Joe Grimshaw
Who’re the UK’s mortgage debtors, and the way do their traits differ? Regardless of intensive literature on mortgage profiles, loan-level segmentation stays restricted, present work depends on aggregates or predefined classes. I handle this hole by making use of unsupervised machine studying to twenty years of knowledge, permitting the mannequin decide segments with out prior assumptions. Three clusters emerge: one with low leverage, and two with excessive leverage however notably completely different earnings profiles. Lending composition has shifted step by step. Excessive leverage, high-income debtors now account for a bigger market share, and first-time patrons more and more fall into extra leveraged segments. Machine studying is essential for monetary stability, revealing concentrations of traits, and developments, that aggregates and easy splits can’t, providing richer and earlier indications of potential vulnerabilities.
How can we kind UK mortgage debtors into significant teams?
To know how borrower traits are distributed throughout UK owner-occupier mortgages, I apply k-means clustering to the FCA’s Product Gross sales Database, a loan-level information set protecting all mortgages issued between 2005 and 2025. The algorithm processes 10 variables in complete: mortgage to worth (LTV), mortgage to earnings (LTI), gross earnings, mortgage worth, property worth, rate of interest, mortgage time period, borrower age, debt-servicing ratio (DSR), and web earnings. The groupings are then characterised utilizing three core options: LTV, LTI, and time period size. In my diagnostics, these present the strongest affect on how observations separate into distinct teams, and are sometimes recognised as key determinants of family vulnerability.
Okay-means is an unsupervised machine studying algorithm that teams loans so these inside a cluster are extra related to one another than to these in different clusters. Consider it as an algorithm that appears for pure groupings within the information, quite than being advised upfront what these groupings ought to appear like. It iterates, reassigning loans to their nearest cluster centre and recalculates centres till groupings stabilise. All inputs are standardised earlier than becoming to make sure truthful comparisons throughout completely different scales, and outliers are excluded by eradicating observations above the 99.ninth percentile for every variable.
I take a look at options. DBSCAN identifies clusters based mostly on the density of close by observations, making it nicely suited to irregular shapes however delicate to parameter selections. Hierarchical clustering builds a tree of nested groupings, helpful for visualising construction however computationally demanding at this information scale. Okay-means proves probably the most strong and interpretable for this job, with steady allocations throughout reruns and clear separation that may be communicated to each technical and non-technical audiences. The variety of clusters is guided by the necessity for tractable interpretation, and the utilization of two diagnostics: the elbow technique, which identifies the place including extra clusters yields diminishing enhancements, and silhouette evaluation, which checks how cleanly every mortgage suits its assigned group relative to the others in addition to. All level to a few clusters because the pure answer.
Three distinct borrower segments emerge from the info
Chart 1 exhibits the cluster centre for every group throughout the three core options (LTI, LTV, mortgage time period) and gross earnings.
I assign cluster labels after estimation, based mostly on the three core metrics. The cluster with the bottom common is labelled Group A; the best, labelled Group C; and the opposite, labelled Group B. Whereas these labels are decided mechanically, the ensuing teams exhibit clear and steady profiles. Group A constantly corresponds to low‑leverage lending, Group B combines comparatively high-leverage with higher-incomes, and Group C captures excessive‑leverage borrowing concentrated amongst decrease‑earnings debtors.
Group A (Low Leverage): This phase is outlined by extra conservative borrowing and accounts for a couple of third of the circulation of mortgage lending. The median LTV is roughly 39%, the median LTI is 1.9, and the median mortgage time period is 15 years. The share of lending at increased thresholds, equivalent to loans exceeding 90% LTV or 4.5 occasions earnings, is negligible. Debtors on this group are sometimes older, with increased incomes and substantial deposits, and are much less more likely to be first-time patrons.
Group B (Excessive Leverage, Excessive Earnings): Between 10% and 15% of recent lending every quarter falls into this phase. Mortgages on this group have a median LTV of round 67%, a median LTI of three.2, and a median time period of 23 years. Round 7.5% of loans exceed 90% LTV, and almost 11% are above 4.5 occasions earnings. This cluster primarily consists of extra prosperous debtors accessing higher-value properties, leveraging their earnings and lengthening phrases to take action.
Group C (Excessive Leverage, Low Earnings): Over half of recent lending falls into this phase every quarter, displaying a median LTV of 80%, a median LTI of three.4, and a median time period of 28 years. Round 13% of loans are above 90% LTV, and about 10% exceed 4.5 occasions earnings. Whereas each teams have the same proportion of excessive LTI loans and comparable common LTIs, this group has increased LTV and time period lengths. Debtors are youthful, with decrease incomes and smaller deposits, and the group features a important share of first-time patrons.
For 3 of the 4 metrics in Chart 1, cluster centres observe a sequential order from Group A to Group C. Earnings is the exception. Group B has the best median earnings, adopted by Group A, with Group C lowest. Group B debtors have the monetary capability to service bigger loans however tackle extra leverage to entry higher-value properties. Contemplate a dual-income couple in London shopping for a £700,000 flat with a £550,000 mortgage. Group C debtors take related leverage however with decrease incomes, like a single first-time purchaser buying a £200,000 dwelling on a modest wage. Each teams are extremely leveraged, however their monetary profiles differ markedly.
Chart 1: Common traits by cluster (2022–25)
Notes: Chart exhibits median values for every cluster. LTV and LTI are expressed as ratios. Gross earnings is in £s. Mortgage time period is in years.
How has the composition of lending modified over time?
Group C has constantly represented the most important share of recent lending, whereas Group B has been the smallest. Nevertheless, Group B has grown in prominence, rising its share from round 7% within the mid-2000s to 11% in recent times (Chart 2), leading to a sluggish discount out there share of Group A over time. This gradual shift displays adjustments in each borrower behaviour and market situations.
Chart 2: Share of completions over time
Each Teams B and C, now embrace a higher share of first-time patrons than earlier than the monetary disaster (Chart 3). Group C particularly has seen its first-time purchaser share develop over the previous 10 to fifteen years. This sample is carefully linked to the rise in home value to earnings ratios. As affordability pressures have mounted, first-time patrons have more and more wanted to tackle higher leverage and mortgage phrases to entry the market.
Chart 3: First-time purchaser share over time by cluster
Mortgage phrases have lengthened throughout all segments (Chart 4). Since 2015, the median time period has elevated by two years in every group. In comparison with the mid-2000s, the rise is much more pronounced, as much as 5 years longer in Teams B and C. Longer phrases permit debtors to unfold repayments, decreasing month-to-month outgoings and easing affordability pressures. Nevertheless, in addition they imply increased complete curiosity prices and longer publicity to market fluctuations.
Chart 4: Common mortgage time period size over time by cluster
Regional variations reveal distinct borrower compositions
London and the South East have the bottom share of Group C debtors (Chart 5), but additionally have the bottom share of Group A debtors. Why? As a result of, a cloth share of lending in these areas falls into Group B, the excessive leverage, excessive earnings phase, which is far smaller elsewhere. Larger property costs imply debtors typically want each excessive incomes and huge loans relative to their earnings, with longer mortgage phrases widespread to handle repayments. Group B’s prominence in London and the South East shouldn’t be a latest improvement however a longstanding function of the UK mortgage panorama.
Elsewhere, the image is extra blended. The North of England, Northern Eire, and Wales have a higher share of Group A lending, reflecting decrease home costs, LTVs, and LTIs. The Midlands, Scotland, and the South West present extra Group C lending. A area might seem unremarkable in common ratios, but masks a really completely different compositional story beneath. Quite than summarising debtors by a single common, loan-level segmentation reveals the distinct teams driving it.
Chart 5: 2025 distribution of clusters inside every area
What does this imply for monetary stability?
Understanding who’s borrowing, the place, and the way a lot, is central to assessing mortgage market patterns. This strategy enhances present frameworks by letting borrower segments emerge from the info quite than imposing predefined classes. Easy splits by earnings or area can point out that leverage is rising; they can not let you know that increased leverage and earnings are more and more clustering collectively, or that first-time patrons are concentrating in additional stretched segments, or that areas with related common ratios maintain very completely different borrower mixes beneath. These mixtures matter. Surfacing them early, earlier than they seem in mixture statistics, strengthens the toolkit and information out there for safeguarding monetary stability in one of many UK’s most systemically necessary markets.
Joe Grimshaw works within the Financial institution’s Macro-Monetary Dangers Division.
If you wish to get in contact, please e-mail us at bankunderground@bankofengland.co.uk or depart a remark beneath.
Feedback will solely seem as soon as authorized by a moderator, and are solely revealed the place a full identify is provided. Financial institution Underground is a weblog for Financial institution of England employees to share views that problem – or assist – prevailing coverage orthodoxies. The views expressed listed here are these of the authors, and usually are not essentially these of the Financial institution of England, or its coverage committees.
Share the put up “Demographics, deposits and information: utilizing machine studying to phase UK mortgages”
