Post
Analysis: what the UK P2P numbers actually showed
For about two years this blog published data work rather than opinion. Several UK platforms released open data on their loan books, some of it genuinely detailed, and it was possible to pull it apart and see things the marketing did not say.
The tools have gone. They were built on data feeds from platforms that no longer exist, in a JavaScript charting library that stopped being maintained, and resurrecting them would produce interactive views of a market that ended. What is worth keeping is what they showed, and more usefully what they did not.
What the data work covered
- Loan book analysis on the platforms that published open data, looking at concentration, loan to value distributions and origination volumes over time.
- Provision fund coverage, scraped every few hours from the platforms that published a coverage ratio, to see how the stated cushion moved against expected defaults.
- Secondary market pricing, using one equity platform's published trade data, to see what shares in individual properties actually changed hands for against their quoted valuations.
- Relative platform popularity, using activity on the main independent investor forum as a rough proxy for where retail money and retail attention were going.
- Monthly income reports showing exactly what my own portfolio earned across around twenty platforms.
What it got right
Origination slowdowns were visible before announcements. More than once, a platform's own published data showed lending volumes falling months before anything was said publicly. That turned out to be a genuinely useful early signal, because a platform that cannot originate loans has a funding problem, a demand problem or an underwriting problem, and all three matter.
Concentration was measurable and alarming. Where loan books were published with borrower identifiers, it was often possible to see that a supposedly diversified book was a handful of connected borrowers wearing different names. This was one of the clearest warning signs available, and it was sitting in public data.
Secondary market pricing told the truth before the platforms did. Shares trading persistently below their quoted valuation is the market saying it does not believe the valuation. That gap widened well before the platforms concerned acknowledged an exit problem.
What it missed, and why
Provision fund coverage ratios were the wrong metric entirely. I tracked them carefully, charted them, and treated a healthy ratio as reassuring. The problem is that the ratio compares the fund against expected defaults, and expectation was the platform's own estimate. A fund covering 150 per cent of what the platform thinks will go wrong tells you nothing about a year when three times as much goes wrong. The metric was precise, updated frequently and almost entirely uninformative, which is a combination that inspires far more confidence than it deserves.
Nothing in the loan data showed platform risk. Every tool here analysed the loans. What destroyed capital was platforms failing: an operating company running out of money, or in one case operating on somebody else's regulatory permission. None of that is in a loan book. It is in the filed accounts, the FCA register and the corporate structure, and I was not looking there.
Valuations were taken as inputs. Every loan to value figure in every chart assumed the valuation was sound. That assumption was the single biggest source of loss in UK property lending over the period, and no amount of analysis of the resulting ratios could see past it.
Publishing my own returns encouraged the wrong behaviour. The monthly income reports were popular and they were honest, and they were also a running advertisement for a strategy whose risks had not yet shown up. Reporting a steady 9 per cent every month for two years is a very effective way of persuading people, including yourself, that a thing is safer than it is.
The lesson worth carrying forward
Good data on the wrong question is worse than no data, because it feels like diligence. The numbers that were easy to get, rates, coverage ratios, loan to value distributions, were the ones the platforms chose to publish. The numbers that mattered, whether the operating company was solvent and whether the valuations were real, were harder to find and less satisfying to chart.
The platform pages on this site now start from the filed accounts and the FCA register, and treat the loan book as the second question rather than the first. That is the main thing eight years changed.