By Jakob Schneebacher
Although economies are increasingly global, consumers still shop locally for many goods and services. Groceries, cinemas and restaurants are just some examples of retail categories where consumer location and travel distances affect the choice of retail establishments available to consumers.
For researchers and policymakers, this geographical dimension of markets and competition may matter a great deal. But traditional firm-level survey and administrative data can’t usually tell us which consumers retailers are competing for. Customer surveys can be used to get a sense of local markets. But ad-hoc surveys are costly, time consuming and noisy, and only measure specific markets of interest at a particular point in time.
A new ESCoE discussion paper with Samir Doshi, Vicky Hoolohan and Tabitha Lewis, “Estimating geographical retail markets from card spending data”, offers an alternative approach to estimating local markets, using consumer card spending data in narrowly defined retail merchant categories (such as traditional sit-down restaurants, or clothes shops).
How did we do it?
We clustered locations based on card payment flows from consumer locations to merchant locations, within each retail merchant category. We call these clusters “local retail markets” for a given retail category. They capture the intuition that retailers within a market compete for the same customers, and customers within a market choose between the same retailers.
To give a concrete example, Figure 1 shows our estimated local markets for traditional sit-down restaurants in Edinburgh in March 2024. These local markets are geographically connected not by assumption, but as a reflection of consumer choices. Their size reflects transport links and consumer travel behaviour. Figure 2 shows the same markets in the same month, but for Great Britain as a whole. Markets differ significantly in size, shape and retailer counts.
What did we find?
Our method allowed markets to differ by shape and size across locations, without needing to be contiguous or symmetrical. This is an improvement over current local market estimates in competition casework, where the area under investigation is often divided into equally-sized local “markets”. It is also an improvement on industry-based market measures, which are implicitly national.
Since the market definitions in this paper are based on existing card spending data, market definitions can be computed retrospectively, and more frequently, in contrast to ad-hoc consumer surveys. We set threshold values for the hierarchical clustering algorithm using market-specific survey estimates of distance travelled. This ensures that the market definitions align with realistic consumer behaviour.
New insights on local markets:
- We found retail markets that differ systematically by retail merchant category and across space. Demographic and economic characteristics are also highly predictive of market size.
- Markets for different retail goods are also spatially correlated in predictable ways. For instance, apparel and accessories markets are highly correlated with discount stores, home improvement stores and entertainment venues, perhaps capturing suburban shopping malls.
- Geographically, markets are smaller in cities and larger for postcodes with larger car ownership shares. There are important regional differences too.
These findings suggest important improvements are possible over the nationally-uniform, radius-based local market estimates currently in use.
Why does this matter?
Beyond methodological improvements, the new local market estimates improve our understanding of consumer search and behaviour, as reflected in travel distances and spending amounts for retail shopping purposes across the UK, and of local competition in UK retail markets. Estimates of market concentration have played a large role in the recent debate on the causes and consequences of aggregate market power, but concentration measures can be misleading if markets are defined incorrectly. Our paper can help shed light on these measurement issues.
It also has several key uses in policymaking. First, our method allows us to understand how geographic retail markets in the UK have changed over time (including during the Covid-19 pandemic). Second, the degree of spatial correlation between consumers’ retail spending patterns across retail categories gives insight into joint purchase decisions and their drivers, including the product portfolio choices of retailers. Finally, merchant counts within each local market provide a new measure of local competition in local UK retail markets.
Alongside this paper, we have made our local market estimates available in full to the research community. We hope this enables researchers to link to other data sources. We also hope that it generates further insights, from a better understanding of pricing behaviour in non-tradable goods and services, to new insights into the causes and effects of the UK’s increasing service share in the economy.
ESCoE blogs are published to further debate. Any opinions expressed are those of the authors only and do not necessarily reflect those of the Competition and Markets Authority (CMA), the Office for National Statistics (ONS) or the data owner, Visa. This paper uses ONS statistical research datasets via the Secure Research Service (SRS). Outputs may not exactly reproduce National Statistics aggregates.