Payment Data Generally Improves Economic Nowcasting Across Comparative Empirical Studies

A systematic literature review published in Frontiers in Big Data reveals that incorporating payment and transaction big data into economic nowcasting systems generally improves out-of-sample predictive accuracy for household consumption, retail trade turnover, and broader macroeconomic activity. Led by researchers from the University of Patras in Greece, the investigation systematically examined peer-reviewed empirical studies published between January 2015 and August 2025. Across the analyzed literature, comparative forecasting evaluations demonstrated that high-frequency transaction flows provide vital incremental information before conventional national accounting statistics and official surveys are published. However, the synthesis established that the magnitude and timing of these predictive improvements vary substantially across national settings, underlying data sources, target variables, and empirical modeling choices. Crucially, the authors found that apparent forecasting gains depend heavily on the realism of the evaluation framework, particularly regarding whether researchers rigorously account for real-time publication lags, data vintages, and historical revisions.

The underlying motivation for leveraging payment records stems from the persistent temporal lag that characterizes traditional macroeconomic reporting. Standard official indicators, including quarterly gross domestic product and monthly private consumption statistics, are typically released weeks or months after the close of the reference period, leaving economic policymakers and corporate planners in an informational void during periods of rapid structural change. Economic nowcasting addresses this challenge by generating contemporaneous estimates of the present, very recent past, or near-term trajectory of economic activity before official statistical agencies release their first estimates. As modern economies transition increasingly toward electronic transactions, records generated through debit and credit card swipes, automated teller machine withdrawals, point-of-sale terminals, and interbank clearing systems capture activity at daily or weekly frequencies. These granular transactional streams offer an operational timing advantage over conventional surveys, providing high-frequency behavioral snapshots that enable forecasters to track aggregate demand dynamically as transactions occur.

To establish the evidentiary strength of transaction-based forecasting, the research team conducted a systematic literature review adhering strictly to the PRISMA 2020 reporting guidelines. Systematic searches were executed across major multidisciplinary databases, primarily Scopus and the Web of Science Core Collection, supplemented by targeted searches across publisher platforms including ScienceDirect, SpringerLink, Emerald Insight, and Taylor and Francis, along with forward and backward citation chaining. The search protocol yielded 1,580 initial records, which were deduplicated to 1,538 entries and screened across titles, abstracts, and keywords. After excluding non-empirical publications, conference papers, books, and out-of-scope topics, the reviewers retrieved 677 full-text reports, ultimately identifying 11 peer-reviewed empirical studies meeting all predefined eligibility criteria. Because of substantial variation in target metrics, sampling frequencies, and performance measures, a quantitative meta-analysis was deemed inappropriate, prompting the team to apply a structured narrative synthesis and domain-based methodological quality appraisal.

The 11 included empirical studies spanned diverse national economies, reflecting marked variations in payment infrastructures, consumer spending habits, and data-collection architectures across Europe and North America. Direct applications to household consumption and retail trade turnover were conducted in Norway, Germany, Spain, Portugal, and Latvia, while broader macroeconomic evaluations targeting gross domestic product and monetary aggregates were undertaken in Canada, Turkey, Italy, and the United States. The underlying data sources exhibited substantial structural heterogeneity, ranging from institution-specific banking datasets, such as Spanish BBVA card records and Turkish Garanti BBVA transactional flows, to proprietary commercial panels like German Fable Data credit card transactions. At the opposite end of the spectrum, several investigations leveraged comprehensive, near-system-wide settlement infrastructures, including the BankAxept domestic debit-card network in Norway, central bank card records covering approximately 98 percent of domestic transactions in Latvia, and nationwide clearinghouse settlement aggregates within Portugal, Spain, Italy, and Canada.

Across the comparative empirical forecasting literature, incorporating transaction signals produced statistically and economically meaningful reductions in forecast errors relative to conventional baseline benchmarks. In Norway, an evaluation by Aastveit and colleagues analyzing weekly domestic debit card volumes through autoregressive distributed lag models identified accuracy improvements of up to approximately 60 percent compared to quarterly autoregressions by quarter end. In Portugal, Duarte and coauthors demonstrated that augmenting models with automated teller machine withdrawals and point-of-sale payments yielded forecast error reductions of approximately 35 percent for quarterly private consumption nowcasts. Similarly, in Canada, Chapman and Desai observed that dynamic factor models incorporating payment system aggregates decreased predictive errors by up to 25 percent, while machine learning algorithms trimmed errors by an additional 20 percent at the nowcast horizon. Galbraith and Tkacz further confirmed that Canadian debit card transaction flows significantly lowered nowcast errors for gross domestic product and retail sales growth.

A critical theoretical insight highlighted by the review is that the predictive value of payment big data is not concentrated at a single point within the nowcasting cycle. Instead, the evidence reveals two distinct and complementary informational patterns across the evaluation horizon. First, transaction datasets provide their greatest marginal advantage during the earliest phase of the reference period, when conventional official surveys and hard statistical indicators remain virtually nonexistent. For example, Barlas and colleagues observed that transaction-based proxies in Turkey generated their strongest predictive gains during the initial 45 days of the quarter. Second, repeated intra-period evaluations demonstrate that absolute forecast accuracy steadily improves as additional transactional observations accumulate toward the official release date. Studies by García and coauthors in Spain, Brinke and colleagues in Latvia, and Conesa and associates in Spain confirmed that tracking incoming card flows progressively sharpens precision, directly reflecting the growing completeness of the transactional signal.

Regarding analytical methodology, the systematic review demonstrated that algorithmic complexity is not a strict prerequisite for achieving superior nowcasting performance. Well-established econometric specifications, particularly bridge equations, mixed-data sampling regressions, and dynamic factor models, consistently delivered strong predictive gains across multiple macroeconomic applications. By accommodating different data sampling frequencies without requiring crude temporal aggregation, mixed-data sampling specifications effectively captured the relationship between high-frequency transactional inputs and lower-frequency quarterly consumption targets in Portugal and Norway. While machine learning techniques, such as random forests, gradient boosting, and artificial neural networks, provided valuable non-linear modeling capabilities in high-dimensional settings—most notably in Canadian macroeconomic studies—the empirical gains were primarily driven by the rich informational content of the payment data rather than algorithmic sophistication alone. Consequently, the authors concluded that parsimonious mixed-frequency econometric frameworks remain fully sufficient for extracting timely signals when datasets are closely aligned with target variables.

The review emphasized that the methodological credibility of nowcasting research depends fundamentally on the realism of the out-of-sample evaluation design rather than model choice in isolation. In operational forecasting environments, practitioners must generate projections using only contemporaneously available information sets, whereas retrospective evaluations relying on final-vintage data risk introducing look-ahead bias because official macroeconomic series frequently undergo substantial revisions after initial publication. Studies implementing genuine real-time or carefully reconstructed pseudo-real-time designs, such as those conducted by Barnett and colleagues in the United States and Chapman and Desai in Canada, explicitly accounted for historical data releases and revisions. Furthermore, Brinke and colleagues in Latvia implemented an exceptionally stringent out-of-sample evaluation framework replicating authentic forecasting workflows. Methodological quality appraisals indicated that studies combining recursive out-of-sample updates, explicit benchmarks, and formal statistical accuracy tests provided significantly more reliable evidence than evaluations relying on contemporaneous in-sample correlations.

Despite the widespread predictive benefits reported in the literature, the authors identified critical constraints regarding data coverage, representativeness, and cross-country generalizability. Payment datasets are operational administrative byproducts created for transaction execution rather than statistical inference, meaning that large transaction volume cannot automatically compensate for demographic or merchant selection bias. Data derived from single financial institutions inherently reflect the institution’s proprietary market share, regional client mix, and customer profiles, which may diverge from the broader population. Moreover, consumer payment preferences vary dramatically across borders; economies with near-universal digital card adoption exhibit structural relationships between electronic transactions and consumption that cannot be assumed in cash-reliant economies. The review also highlighted that structural shocks and shifts in consumer habits can alter baseline parameters over time. Additionally, because many included investigations relied on proprietary commercial bank records, independent replication and standardized cross-country comparisons remain constrained by data accessibility barriers.

Addressing the broader implications for real-time decision support, the synthesis carefully separated demonstrated predictive accuracy from organizational decision outcomes. The authors concluded that payment-based nowcasts currently provide valuable empirical inputs for enhancing situational awareness, macroeconomic monitoring, and short-horizon risk assessment, allowing organizations to update expectations weeks ahead of official statistical releases. However, none of the 11 reviewed studies empirically tested whether managerial actions taken in response to these early signals—such as adjustments to corporate budgets, operational staffing, or marketing expenditures—actually improved business performance. Timeliness provides practical utility only when paired with demonstrated signal stability, transparent communication of uncertainty, and rigorous model validation. Ultimately, the researchers established that while transaction big data represent an indispensable, timely complement to traditional economic statistics, future empirical research must conduct prospective decision trials to establish whether acting on high-frequency nowcasts translates into measurably superior organizational outcomes.

Subject of Research: Systematic literature review evaluating payment and transaction big data for economic nowcasting and real-time decision support

Article Title: Payment and transaction big data for economic nowcasting and decision support: a systematic review of predictive performance and real-time evaluation

Article References: Theodorakopoulos, L., Theodoropoulou, A., Klavdianos, C., Garmpis, S., & Garmpis, A. (2026). Payment and transaction big data for economic nowcasting and decision support: a systematic review of predictive performance and real-time evaluation. Frontiers in Big Data, 9, Article 1729674. https://doi.org/10.3389/fdata.2026.1729674

Image Credits: AI Generated

DOI: 10.3389/fdata.2026.1729674

Keywords: economic nowcasting, payment big data, transaction data, real-time evaluation, private consumption, retail trade turnover, PRISMA 2020, mixed-frequency forecasting, MIDAS specifications, bridge equations, dynamic factor models, data vintages, situational awareness, decision support systems, out-of-sample forecasting, high-frequency economic indicators, bank transaction data, macroeconomic monitoring