Tracking the COVID-19 crisis with high-resolution transaction data

GAICIL Insights

Official statistics describe last quarter.¹ Crises unfold this afternoon. Tracking the COVID-19 crisis with high-resolution transaction data closes that gap with an asset hiding in plain sight: the digital exhaust of everyday payments. The authors use billions of card transactions from a large bank, Banco Bilbao Vizcaya Argentaria, as an alternative source for measuring consumption. Validated against official consumption measures, the data complement national accounts and surveys with a timely, granular view of economic activity.

Point that lens at the first Spanish lockdown in early 2020 and the picture is vivid. Consumption responded strongly to business closures, with smaller effects from capacity restrictions. Spending fell more steeply in rich neighborhoods. Residents of lower-income neighborhoods had higher mobility, which correlated with increased disease incidence. The closing and opening of establishments had a dramatic effect on spending, which reacted abruptly to both.

Each finding is a policy memo. The gap between closures and capacity limits shows that the design of public health measures shapes their economic cost. The steeper drop in rich neighborhoods reveals that the composition of consumption matters for reading a shock. The link between mobility, income and disease exposes how economic constraints shape exposure to risk. None of this could have been seen as quickly or as sharply through traditional surveys.

The bigger story is a revolution in measurement. The growing abundance of detailed financial transactions stored by banks and payment systems, the authors write, is potentially transformative for economic measurement. National statistics agencies are at the earliest stages of engaging with non-traditional data, and the results show the value of pairing surveys with naturally occurring transaction data. Such data give policymakers timely signals about shocks and interventions, vital in moments of high uncertainty and rapid change.

Add artificial intelligence and the possibilities multiply. Billions of transactions demand computational methods to clean, classify and interpret them. Machine learning can map merchant categories, flag anomalies and produce near-real-time indicators at fine geographic levels. The study proves the value of the raw material; AI makes it feasible to turn it into routine public statistics. The result could be an economic weather service, issuing forecasts and alerts as conditions shift, so that crisis policy runs on evidence.

The most exciting implication is for poorer countries. The authors stress that such efforts are especially important in low- and middle-income countries, where standard high-quality, high-frequency consumption indicators may be too costly to produce. Digital payments are spreading fast, and with them the potential for cheap measurement. AI-assisted analysis of transaction data could hand such countries statistical capabilities that would otherwise take decades and great expense to build. A country could leapfrog straight to a real-time statistical system.

Governance is the hinge. Transaction data are private and sensitive. Public use depends on aggregation, anonymization and clear access rules. The study shows that data aggregated at the level of neighborhoods and categories deliver powerful insights. Frameworks permitting such aggregate uses while protecting individual privacy will determine how much value is realized. Rules banning any secondary use of payment data would forfeit the gains documented here; rules ignoring privacy would erode trust. Well-designed partnerships between banks, statistical agencies and researchers offer a constructive path between those failures.

The final lesson is about preparedness. The authors frame their findings as relevant whenever economic activity is restricted to control disease. Measuring an economy in real time is a form of resilience, and building it takes investment in data partnerships, analytical methods and institutional willingness to use unconventional sources. Next time, the question is whether that capacity will already be in place when the shock arrives, or improvised after it.

Something for policymakers to sit with: the best economic indicator of the next crisis may already be sitting on a bank's servers. The raw material exists in the payment systems of modern economies. Innovative measurement, increasingly powered by AI, can turn it into public knowledge at the speed decisions require.