We review the paper “Macroeconomic Predictions Using Payments Data and Machine Learning,” published by the Bank of Canada.

QUICK VIEW

• This paper tests whether integrating transaction data into machine learning models improves the accuracy of short-run forecasts for GDP, Retail Trade Sales (RTS) and Wholesale Trade Sales (WTS).
• This hypothesis is tested in Canada using data from 2004-2020, encompassing periods of crisis, namely, the COVID-19 pandemic and the Great Financial Crisis of 2008.
• The authors find that including payments data with traditional economic measures (e.g. unemployment) enhances the accuracy of short-term forecasts for three macroeconomic metrics: GDP, RTS and WTS. Furthermore, the benefit of integrating payments data is shown to be strongest during periods of economic crisis. 
• Finally, the authors show that certain payment types, such as point-of-sale transactions (POS) and pre-authorised debit payments (e.g. mortgage repayments and utility bills), provide the most significant improvement to model forecasts.