Augmenting the availability of historical GDP per capita estimates through machine learning

Area of Science:

• Economic History
• Computational Social Science
• Historical Demography

Background:

• Estimating historical Gross Domestic Product (GDP) per capita is crucial for understanding long-term economic development.
• Traditional data sources for historical GDP are often scarce, incomplete, or geographically limited, especially for pre-modern periods.
• Biographical data on historical figures offer a potentially rich, yet largely untapped, source of information about past societies.

Purpose of the Study:

• To develop and validate a machine learning methodology for estimating historical GDP per capita using biographical data.
• To generate novel GDP per capita estimates for European and North American countries and regions over the past seven centuries.
• To assess the reliability of these estimates through external validation with diverse historical proxies.

Main Methods:

• Utilized an elastic net regression model for feature selection and prediction, trained on biographical data (birthplace, death place, occupations) of hundreds of thousands of historical figures.
• Generated out-of-sample GDP per capita estimates for countries and regions lacking traditional economic data.
• Externally validated estimates against urbanization rates, average body height, self-reported well-being, and church construction activity.

Main Results:

• The machine learning model explained 90% of the variance in known historical income levels.
• Generated novel GDP per capita estimates for numerous regions and time periods.
• Validated estimates against four independent proxies, demonstrating strong correlations.
• Successfully reproduced the historical 'reversal of fortune' between Southwestern and Northwestern Europe, linking it to Atlantic trade.

Conclusions:

• Fine-grained biographical data, analyzed with machine learning, provide a robust method for augmenting and extending historical GDP per capita estimates.
• This approach significantly enhances the ability to study long-term economic trends and regional disparities.
• The comprehensive dataset of estimates and source data is made publicly available for future research.