widespread inefficiencies in current econometric research. The relative efficiency of machine learning is rooted in its theoretical foundation, particularly the concept of Probably Approximately Correct (PAC) learning. A thorough examination reveals that PAC learning theory outperforms Haavelmos probability approach in achieving the goals of applied economic modelling research. Consequently, econometrics should abandon its outdated foundation and rebuild itself upon PAC learning theory to unlock its untapped research potential. This book is designed for applied economists, econometricians, economists specializing in the history and methodology of economics, advanced students, and philosophers of social sciences.

Dimension: 234 x 156 (mm)
ISBN-13: 9781032586052