Open banking data modelling of gambling spend thresholds

Summary

This publication sets out the results of an analysis of a sample of open banking data. A data modelling approach was taken to consider the proportion of individuals whose rate of gambling spend exceeded certain thresholds. We consider the overlap between the groups of individuals exceeding the different thresholds, as well as whether the thresholds are still reached when considering spend at each gambling operator separately compared with overall gambling spend.

Introduction

In July 2023, the Gambling Commission opened a consultation (opens in new tab) connected to the Gambling Act Review, which closed in October 2023. One aspect of this was a set of questions about new obligations on operators to conduct checks to understand if a customer’s gambling is likely to be harmful in the context of their financial circumstances, in the form of financial vulnerability checks and financial risk assessments.

It was proposed that these would be conducted at certain net loss thresholds. The proposed thresholds for vulnerability checks were £125 in a rolling 30-day period, and £500 in a rolling 365-day period. The proposed thresholds for financial risk assessments were £1000 in a rolling 24-hour period, and £2000 in a rolling 90-day period.

This technical report describes data modelling of these thresholds, as well as describing analyses with the 30-day threshold of £150, conducted with open banking transaction data. The results of this modelling forms part of the evidence base for the consultation response.

Data and model

Open banking protocols enable individuals’ financial data to be shared between banks and third-party service providers.

In 2023, the Gambling Commission purchased a sample of open banking data from YouGov Finance. This dataset includes the details of banking transactions from 4000 individuals in Great Britain, each of whom have given their consent for their data to be used for research. It comprises over 12 million individual transactions, of which 253,916 are non-lottery gambling transactions, by 2034 unique gamblers. At least one year of transactions is available for each gambler, with an average of 966 days and a standard deviation of 222 days. The data spans the period up to August 2023.

The sample was designed to over-represent gamblers and it is not demographically representative. Certain groups, such as men, are over-represented. Others, such as those over 50 years old, are under-represented.

It is not possible to see exact wins or losses in this dataset, as there is no information about gambling account balances. Instead, net deposits to gambling brands over time (debit transactions minus credit transactions) is used as a proxy. Gambling transactions were identified by finding gambling brands and operators in the transaction descriptions. In total, 207 brands, associated with 65 unique operators were identified, including all the largest operators in Great Britain. However, it is possible that not all gambling transactions were identified.

As the exact time that each transaction took place is not available, we consider net spend by calendar day. Thresholds are calculated over one calendar day more than the consulted rolling period to account for transactions occurring either side of midnight. For example, the £1000 threshold is calculated over 2 calendar days and the £500 threshold is calculated over 366 days.

Results

The net spends at each gambling operator over the previous 2, 31, 91 and 366 days were calculated for each gambler in the data, on each day on which they had a gambling transaction. This was compared with the proposed thresholds to identify every individual who would have triggered either a financial risk assessment or a vulnerability check had the proposed policies been in place.

These calculations were repeated for net spend across all gambling businesses, to see how many additional individuals would have been captured by either a vulnerability check or a financial risk check if their total gambling spend was considered. The number of unique individuals exceeding the thresholds was compared for the 2 sets of calculations.

We also considered the number of unique individuals identified as exceeding either of the financial risk check thresholds and the additional coverage provided by adding the other, as well as the equivalent calculation for the vulnerability check thresholds.

The results of these calculations are presented in the following table:

Number of individuals exceeding the financial risk check thresholds

Number of individuals exceeding the financial risk check thresholds

Threshold evaluated and the impact of threshold being omitted	Number of unique individuals exceeding net spend threshold at any single operator	Number of unique individuals exceeding across all gambling transactions	Difference (percentage)
Risk assessment: £1000 in 24 hours	93	98	5%
Risk assessment: £2000 in 90 days	95	114	20%
Either or both risk assessment thresholds	120	135	13%
Reduction in number of unique individuals captured if £1000 threshold was omitted (percentage)	21%	16%	N/A
Reduction in number of unique individuals captured if £2000 threshold was omitted (percentage)	23%	27%	N/A
Vulnerability check: £125 in 30 days	539	554	3%
Vulnerability check: £500 in 365 days	342	350	2%
Either or both vulnerability check thresholds	543	559	3%
Reduction in number of unique individuals captured if £125 threshold was omitted (percentage)	37%	37%	N/A
Reduction in number of unique individuals captured if £500 threshold was omitted (percentage)	1%	1%	N/A

Discussion

When considering the 2 proposed risk assessment thresholds for single operator net spend, we see that similar numbers of individuals exceed each, with 93 and 95 unique individuals for the £1000 and £2000 thresholds respectively. There is a large overlap in the sets of unique individuals identified by each. However, for each of these thresholds between a fifth and a quarter of individuals identified did not exceed the other threshold. This results in a notably higher number of unique individuals exceeding at least one risk assessment threshold, 120, than either threshold alone.

For the 2 proposed vulnerability check thresholds for single operator net spend, we see that a far greater number of individuals exceed the £125 threshold than the £500 threshold, with 539 and 372 unique individuals identified respectively. We note that almost every unique individual exceeding the £500 threshold also exceeds the £125 threshold, with only 1 percent of individuals captured by £500 threshold not also exceeding the £125 threshold.

When considering the differences between evaluating the thresholds separately for each operator to evaluating across all gambling transactions identified in the data, we see relatively small differences in the number of individuals captured by the 2 vulnerability check thresholds and only a 3 percent increase in the numbers exceeding at least one of these. However, the equivalent increase for the risk assessment thresholds is larger, with 13 percent more individuals exceeding at least one of the risk assessment thresholds across all gambling spend than when measuring spend separately for each operator. This difference is less extreme for the £1000 threshold than for the £2000 threshold, with 5 percent more individuals captured by the £1000 threshold when considering all gambling spend compared to 20 percent for the £2000 threshold.

In a follow-up analysis exercise, the 30-day threshold was adjusted from £125 to £150. This resulted in only a minor change to the number of individuals that would meet each threshold (as defined in this exercise). For the single operator net spend with the increased £150 threshold value, 171 individuals would have met only the £150 in 30-days threshold, 8 individuals would have met only the £500 in 365-days threshold and 336 would have met both thresholds (as defined in this exercise). This equates to 1.6 percent of individuals meeting either or both thresholds doing so only through the 365-day limit. These calculations were produced using a slightly updated dataset after a relatively small number of additional gambling transactions (comprising approximately 0.4 percent of the total transactions) had been identified following the initial analysis.

Author:
Steven Murphy
Postdoctoral Data Scientist, Gambling Commission
Honorary research fellow, Warwick Business School