Abstract: Machine learning models can extract information in a systemic, comprehensive, and replicable way, creating synthetic proxies for a wide range of variables that cannot be measured otherwise. In this paper, we emphasize that a lot more information and correlation patterns can be extracted from existing historical data using these models. To illustrate our methodology, we study the effects that the Latin Monetary Union had on financial flows among its members in the 19th century, a natural question that has not been addressed because of the lack of data for financial flows during that period. Relying on machine learning techniques, we are able to circumvent these data limitations by reconstructing a proxy for financial flows across 14 countries between 1861 and 1913. Making use of our proxy, we use standard casual inference methods and find that bilateral financial flows increased by 5% between 1865 and 1913 among members of the LMU, and by approximately 15% between 1865-1885, the period during which the Union was most active. Overall, these results provide new insights about the history of the LMU, showing that it did help member countries achieve part of the goals that had pushed them to join the Union in the first place.