As part of its AI research series, the FCA has published a report on the potential for pricing differences by demographic characteristics in the mortgages market.
The FCA compared mortgage features like interest rates, loan-to-value (LTV) ratio and lender fees, across the following characteristics:
- age
- sex
- sexual orientation
- ethnic group
- health condition
The research found that there are differences in the types of mortgage products taken out by different groups, which may affect the overall price paid. Statistical analysis showed that:
- Those with a health condition appear to have mortgages with higher initial gross rates of interest on average, but lower upfront lender fees, lower property values and lower household incomes.
- This could suggest that they are more likely to take out products where payments are spreads over time resulting in slightly higher overall prices paid.
- The FCA was unclear whether this difference was driven by consumer choice, or due to the types of mortgages the consumers were able to access, although noted that under the FCA’s rules, customers of regulated mortgage products should always have the choice of paying lender fees upfront or including them in the loan amount.
- Women appeared to have mortgages with similar interest rates and LTV ratios to men, but lower household income and property values. These were likely indicators of wider structural inequalities, such as the gender pay gap.
- Customers who identified as either lesbian, gay, bisexual or asexual appeared to have mortgages with higher interest rates and LTV ratios, and lower lender fees and property values, compared to customers who did not identify with these characteristics.
- People from minority ethnic groups appeared to have mortgages with marginally lower interest rates, higher loan amounts, higher household incomes and higher lender fees.
- They were also slightly more likely to take out mortgages with higher loan-to-income ratios.
- The FCA was unable to explore nuances across minority ethnic groups due to small sample sizes in the dataset.
The FCA concluded that it did not find any evidence of differences in mortgage pricing across different demographic groups, but rather that different groups appeared to have different types of mortgage products. However, it did note that it was unable to conclude that there were no issues with the availability of products to customers from different groups.
The FCA built a machine learning model to quantify how influential demographic characteristics were in determining mortgage price. The model worked by predicting the initial gross rate of interest based on a range of mortgage product features, borrower features and macro-economic variables.