Identifying Business and Consumer Payment Patterns from Bank Account Data: Estimating Partially Observed Distributions from Survey Data

• Activity Number: 119
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Xuemei Liu*+ and Geoffrey R. Gerdes
• Companies: Federal Reserve Board and Federal Reserve Board
• Keywords: Bayesian Estimation ; Imputation ; Allocation ; Regression ; Survey Data ; Financial Data

Abstract

There are around 15,000 commercial banks, savings institutions, and credit unions (collectively, banks) in the United States that handle all of the deposit money and process virtually all of the noncash payments for consumers and businesses, summarizing the aggregate behavior of their customers. In our survey, a large representative sample of banks reported aggregate payment flows, along with the number of customers they served, complimenting the existing data. The reported information can be thought of as grouped sets of observations drawn from two independent bank customer distributions, and variations may help identify the customer distributions. Simple cross-sectional regressions produce reasonable allocations of payments to the different customer types, essentially estimating mean behavior, but do not use all the available information. Using an iterative Gibbs sampling approach, we develop estimates of the separate distributions of consumer payments and business payments using observed subsamples of the distributions with varying and known levels of mixing. The approach leads to a more complete description of payment choices being made by consumers and businesses.