Nowcasting Monthly Consumption Indicators Using Credit Card Transactions: Evidence from Korea

Rising economic uncertainty, together with the publication lag of official statistics, makes timely and accurate assessment of consumption conditions essential for effective and responsive policymaking. As conventional indicators are released with a delay, they hinder early recognition of shocks and prompt diagnosis of the business cycle, which in turn raises the risk of mistimed policy responses. This paper develops a real-time nowcasting framework for Korean private consumption that exploits high-frequency credit card transaction data to anticipate the official Retail Sales Index ahead of its release. Monthly sums of credit card approvals in retail-related categories serve as predictors. The pipeline integrates reversible instance normalization (RevIN) to mitigate distribution shift arising from non-stationarity, a suite of modern deep-learning forecasters (e.g., N-BEATS, TiDE, and Transformer- and linear-based architectures), and a stacking ensemble whose meta-learner aggregates sub-model predictions to improve robustness over any single forecaster. The selected ensemble closely tracked the official Retail Sales Index across nominal, real, and seasonally adjusted measures, and reproduced the direction and turning points of year-on-year and month-on-month growth throughout the evaluation period. Coupling credit card data with normalization-augmented deep-learning ensembles thus enables accurate and timely consumption nowcasts, supporting earlier business-cycle diagnosis and more responsive policy. The framework is being extended toward higher-frequency weekly forecasting and to the nowcasting of investment and production.