The Great British Honesty Tax

Derek has, across the course of this series, suffered a procession of indignities. We took the No Claims Discount from his kitchen drawer and demonstrated it to be a marketing construct in actuarial costume. We took the meerkat from his mantelpiece and demonstrated it to be a revenue-extraction mechanism in a plush suit. Colleagues wrote in, on both occasions, to complain that this was hard on Derek. They were, on both occasions, not wrong. But I come today not to bury Derek. I come today to explain to the industry why it should be afraid of him.

The Innocent Question

When Derek next opens a price comparison website, he will be presented, amid the usual catechism of vehicle registration and postcode and date of birth, with one apparently innocent question.

How do you normally pay for your insurance? Monthly? Or annually?

Derek pays monthly. He does not have seven hundred pounds lying about in April, or indeed in any other month, and he never has. He will tick monthly. And at the moment he does so — before a single quote has been returned, before a single underwriter has glanced at his postcode, or his parking arrangements, or his blameless fifteen years behind the wheel — his premium will rise.

Not his interest rate. His premium. The base price of the policy itself will be inflated by the mere act of answering, honestly, a question about his future intentions. He will then be charged interest on the inflated figure. He is, in the elegant language of the trade, being double-dipped: charged more for the product, and then charged more for the credit on the product for which he has already been charged more.

Let us be specific, because vagueness is the natural habitat of this industry and specificity its natural predator. In June 2024, Insurance Post ran identical quote requests on a major comparison site. Same man. Same car. Same address. Same date. Two browser windows, differing in precisely one respect: one said annual; the other said monthly. Many household names all returned a higher annual premium — a higher lump-sum price for identical cover — for the customer who had confessed to a preference for paying in instalments. He was penalised for the confession, not the act. The confession was the act.

A spokesperson for one of the participating insurers, in a moment of candour so startling one suspects it was accidental and cannot have been repeated, told Insurance Post that major market participants were loading core premiums for monthly-declaring customers “in addition to the interest charged,” and that the loading was, in some instances, “greater than the interest payment.” The hidden charge is larger than the disclosed one. The interest rate — the number Derek might conceivably notice and resent — is the smaller extraction. The larger one lies buried in the premium he was quoted before the word “interest” had even been uttered.

The likes of Admiral and Allianz charge the same premium regardless of the answer. They have demonstrated, by the simple expedient of remaining solvent, that the practice is not actuarially necessary. It is merely profitable.

The Defence Considered

Now. The actuarial defence. Before I dismantle it, I should declare my interest. I write as an actuary; I have spent my working life around the business of pricing UK general insurance, and what follows draws on what is a matter of public record — the FCA’s own Premium Finance Market Study and the trade press — informed by professional understanding of the discipline. It is from that position — and with the particular vehemence reserved for a man watching his own profession making a grave mistake — that I offer what follows. Let us first do the industry the courtesy of stating its case at its strongest.

Monthly payers, as a class, exhibit higher claims frequency and severity. This is not credit risk — that is already priced, quite openly, in the APR — but claims risk. Preference for monthly payment, the argument runs, signals financial stress; financial stress correlates with incidents; incidents cost the insurer money; the loading is therefore justified.

One must concede the correlation. There usually is one, between poverty and misfortune. This is almost the definition of poverty. The question is not whether the correlation exists. The question is whether payment preference tells the insurer anything it does not already know.

It does not. Payment preference correlates with income. Income correlates with postcode — already rated. With age — already rated. With occupation — already rated. With vehicle type — already rated. With credit score, which the insurer has already pulled through a soft search — already rated. Every channel through which monthly preference might predict claims is already running through the pricing model in thicker and more informative form. The standard test — regress the disputed variable on everything else in the model and ask whether the residual still predicts anything of value — has, to my knowledge, never been published for UK motor payment preference. Not by any insurer. Not by any independent actuary. Not by the Institute and Faculty of Actuaries, of which I am a member, and which has produced working papers on seemingly every other rating variable in creation, down to and including the model of one’s telephone. The entire justification for the loading rests on internal model outputs that are asserted but never shown, cited but never scrutinised, referenced but never subjected to a single peer-reviewed squint. What can be asserted without evidence, it turns out, can also be priced without apparent evidence.

The FCA (and here one detects the distinctive scrape of a regulator backing carefully away from its own findings) observed that firms themselves “had different views on whether payment preference was statistically predictive of claims risk.” One firm loads it for vans but not cars. Another loads it for cars but not vans. Some firms do not load for it at all. This is not the signature of a robust actuarial signal. It is the signature of an industry that has stumbled upon a variable it can charge for, and has decided, with commendable corporate discipline, not to investigate too closely whether it should.

The Variable That Cannot Survive Scrutiny

But here is where the matter becomes a problem. Because the question being asked of Derek is not, properly speaking, about Derek’s risk. It is about Derek’s intentions. And intentions, unlike his date of birth, his postcode, or his disc-brake configuration, are subjective, declared, and entirely within Derek’s gift to revise.

The architecture of the question is, on inspection, breathtakingly fragile. The customer is being asked for a stated preference. He is not being asked for a contractual undertaking. He is not signing a declaration. He is not being placed under any obligation, legal or moral, to pay annually if he says he prefers to. The FCA’s own report into this market makes the point with admirable clarity: “Customers who click through on an annual policy can still be offered premium finance by the insurance provider.” The route from a stated annual preference to an actual monthly payment is not closed. It is not even narrow. It is, by the regulator’s own observation, wide open and trafficked daily.

In other words: the loading is being levied on a variable which the customer can — entirely without misrepresentation, entirely within the architecture the industry itself has built — restate at any moment. The question asks for a preference. A preference is what a person would do if circumstances allowed. If Derek would prefer to pay annually but cannot afford to, Derek prefers to pay annually. The question is not material to risk. It is not material to the contract. It is material only to the margin.

This is not a defect that careful drafting can repair. It is the variable’s essential nature. You cannot collect a contractually binding declaration of payment intention at the quote stage, because the quote stage is, by definition, prior to contract formation. The industry has elected to charge a meaningful sum of money on the basis of a non-binding preference statement, and the existence of the click-through-then-switch route means that anybody who notices what is happening can render the loading inapplicable to themselves with a single tickbox change.

The loading depends, in other words, on Derek’s not noticing.

What Happens When Derek Notices

And now we come to the real point. Because Derek, eventually, will notice. He has children. They have phones. The information is not buried in a regulatory annexe; it has been printed in Insurance Post, examined in an FCA market study, and is now being discussed in a national newspaper. The half-life of asymmetric information in a market with WhatsApp groups and MoneySavingExpert is not what it once was.

When Derek notices — and his neighbour notices, and his neighbour’s daughter notices on his behalf — the pricing variable on which the loading is built will begin to behave very strangely indeed. An actuary at one of the offending firms reading the foregoing will feel a statistical chill run down the back of his neck. Because the payment preference variable is about to eat itself, and an actuary can sense an endogeneity crisis coming the way a farmer can sense a storm.

The pool of monthly-declarers will not shrink uniformly. It will shrink from the top. The first to migrate to the annual-declarer pool will be the lowest-risk members of the monthly cohort: the people who pay monthly from convenience rather than constraint, who subscribe to MoneySavingExpert, and who respond to incentives because they possess the bandwidth to recognise one. What remains in the monthly-declarer pool, once these have migrated, is those who cannot or do not migrate: the genuinely poor, the digitally disengaged, the elderly on fixed pensions who do not read investigations into insurance pricing.

When the actuary refits his model on next year’s data, the claims differential between those who declare monthly and those who declare annually will have widened. Not because monthly payment has become a more reliable predictor of accident risk, but because the pricing itself has reshaped the pool. The signal will glow brighter. It will mean less. The loading will increase, and it will mean less still. And the people left paying the loading will be, with the precision of a laboratory experiment, those whom any regulatory framework with the slightest claim to seriousness would most wish to protect.

And there is a further twist, of the kind that those who have built these models from the inside tend to notice: in a great many UK pricing systems, the variable on which the loading is calibrated is not the variable being asked at the tickbox. The model has been trained on whether the customer in fact paid monthly or annually — recoverable from the policy administration system, observable in the back book, factual rather than declarative. The customer at the quote stage, however, is being asked his stated preference. These are not the same variable. They have never been the same variable. They have never been measured on the same population. The industry has been pricing one quantity on the strength of evidence drawn from another, and calling the result actuarial.

And once Derek’s neighbours begin migrating between declared-preference pools while paying in unchanged ways, the distance between the two will widen further still. The customer who declares annual and pays monthly will, in next year’s training data, sit comfortably in the actual monthly pool while having been quoted at the annual preference tariff. The model will learn from him as a monthly payer. The pricing engine will not have charged him as one. A cleaner source of confounding could not be devised if one sat down to design it on purpose. An undergraduate statistics student would catch it. The pricing committees of some large UK motor insurers, evidently, have not.

This is Goodhart’s Law — that when a measure becomes a target it ceases to be a good measure — applied with the serene indifference of a pricing algorithm to the household budgets of the working poor. It is also, one notes with no small satisfaction, a failure entirely of the industry’s own design. A rating factor that depends on the customer not understanding the question is not a rating factor. It is a tax on honesty. And taxes on honesty have a predictable and well-documented failure mode: honest people stop being honest, and the tax comes to rest, at length, only on those too poor or too trusting to recognise that they were being asked a trick question.

The Industry’s Choice

The industry, at this point, has two options, and only two.

The first is to do nothing. To carry on loading core premiums on the basis of a stated preference that any customer can costlessly restate, and to wait for the consumer education which is now well under way to reach the household-budget-spreadsheet apps, the comparison-site review channels, and the consumer-rights coverage on the news. This path leads to the slow collapse of the variable described above: a loading that increases as it explains less, levied on a population whose composition is increasingly indefensible, until the FCA — eventually, reluctantly, after a thematic review or two — concludes that the loading no longer has the “objective and reasonable basis” that its own rules require, and the practice ends not because the industry abandoned it but because it became too obvious to defend.

The second option is to abandon the loading now, while the variable still has the pretence of meaning, and to charge for premium finance through the APR alone — which is the honest mechanism, transparently disclosed, and which adequately compensates the insurer for the genuine credit risk involved. Admiral, Allianz, and some others have demonstrated that a competently run insurer can do this without going bust. The path is not theoretical. It is being walked, profitably, by competitors today.

I do not expect the industry to choose the second option. I expect it to choose the first, because pricing committees do not voluntarily abandon margin, and because the timeline of consumer awareness is longer than the timeline of next quarter’s results. But the industry should be under no illusion about which path it has chosen. The loading is fragile. The signal is corrupting itself in the training data. The customer’s ability to render the loading inapplicable is built into the architecture of the question. And the only thing currently sustaining the variable is the ignorance of the customer.

Quo Vadis, Derek?

Derek, meanwhile — dear, incorruptibly honest Derek — has been sitting at the centre of this arrangement for years without knowing it. His No Claims Discount was a decoy. His price comparison was a rigged auction. And the answer he gave to a genial little question about how he likes to pay was used to inflate the price of his cover before the auction had even begun. He has been, throughout, the willing participant in a market structure that took his honesty as an opening bid.

But he is not the willing participant any more. The literature has reached him. The loophole — which is to say, the fragility built into the variable by the industry’s own design — is now public knowledge. The loading depends on his not knowing. And now he knows.

The meerkat, on the mantelpiece, says nothing. It has, for more than fifteen years, watched Derek being fleeced with a faintly amused expression. It is about to witness something for which this long-time mantelpiece duty has not prepared it: the customer, having understood the game, deciding whether to keep playing.