Paid In, Priced Out

Reform UK published Making Welfare Work this week. Labour called it fantasy economics. Mel Stride said the numbers were simply not plausible. Our own Director, Damian Pudner, argued the opposite: that the scale is right and the critics should say which figures are wrong.

So we did what we did with Restore Britain's economic manifesto a fortnight ago. We read the document, went to the appendices, and checked whether the sums reconcile.

They do. That is the first surprise, and it is not what either side of this argument has told you.

First, the credit

The diagnosis is sound and the paper is more candid about it than most. As of August 2025, 9.3 million working-age people in England and Wales were claiming some form of welfare, of whom 6.6 million were on out-of-work benefits. Health and disability spending alone is forecast to reach £110 billion a year by 2031. Reform makes an observation that ought to be uncomfortable for everyone: with unemployment at 4.9%, welfare spending per household is roughly where it was in 2009/10, when unemployment was 7.9%.

That is the real finding, and it is Reform's, not ours. Spending is no longer tracking the labour market. Something structural detached the two, and no party has a convincing account of what.

The appendices are also better than the genre usually manages. They show their working on Törnqvist index reconstruction, cite Feenstra on new-variety bias, disclose a 9.8% appeal-reinstatement assumption, and admit where a number is an estimate. That is more transparency than the Conservatives or Labour have offered on welfare in a decade. It is precisely because the appendices are detailed that the problems in them are findable.

The denominator, including the one we got wrong

Start with what £50 billion is a share of, because most of the coverage has this wrong, and so did we.

Damian's piece measured the saving against a welfare bill of "roughly £334bn a year". That is the total, and the total includes pensioners. Reform's savings come almost entirely from working-age benefits, and the plan explicitly protects those over State Pension age. Measuring a working-age cut against a bill that is majority pensioner spending understates it badly.

Reform's own costing table gives the right denominator, and gives it plainly: forecast welfare spending of £210.7bn in 2030-31, against total net savings of £52.6bn. The table states the resulting figure itself.

That is a far more radical proposition than "£50 billion" conveys, and considerably more radical than "15% of £334bn". Credit where it is due: the number is not buried. It is row three of the summary table. Almost nobody reporting this plan has quoted it.

Problem one: the index that does not exist yet

This is the part of the paper almost nobody has read, and it matters more than the citizenship row.

Buried on page 25 under "Reforming Indexation" is a proposal to change the inflation measure used to uprate working-age benefits every year. The argument is that CPI overstates true cost-of-living inflation, because the ONS does not adequately capture how people switch to cheaper alternatives when prices rise.

Reform would "establish an independent commission to consider updates to the CPI methodology", which would "recommend ways for the ONS to improve the CPI". Then, in the paper's own words:

Read that again. The index is not named, because it does not exist. It is whatever a commission that has not been convened would recommend at some future date. Reform concedes it directly: "The exact savings will depend on the commission's recommendations."

And then it costs it anyway, at 0.6 percentage points a year.

The 0.6 is assembled from three components: 0.16pp of upper-level substitution bias from a Törnqvist reconstruction, 0.30 to 0.40pp for new-variety bias after a 25% haircut, and 0.05 to 0.10pp for substitution between market purchases and household production. The academic scaffolding is real. The problem is what happens next.

It compounds, and that is admitted only in the appendix

The policy section says the effect would "add up to very substantial mismeasurement". The appendix is blunter:

"The saving then compounds. In the first year, the affected benefits are around 0.6% lower than under existing CPI uprating. In the second year, they are around 1.2% lower because the second year's smaller uprating is applied to a benefit level already reduced in the first year, and so on thereafter."

"And so on thereafter" is the entirety of the long-run treatment. There is no cap, no floor, no real-terms protection and no stated cumulative effect at ten, twenty or thirty years. The costing table shows the shape of it. The IFS, which ran it forward independently, found the switch saves £4.8bn in 2033-34 and "would continue to shrink the size of the benefit system indefinitely".

A permanent, compounding reduction in working-age benefit levels, measured against the CPI-uprated baseline the government currently uses, is a major policy. It appears in this document as a technical annex.

Three things that do not hold

One: if CPI is wrong, it is wrong for everyone. The alleged mismeasurement is a property of the index, not of the claimant. If the ONS overstates inflation by 0.6pp a year, it overstates it wherever CPI is used, the State Pension included. Reform applies the correction to working-age benefits alone, and to pensioners not at all. No justification for the asymmetry appears anywhere in 50 pages. A measurement correction applied only where it saves money is not a measurement correction.

Two: they book jobs from it. The employment table on page 28 attributes +22,000 jobs to "Indexation Reforms". Reform could argue that a more accurate index still changes replacement rates against wages, and that is a fair answer. What the line concedes is the thing the policy section avoids saying: this is not statistical housekeeping. Reform expects a lower benefit path to change behaviour, which means it expects the path to be lower.

Three: the home-production component is circular. Part of the 0.6pp rests on evidence that unemployed households shift time into home production, cooking and repairing rather than buying. The appendix cites "the smaller group of people who are persistently unemployed or weakly attached to the labour market and therefore have more scope to substitute time for market purchases". That reasoning is then used to justify a smaller uprating for the unemployed. The claimant's unemployment becomes the evidence for cutting the claimant's benefit.

There is also an odd appeal to authority. Reform cites the Johnson review, which recommended CPIH as the headline measure in 2015. CPIH includes owner-occupiers' housing costs and is not systematically lower than CPI. It is invoked as precedent for a downward correction it does not support.

We have written before about CPI understating what households actually pay, particularly on housing. Reform's argument runs in the exact opposite direction to the one most commonly made about this index. Both cannot be right.

Problem two: a contributory principle with no contributions in it

The citizenship restriction is one of the two dominant savings in the plan, alongside disability, and the language around it is the language of contribution. It describes "foreigners who haven't paid in a penny", people with "no particular ties to this country nor an individual record of contribution", and a welfare state funded "on the understanding that it is there to support our own on a reciprocal basis".

We have a good deal of sympathy with the principle. We have argued it ourselves.

But the mechanism is not a contributions test. It is a nationality test. Nowhere in the 50 pages is there a test of National Insurance record, years of contribution, amount paid in, or length of residence. There is no qualifying period and no transition: the rules "will apply to both new and existing claimants".

Which produces results Reform does not appear to have noticed.

Contributory ESA is the clean case. It is payable only to someone with a qualifying National Insurance record, which is the definition of having paid in. Reform withdraws it from non-citizens in a chapter whose argument is that the problem is people who have not paid in. Bereavement benefits, which normally rest on the contributions of the person who died, go the same way.

The only contributory carve-out that survives is the State Pension. So the plan produces this: a person with thirty years of National Insurance contributions who is not a British citizen loses Universal Credit, Housing Benefit, ESA, PIP, Carer's Allowance, Pension Credit and Child Benefit. A British citizen who has never worked a day keeps all of it.

There is a third slippage worth naming. In the Welfare to Work chapter, "contribute" stops meaning National Insurance and starts meaning twenty hours a week of unpaid work for the council. The same word does two jobs in one document: it justifies excluding foreigners on the grounds of fiscal contribution, and it justifies workfare for citizens on the grounds of civic contribution. The switch is never acknowledged.

How Britain actually lost the contributory principle

The principle Reform wants to restore was dismantled long before any of the people it is targeting arrived.

Beveridge was explicit. The 1942 plan was "first and foremost a plan of insurance, of giving in return for contributions benefits up to subsistence levels, as of right and without means test". You paid in, you drew out, and nobody inspected your bank account.

In 2019/20, the last year the House of Commons Library costed it, contributory benefits accounted for around 9% of social security spending on working-age adults and children, some £8bn. The Library states plainly that the post-war insurance system "has been gradually superseded by means-tested benefits, culminating in the introduction of Universal Credit from 2013". Universal Credit, which has absorbed the legacy income-related benefits and now sits at the centre of working-age support, contains no contributory element at all.

Where contribution still buys something, it buys very little. New Style JSA requires a National Insurance record and is then payable for 182 days. Contributory ESA for those in the work-related activity group was time-limited to 365 days by the Welfare Reform Act 2012. And for the State Pension, as we set out in May, 35 years of paid contributions and 35 years of credits produce the same entitlement.

What replaced contribution was not nothing. It was a set of tests based on time and status.

Look at the last row against the five above it. Reform's proposal is not a break from the drift away from contribution. It is the furthest point along it. Thirty years of policy replaced "what did you pay in" with "how long have you been here", and Reform's answer is to replace it with "what nationality do you hold". The contributory principle is invoked throughout and restored nowhere.

Damian made this point in his piece and it is the right one: lawful residence matters, citizenship matters, and contribution should matter too. A plan serious about reciprocity would build an actual contributions gateway. This plan books £21 billion from a nationality check and calls it reciprocity.

The arithmetic, checked

We went to the consolidated costing on page 27 and read it line by line. What we found is not what we went looking for.

Reform's table is properly constructed. Every measure is shown across the same eight fiscal years. Programme costs are deducted rather than ignored. There is an explicit "Overlaps" line, running from £494m to £723m, netting off the double-counting between measures that interact. Devolved consequentials for Scotland and Northern Ireland are shown separately. This is more disciplined than most costings that reach us from any party.

So the headline reconciles. £52.6bn is the 2030-31 column, and £63.2bn is where the plan lands by 2033-34. The critics calling this fantasy economics have not engaged with the costing, because on its own terms it adds up.

Two things in it are worth pressing on.

The denominator runs out before the savings do. Forecast welfare spending stops at 2030-31, because that is where the OBR forecast stops, and Reform says so. The savings do not stop. They keep climbing to £63.2bn by 2033-34, in years where the "Savings as %" row reads only "See note". The last three columns of the plan are therefore a number without a denominator. That is not a trick, but it does mean nobody, Reform included, can tell you what share of the welfare bill the mature policy takes.

The second-largest measure in 2033-34 is not what anyone thinks it is. Watch the indexation row across the table.

That is the compounding, in Reform's own figures. The measure contributes nothing in the first year and nearly £4.8bn by the eighth, a nine-fold increase, and the table gives no reason to think it stops there. The IFS reached £4.8bn for 2033-34 independently, which is a good sign for both sets of arithmetic.

It is the only line in the table that behaves this way, and it is the one resting entirely on an index that has not been designed.

The assumptions carrying the weight

Reform assumes 241,000 more people in work by 2029/30, and states that "the fiscal benefits of this are accounted for in the costing above". So the headline saving is not a spending cut alone; it includes tax receipts from jobs that have been assumed into existence.

Of those 241,000, some 198,000 come from the disability reform. That is 82% of the total employment effect, arising from people losing disability benefits and being assumed to move into work. No labour-demand constraint is modelled. It says so outright: "We assume that these are net new jobs in the economy."

The IFS is pointed about the precedent here, noting there is "relatively little detail on what the new assessment would actually entail or on how they would avoid the experience of previous attempts to toughen up the system, which saw savings falling significantly short of government's expectations". Britain has run this experiment. The Work Capability Assessment and the PIP replacement of DLA were both sold on tougher gateways producing large savings, and both under-delivered.

On the employer insurance scheme, two things stand out. The body text leads with the Dutch result, a 40% fall in disability applications. The model then uses 2.7%, "scaled down for the much lower liability imposed under our system". The flagship evidence is discounted by roughly 93% in the actual costing, which is defensible modelling and indefensible presentation. And the scheme nets £200m in 2030-31, rising to £232m by 2033-34: roughly 0.4% of the headline, for something presented as one of the plan's four pillars.

One line we would still like explained. Employers are compensated for the new liability by a 0.2 percentage-point cut in employer National Insurance, worth about £1.6bn a year. Reform is right that the £1.6bn liability is not an Exchequer saving. But the NICs cut that offsets it is foregone revenue, and the consolidated table carries no line for it: the Return to Work Cover row shows only the net £200m saving to the Crown. Either the cost is netted inside that figure, in which case the arithmetic is doing something the table does not show, or it is missing. Reform should say which.

Damian raised a separate concern about this scheme that we share: insurance prices risk. Reform anticipates it, banning insurers from experience-rating on the individual health histories of a firm's workers. But it explicitly allows experience-rating at employer level, and concedes the asymmetry in its own words: "an employer receives the NI reduction regardless, but the firm or its insurer bears a direct cost if one of its workers leaves employment". Cost-neutral on average means a transfer from firms whose staff fall ill to firms whose staff do not. The hiring incentive that creates is not addressed.

What actually survives

Quite a lot, and this is where we part company with Labour and Mel Stride.

The disability reform is the strongest part of the plan. It is detailed, it protects severe and lifelong conditions explicitly, and the shift from cash to verified additional costs is a coherent answer to a real problem. Whether it delivers is an open question, given the history. That it is a serious proposal is not.

And Damian was right about the convergence, which we can now confirm against the table. He put his own sickness and disability figure at £22bn. Reform's disability line comes in at £22.116bn in 2030-31. Two people working separately, by different methods, from different starting assumptions, landed within £116m of each other on the largest single number in the welfare debate. That deserved more attention than it got.

The community work requirement is sensible and modest in fiscal terms. Reform scores the whole Welfare to Work programme at "more than £1 billion a year" once matured, and its own appendix nets each placement down to £87 after the value of the work done. Almost all of the saving comes from the assumed employment effect rather than the placements. The fraud enforcement is reasonable where the return exceeds the cost, and the paper's own steady-state figure of £925m is the one to use.

The citizenship restriction has a real principle underneath it, executed with the wrong instrument. And 58% of the affected caseload depends on successfully renegotiating the EU Withdrawal Agreement's equal-treatment provision. Reform says it "would renegotiate the treaty and disapply this provision". It provides no fallback if the EU declines. A £500m contingency is set aside for British citizens returning from the EU, but nothing at all for the possibility that the central legal manoeuvre fails.

The point

Reform has done something the other parties have not, which is to publish a costed plan with its assumptions visible. The diagnosis is right, the direction is defensible, and the willingness to show working deserves acknowledgement rather than the reflexive dismissal it received.

But two things in this paper are not what they are presented as.

The indexation change is presented as a technical correction and is in fact a permanent, compounding reduction in working-age benefit levels against the baseline the government currently uses, applied selectively where it saves money, justified partly on the grounds that unemployed people have time to cook.

The citizenship restriction is presented as restoring the contributory principle and in fact completes its destruction, withdrawing contribution-based entitlements from people who did pay in, while a citizen who never contributed keeps everything.

Labour called this fantasy economics. Mel Stride called it implausible. Neither appears to have opened the costing table, which reconciles. The real objections are elsewhere: an index that has not been designed, 198,000 jobs assumed into existence with no labour-demand constraint, £20bn riding on a treaty renegotiation with no stated fallback, three years of savings with no published denominator, and a rhetoric of contribution delivered through nationality. Those are arguments about assumptions. The spreadsheet is not the weak point, and saying it is has let Reform off lightly.

Reform has the scale of the problem right. On this version of the cure, we are not sold. Not yet.