Predictive modelling the past: a new machine learning method applied to seven centuries of wages
All debates about living standards and wages in the very long run are hampered by a lack of data. The smooth time series that economists and historians frequently use have often been generated with regression models to, quite literally, fill in the gaps left by history. In this paper, we introduce a new method, based in established machine learning techniques, to peer into the past despite the increasing scarcity of data. The key intuition is that, just as data in the future are unknown to today’s modelers, data in the distant past are unknown to historians. We can therefore use state-of-the-art predictive modeling methods and best-practice forecasting techniques to make predictions of historic economic time series with improved accuracy and generalizability. We apply this machine learning approach to the seven centuries of English wage data at the heart of debates on long-run living standards. We find our new predictions deviate from established series, with the largest variation at key points of social change (in the early modern period). We consider the implications for economy history methods and narratives.
Meredith Paker is Assistant Professor of Economics at Grinnell College
this event is free to attend but booking is required.
This page was last updated on 30 June 2024
