Estimating geographical retail markets from card spending data

Accurate market definitions are important for competition agencies, but traditional survey based measures are costly, time-consuming and noisy at low aggregations. This paper explores the use of consumer card spending data to improve the timeliness and accuracy of retail market estimates. With the help of a standard machine-learning algorithm, we cluster spending flows from cardholder postcode sectors to merchant postcode sectors for detailed categories of retail merchants in the UK at a monthly frequency. To decide the thresholds for the clustering algorithm, we use estimates of average distance travelled from traditional survey tools. We find geographical retail markets that differ systematically by merchant good category and across space. Market size is also predicted by demographic and economic characteristics. Over time, market size is relatively stable but shrinks during periods of pandemic-induced travel restrictions. Markets for different retail goods are spatially correlated in predictable ways. Beyond applications to competition agency casework, this method allows researchers to investigate local competition and the impact of technology and government policies on spatial consumer search and purchasing behaviour.