Methodology
Consumer Current transforms millions of anonymous credit and debit card transactions into actionable sales estimates for companies. Our methodology includes several key modeling steps such as panelization and projection. These steps allow us to take raw transaction data from over 250 million consumer accounts and produce estimates that reflect the broader population's spending behavior. The large sample size of the underlying panel ensures that even smaller demographic groups can be analyzed with a high degree of accuracy. Additionally, our projection model extends beyond the tracked data to account for unobservable channels such as cash transactions.
User Selection and Panelization
The composition of a panel changes over time, both in terms of panelist numbers and demographics. Cohorting stabilizes the composition and count of panelists within a cohort and excludes those who are infrequently active or only active for a short period. Cohorting allows us to generate stable, comparable sales growth estimates that are independent of fluctuations in the size of the underlying consumer panel. To be included in the panel, a card must have been active for at least 24 months.
Weighting and Projection
Weighting is the process of adjusting the panel data to ensure it accurately reflects the broader population. This is done by rebalancing the composition of panelists based on known attributes, such as state-level population estimates reported by the Census Bureau. Projection, on the other hand, involves extrapolating sales estimates from the panel to represent the entire U.S. population. By combining these two methods, we generate consumer spending estimates that are representative of sales across a wide range of businesses, ensuring that even smaller demographic or regional differences are captured effectively.
Nowcasting
To provide a real-time view of company performance we use a nowcasting approach. This method predicts the full current month, quarter, or year based on data from only part of the period. Note that we do not apply nowcasting to weekly data, and we only nowcast longer periods when we have at least seven days of complete data.
Benchmarking
We benchmark the data to publicly available government sources, filings from publicly traded companies, and proprietary data from merchants. An example for benchmarking Consumer Current estimates with SEC filings is available here