Foreword

The Office for Budget Responsibility (OBR) was created in 2010 to provide independent and authoritative analysis of the UK’s public finances. In December 2013, the Government asked the OBR to take on additional responsibilities in relation to its newly announced cap on a subset of welfare spending. This request was in two parts: to assess performance against a cap on a subset of welfare spending, and to “prepare and publish information on the trends in and drivers of welfare spending within the cap”, so as to facilitate open and constructive debate. Parliament formally included these requirements in the 2015 edition of the Charter for Budget Responsibility, which since its 2022 update has required us to publish a Welfare trends report (WTR) once every two years.

We have explored several issues in our successive WTRs, ranging from a broad historical sweep of trends in UK welfare spending and international comparisons of welfare spending in our first two reports; to topics including universal credit, disability benefits, the Summer Budget 2015 welfare cuts, changes in welfare spending during and after recessions, and trends in working-age incapacity benefits spending across our subsequent seven reports. With rates of fraud and error in welfare having fluctuated substantially over recent years, this 10th WTR focuses on the drivers of these recent trends and what they imply for our approach to forecasting fraud and error in the welfare system.

The analysis in this report represents the collective view of the OBR’s Budget Responsibility Committee. We take full responsibility for the judgements that underpin it and for the conclusions we have reached. We have, of course, been supported in this by the full-time staff of the OBR, to whom we are enormously grateful, as we are to officials in the Department for Work and Pensions and HM Revenue and Customs who provided their help and expertise. We are also grateful to external stakeholders who gave their time and insights, including staff at the National Audit Office, Institute for Fiscal Studies, Institute for Government, Resolution Foundation, and National Institute of Economic and Social Research. In line with our Memorandum of Understanding, we provided a final copy of this report to HM Treasury two working days in advance of publication.

As with all our reports, the WTR remains a work-in-progress. We would be pleased to receive feedback on any aspect of the content or presentation of our analysis. This can be sent to [email protected].

The Budget Responsibility Committee
Professor David Miles CBE and Tom Josephs

Chapter 1: Executive summary

1.1 Fraud and error in welfare refers to overpayments where a claimant receives more than they are entitled to. The overall rate of welfare fraud and error rose sharply from 3.1 per cent of spending in 2019-20 to 4.3 per cent of spending during the Covid years of 2020-21 and 2021-22, an increase of around two-fifths (Chart 1.1). Over the subsequent four years the rate has fallen back to 3.2 per cent, close to its pre-pandemic level.

Chart 1.1: Fraud and error rate across welfare by benefit group

Note: Total welfare refers to all DWP welfare spending and HMRC spending on child and working tax credits. UC and legacy includes universal credit, working-age housing benefit, income-based employment and support allowance and jobseeker’s allowance, child and working tax credits, incapacity benefit, and income support. Pensioner benefits include state pension, pension credit, and pensioner-age housing benefit. Disability benefits include disability living allowance, personal independence payment, and attendance allowance. Tax credit fraud and error includes in-year overpayments, as outlined in Box 2.1.

Source: DWP, HMRC, OBR

1.2 This Welfare trends report investigates the drivers of these recent trends, with a particular focus on the pandemic and the move to universal credit (UC), and considers how this analysis should be reflected in our fraud and error forecasts. Data on fraud and error comes from sampling exercises and there is limited evidence on the underlying drivers of trends. This means there is inevitably some uncertainty around the conclusions reached in this report.

1.3 We focus our analysis on fraud and error in universal credit. The 2020-21 rise and subsequent fall in welfare fraud and error was entirely concentrated within UC. The UC fraud and error rate rose from 9.4 per cent of UC spending in 2019-20 to 14.7 per cent in 2021-22, before then falling to 8.5 per cent in 2025-26. Fraud and error rates across other large benefits – including for legacy benefits that were being progressively phased out in favour of UC – were largely stable over this period.

1.4 The initial rise in UC fraud and error in 2020-21 and the subsequent decline were primarily concentrated among the cohort of claimants joining UC in the first year of the pandemic (Chart 1.2). By contrast, the fraud and error rate among pre-existing cohorts was relatively stable. The fraud and error rate among the ‘Covid cohort’ was 25.8 per cent of spending in 2020-21. This compares to the overall 2019-20 rate of 9.4 per cent, and the rate among pre-existing cases in 2020-21 of 9.8 per cent. Likewise, the rate of fraud and error among the Covid cohort fell by two-thirds between 2020-21 and 2025-26, while the rate among all other UC cases fell by less than a sixth over the same period.

Chart 1.2: Universal credit fraud and error rate in the Covid and non-Covid cohorts

Note: The Covid cohort refers to new claims to UC made between 14 March 2020 and 27 February 2021. The non-Covid cohort refers to all other cases on UC in each year.

Source: DWP, OBR

1.5 The initially high fraud and error rate in the Covid cohort is likely to be primarily explained by operational easements introduced by the Department for Work and Pensions (DWP) at the start of the pandemic, which aimed to reduce restrictions and speed processes at a time of acute need. This included relaxing departmental checks normally in place to process a benefit claim, and the suspension of the minimum income floor and gainfully self-employed checks for self-employed claimants. The nature of the Covid shock to the economy also led to a significant rise in self-employed cases on UC, which have substantially higher rates of fraud and error than the rest of the caseload.

1.6 The subsequent fall in fraud and error is very likely to be largely driven by the removal of these measures in the following years along with additional departmental activity to combat fraud. In response to the Covid spike, DWP increased staff addressing fraud and error from 4,700 in 2020-21 to 15,000 in 2025-26, with plans for further increases in the coming years which we expect to lead to further falls in the medium term.

1.7 It is possible that there was also a greater willingness to commit fraud among potential claimants during the Covid period, though the extent of this is uncertain. Prior to the pandemic, there is some evidence that societal views towards fraud were softening, and that fraud and error across society was on the rise. The onset of the pandemic may have accelerated these trends. First, due to the greater incentive for fraud among those facing economic hardship. And second, due to the greater opportunity for fraud from the operational easements mentioned above alongside wider Covid-related changes, such as the suspension of face-to-face appointments to monitor compliance with conditionality requirements, and more publicity around available benefits and the digital claims system.

1.8 The pandemic hit while the working-age welfare system was undergoing significant structural change, in particular from the transition to UC. While UC had been fully rolled out for new claims in 2019-20, it made up just 31 per cent of combined UC and legacy spending in that year. By 2025-26, this share had risen to 91 per cent. Significant structural differences between UC and the legacy system are also likely to have contributed to changing patterns of fraud and error in this period, although our analysis suggests these changes may have been largely offsetting overall. The differences include:

• Growth in capital- and housing-related fraud and error in UC relative to the legacy system as a result of tighter restrictions on savings levels, and centrally administered housing support in UC. Overpayments related to these factors were small in the legacy system – averaging around 0.4 per cent of UC and legacy spending between 2011‑12 and 2018‑19 – but have become more significant since 2019-20, averaging 2.1 per cent.
• Evidence of higher fraud and error among those with self-employment income in UC, as monthly assessment and other changes have increased reporting requirements for self-employed claimants. In addition, stronger incentives to work at low hours and earnings levels in UC have increased the prevalence of low levels of self-employed income among those who would have been on out-of-work legacy benefits.
• Pushing in the other direction is lower in-work fraud and error for employees in UC than the legacy system. In large part this is because UC’s monthly assessment has removed the possibility of in-year overpayments – reconciled at year-end – that were a feature of the annually assessed tax credits system. These in-year overpayments averaged 1.7 per cent of spending between 2011-12 and 2018-19. The use of real-time earnings information to calculate UC entitlement has also reduced the scope for income-related overpayments among those earning via PAYE.

1.9 Based on the findings in this report, we plan to change our forecast methodology for fraud and error at the next forecast. Previously we assumed that the underlying propensity for fraud and error in UC would rise in each year of the forecast. This was based on increased fraud and error rates and some evidence that fraud has risen more widely across society in recent years. However, as the analysis in this report suggests that the rise was concentrated in the Covid cohort, and that fraud and error rates have now returned close to pre-pandemic levels, we will now remove this rising underlying propensity assumption.

1.10 There is uncertainty around this judgement and the wider outlook for fraud and error. Even before making this change the forecast assumed that the recent decline in fraud and error rates would continue for several years, largely based on a judgement that that the ongoing significant increase in DWP resource to combat fraud and error would be effective. We will monitor this and wider developments in fraud and error as new data becomes available.

Chapter 2: Recent trends in fraud and error

Introduction

2.1 Our biennial Welfare trends reports (WTRs) examine trends in social security benefits and tax credits. This year’s report considers fraud and error in the welfare system, with a particular focus on the changing trends in fraud and error in universal credit (UC) following both its full rollout and the pandemic. Other public bodies, such as the National Audit Office (NAO), also examine fraud and error in the welfare system, often with a focus on assessing the value for money and effectiveness of Department for Work and Pensions (DWP) operations. This is not our focus – the role of the WTR is to provide insights that improve our welfare forecasts and our costings of welfare policy – however, we draw on these findings in this report where relevant to our analysis.^[1]

2.2 Fraud and error is an inherently more uncertain topic than those covered in previous WTRs. While benefit expenditure and caseloads are recorded in DWP administrative data, fraud and error is unobserved and therefore analysis depends on sampling exercise data with small samples and complex methodologies. In addition, there is relatively limited evidence on the underlying drivers of trends in fraud and error. For example, there is some evidence that fraud across society has been on the rise in recent years, and we have used this to support our past forecast judgements (see Chapter 3). However, this evidence is very far from conclusive. These features of measuring and understanding fraud and error mean there is inevitable uncertainty around all the conclusions we reach in this report. Box 2.1 explains how we define fraud and error in this report and how we measure recent trends using DWP’s annual sampling exercise data.

2.3 In this chapter, we consider recent trends in fraud and error in the welfare system, with a particular focus on the volatile trends within UC and their key underlying drivers. This analysis then informs the next chapter, where we present our forecast for fraud and error in welfare and describe how the review of recent trends in this report has led to improvements in our forecast methodology.

Fraud and error rates across welfare

2.4 Fraud and error rates have differed significantly across different benefits. As shown in Chart 2.1, the overall level and trend in fraud and error across the welfare system has been driven predominantly by UC and the legacy benefits it replaced.^[2] UC and legacy benefits have a much higher rate of fraud and error than other large benefits, accounting for over 70 per cent of total fraud and error expenditure in 2025-26 despite only representing around 30 per cent of total welfare expenditure.

Chart 2.1: Fraud and error rates across welfare by benefit group

Note: Total welfare comprises all DWP welfare expenditure and HMRC expenditure on child and working tax credits. UC and legacy comprises UC, income-related ESA and JSA, IB, IS, working-age HB, and child and working tax credits. Pensioner comprises state pension, pension credit, and pensioner HB. Disability comprises disability living allowance, personal independence payment, and attendance allowance.

Source: DWP, HMRC, OBR

2.5 Chart 2.1 shows that the overall rate of fraud and error within the welfare system fell between 2003-04 to 2019-20. There were two particular periods of decline in the mid-2000s and early 2010s, which were both concentrated in tax credits. The former was driven by a definitional change, the latter by a policy focus on fraud and error:

• In the early 2000s, HM Revenue and Customs (HMRC) reformed its ‘income disregards’ process, which significantly reduced measured overpayments. Under the tax credits process, claimants would receive a monthly payment based on their income in the previous year or estimated income for the current year, but would then confirm their final annual income after the financial year was complete to calculate their final tax credits entitlement for that year. There was an ‘income rise disregard’, initially set at £2,500, which created a buffer zone within which a claimant’s income could increase over the year without affecting their entitlement. This process resulted in very large corrections to entitlements and accordingly high volumes of overpayments, because in a large number of cases income changed by more than the threshold. In April 2006, HMRC significantly raised the income rise disregard threshold, allowing claimants to revise up their annual income by up to £25,000 without affecting their final entitlement.^[3] This helped to reduce the tax credits overpayment rate from 23 per cent in 2003-04 to 14 per cent by 2007-08.^[4] This decrease therefore reflected a narrower definition of overpayments rather than a change in claimant behaviour.
• In the early 2010s, HMRC was able to drive reductions in the tax credit overpayment rate through a policy focus on its detect and correct activities, which contributed towards a reduction in overpayments from over 13 per cent in 2010-11 to 10 per cent in 2014-15.^[5]

2.6 Turning to the more recent period, fraud and error rates spiked sharply in 2020-21, remained high in 2021-22, and since then have been falling. Almost all of this trend has been concentrated in UC and legacy benefits. The timing of these changes suggests they were likely driven by one or a combination of two major developments: the transition from the legacy system to UC, and the onset of the Covid pandemic. In the remainder of this chapter we explore these factors in more detail and consider their relative importance in explaining recent trends:

• The pandemic: The onset of the Covid pandemic was followed by a 1.2 percentage point (39 per cent) increase in the welfare fraud and error rate in 2020-21, the largest year-on-year change on record. We consider how the pandemic altered claimant behaviour and departmental activity in the short term, to what extent these pandemic effects unwound over the subsequent five years, and how policy announcements following the pandemic have affected fraud and error.
• The move to universal credit: 2019-20 was the first financial year in which universal credit was fully rolled out for new claims across the UK. UC’s share of welfare spending has rapidly increased since then, rising from less than 10 per cent of total spending in 2019-20 to over 25 per cent in 2025-26, and from around 30 per cent of UC and legacy spending in 2019-20 to over 90 per cent in 2025-26. We consider whether features inherent to the UC system may have contributed to higher rates of fraud and error.

The pandemic

2.7 This section explores the short- and medium-term impacts of the pandemic on UC fraud and error, which saw the overpayment rate increase by over 50 per cent in a single year (from 9.4 per cent of UC expenditure in 2019-20 to 14.5 per cent in 2020-21) before then declining, and eventually falling to below its pre-pandemic level to 8.5 per cent by 2025-26. We examine how the pandemic affected new and existing claims differently, how operational easements affected fraud and error, and how additional policy measures following the pandemic have begun to reduce overpayments.

The Covid cohort impact

2.8 Both the sharp rise in UC fraud and error in 2020-21 and the subsequent decline over the following five years was driven predominantly by claimants that joined during the first year of the pandemic. As shown in Chart 2.2, in 2020-21, overpayments accounted for over a quarter of all UC spending on the ‘Covid cohort’,^[6] a rate more than two-and-a-half times higher than the overpayment rate for pre-existing claims, which only increased slightly between 2019-20 and 2020-21, from 9.4 to 9.8 per cent. Over subsequent years, the Covid cohort’s overpayment rate fell by nearly two-thirds (from 25.8 per cent of spending in 2020-21 to 8.9 per cent in 2025-26), whereas the rate for all other UC cases was relatively unchanged, rising slightly from 9.8 per cent of spending in 2020-21 to 11.7 per cent in 2022-23 before falling to 8.4 per cent in 2025-26.

Chart 2.2: Universal credit fraud and error rates in the Covid and non-Covid cohorts

Note: The Covid cohort refers to cases that joined UC between 14 March 2020 and 27 February 2021. The non-Covid cohort refers to all other cases on the UC caseload that joined in the years before and after.

Source: DWP, OBR

2.9 It is typical for claimant cohorts to begin with higher rates of fraud and error which then fall as the cohort matures (Chart 2.3). This is explained by the changing composition of cohorts over time, which typically begin with a large proportion of unemployed and in-work cases (which tend to have substantially higher overpayment rates) before shifting towards a greater proportion of inactive cases as unemployed cases either leave UC or fall into ill health and move into inactivity. While the 2020 cohort follows a similar trend to the other cohorts shown in Chart 2.3, the scale of fraud and error within it far exceeds that seen in any other cohort in the first three years of the claim, underlining the likelihood of a pandemic-specific effect driving higher fraud and error rates in this group.

Chart 2.3: Universal credit fraud and error rates by cohort and duration

Note: Cohorts are split by calendar year of initial claim, and therefore the 2020 cohort is not entirely equivalent to the Covid cohort in Chart 2.2, which spans mainly the 2020 and the early part of the 2021 calendar years.

Source: DWP, OBR

The short-term impact of the pandemic on fraud and error

2.10 The vast majority of the rise in fraud and error in 2020-21 was driven by higher overpayment rates across all conditionality groups.^[7]The fraud and error rates in the Covid cohort for unemployed, in-work and inactive cases were all more than double their non-Covid counterparts (left panel of Chart 2.4). This indicates that the pandemic’s impact on fraud and error was primarily due to factors that affected the entirety of the Covid cohort, rather than being due to differences in the composition of the Covid cohort compared to previous cohorts. These higher rates within conditionality groups explain over four-fifths of the total rise in the UC overpayment rate in 2020-21, with Covid cohort composition, and higher non-Covid cohort fraud and error rates, each explaining roughly a tenth of the rise (right panel of Chart 2.4).

Chart 2.4: Decomposition of change in universal credit fraud and error rate by cohort and conditionality groups, 2019-20–2020-21

Note: The Covid cohort refers to cases that joined UC between 14 March 2020 and 27 February 2021. The non-Covid cohort refers to all other cases on the UC caseload that joined in the years before and after.

Source: DWP, OBR

2.11 It is likely that the high fraud and error rates across all conditionality groups within the Covid cohort are primarily explained by operational measures taken by DWP around the onset of the pandemic. Assessments of the impact of these measures, including by DWP,^[8] suggest the following factors, which particularly affected new claims, contributed to this:

• The relaxation of departmental checks normally in place to process a benefit claim. During the pandemic, verification checks related to identity, household circumstances, and tenancy agreements were eased in order to meet the high demand of new claims and comply with wider public health restrictions. These easements created greater scope for people to make fraudulent UC claims as information pertaining to eligibility was self-reported without verification by the department.
• The suspension of the minimum income floor (MIF) and gainfully self-employed (GSE) checks for self-employed claimants, providing the opportunity for claimants to under-report earnings from their businesses to earn higher UC awards.
• Smaller operational adjustments may also have played some role, including the suspension of jobcentre appointments, for example to monitor compliance with conditionality requirements, and the rapid expansion of staff capacity to process high volumes of new claims and payments.

2.12 In addition to operational measures, the nature of the Covid shock on the economy also contributed to the rise in fraud and error in 2020-21. With many businesses forced to close in response to the pandemic, there were large onflows onto UC of self-employed claimants who otherwise would not typically engage with system, with the self-employed share of the UC caseload rising from 5 per cent in 2019-20 to 13 per cent in 2020-21. Given that fraud and error rates among self-employed claimants (32.5 per cent in 2020-21) are much higher than the rest of the caseload, this shift in composition contributed to a higher overall fraud and error rate within UC.

2.13 Combined with these factors, the rise in UC fraud and error rates in 2020-21 may have reflected an increased willingness to commit fraud in wider society, though this is uncertain. Starting before the pandemic, some survey evidence indicated softening societal views towards fraud,^[9] and there was some evidence of increasing fraud in other parts of society.^[10] The onset of the pandemic may have accelerated these trends, with greater incentives and opportunity to commit fraud due to the economic need and instability the pandemic caused, alongside enhanced publicity around the availability of UC and its digital claims system.

2.14 While new cohorts typically have a higher share of overpayments attributed to income-related fraud (given new cohorts have high volumes of unemployed and in-work cases), the scale of income-related fraud in the Covid cohort was much higher than for other cohorts. Nearly half of all fraud and error in the Covid cohort in 2020-21 was related to income (Chart 2.5), compared to much smaller shares in the non-Covid cohort in 2020-21 (a quarter) and across UC in 2019-20 (a third).

Chart 2.5: Composition of universal credit fraud and error by reason and cohort

Note: The Covid cohort refers to cases that joined UC between 14 March 2020 and 27 February 2021. The non-Covid cohort refers to all other cases on the UC caseload that joined in the periods before and after.

Source: DWP, OBR

2.15 The majority of the rise in income-related fraud and error relates to self-employment and was likely driven in part by specific easements made for self-employed claims. The nature of the Covid shock to the economy drove a lot of self-employed claimants into UC who would not normally engage with the system. This economy shock was paired with the suspension of gainfully self-employed checks to verify self-employed status and the removal of the minimum income floor in March 2020, which together contributed to both high inflows of self-employed claimants and higher fraud and error rates among the self-employed. With these easements, claimants not only no longer needed to demonstrate a credible business plan but also could under-report earnings from their businesses to earn greater UC awards. Even if circumstances improved, claimants had no incentive to report a change in their earnings and so remained on the caseload. DWP estimates that the suspension of the MIF increased UC fraud and error rates by between 0.7 per cent (£0.3 billion) and 1.5 per cent (£0.6 billion) of spending in 2020-21.^[11] Total UC overpayments related to self-employment rose from £0.3 billion in 2019-20 (17 per cent of all UC overpayments) to £1.9 billion in 2020-21 (34 per cent of UC overpayments).

The medium-term impact of the pandemic on fraud and error

2.16 The UC fraud and error rate fell from 14.5 per cent in 2020-21 to 8.5 per cent in 2025-26, a rate lower than the pre-pandemic level of 9.4 per cent. This gradual unwind of the pandemic impact is likely to be the result of a combination of new policy initiatives and changing Covid cohort trends. Chart 2.6 decomposes this fall first into the respective Covid and non-Covid cohort drivers, showing that:

• Roughly half of the fall (3.2 percentage points) was driven by lower fraud and error rates in the Covid cohort across all conditionality groups. The unemployed, in-work, and inactive overpayment rates in the Covid cohort all more than halved between 2020-21 and 2025-26, and are now very close to their non-Covid cohort counterparts (Chart 2.7). This reflects the removal of fraudulent cases in the cohort over time, primarily through claim closure or the removal of overpayments by the department after identification of fraud.
• A further tenth (0.5 percentage points) of the fall shown in Chart 2.6 is explained by a caseload composition shift within the Covid cohort towards inactive cases. The inactive share of spending in the Covid cohort rose from 20 per cent in 2020-21 to 72 per cent in 2025-26. These inactive cases have a significantly lower fraud and error rate (7.5 per cent of spending in 2025-26) than unemployed and in-work cases in the Covid cohort (11.3 and 12.8 per cent of spending respectively).
• Lower fraud and error within the non-Covid cohort explains a fifth (1.1 percentage points) of the fall.
• The remaining fifth (1.2 percentage points) of the fall is due to the Covid cohort becoming a smaller share of the UC caseload. As UC has expanded through both new claims and migrations from legacy benefits, the relative size of spending on the Covid cohort (and therefore its impact on the overall UC fraud and error rate) has diminished, falling from 29 per cent of UC spending in 2020-21 to 12 per cent in 2025-26.

Chart 2.6: Change in universal credit fraud and error rate, 2020-21–2025-26

Note: The Covid cohort refers to cases that joined UC between 14 March 2020 and 27 February 2021. The non-Covid cohort refers to all other cases on the UC caseload that joined in the years before and after.

Source: DWP, OBR

Chart 2.7: Universal credit fraud and error rates by cohort and conditionality group, 2020-21–2025-26

Note: The Covid cohort refers to cases that joined UC between 14 March 2020 and 27 February 2021. The non-Covid cohort refers to all other cases on the UC caseload that joined in the years before and after.

Source: DWP, OBR

2.17 As described above, DWP initiatives also likely contributed to the reduction in fraud and error after 2020-21. In part, this reflected the resumption of business-as-usual departmental activity as the pandemic eased:

• In January 2021, DWP introduced retrospective verification checks for claims that joined during the height of the pandemic. Between January and December 2021 the department reviewed around 600,000 cases within the Covid cohort, with around 8 per cent of reviews leading to claim closures and a further 4 per cent correcting overpayments on claims that remained open after review.
• The minimum income floor (MIF) and gainfully self-employed (GSE) checks were reinstated in August 2021. Cases that had come on during the pandemic without being subjected to a GSE check were invited to a check from October 2021 onwards. Many decided to close their claim rather than present a credible business plan to DWP, which reduced the self-employed caseload by around 11 per cent in 2021-22 (56,000 cases). DWP estimate the reinstatement of the GSE checks to have reduced the measured fraud and error rate by 0.9 per cent of UC spending that year.

2.18 Further, DWP announced new policy measures aimed at reducing fraud and error:

• Targeted case reviews (TCR). In 2022, DWP established the TCR programme to identify and amend incorrect UC claims. The programme was designed to scale to around 6,900 reviewing staff by March 2025, with a phased rollout from 2022 to full national delivery by 2025. In 2025-26, TCR was estimated to reduce UC fraud and error by £0.7 billion (0.9 per cent of UC spending), costing DWP £0.3 billion in additional resource.^[12] More detail on the additional savings we expect from TCR over the forecast (as well as from other additional policy measures) is available in Chapter 3.
• Additional core counter-fraud staff. In October 2021, DWP announced that it would temporarily hire an additional 2,500 full-time equivalent (FTE) staff until March 2025 to undertake more of its core counter-fraud activity, before expanding the policy further and making it permanent at subsequent fiscal events. Between 2020-21 and 2025-26, core FTE staff rose from 4,700 to 8,000. In 2025-26, it is estimated that core counter-fraud activity reduced UC fraud and error by £0.9 billion (1.1 per cent of UC spending), costing DWP £0.4 billion.^[13]

Conclusion

2.19 The pandemic led to the greatest increase in welfare fraud and error on record in 2020-21, but the resumption of business-as-usual departmental checks, the evolution of the Covid cohort over time towards lower-fraud cases, and the expansion of the non-Covid caseload have contributed to UC fraud and error returning to its pre-pandemic levels. Taken together with the impacts of significant additional policy to reduce fraud and error, the UC overpayment rate fell to below its pre-pandemic level for the first time in 2025-26. We consider the implications of these trends for our forecast for fraud and error in Chapter 3.

The move to universal credit

2.20 The introduction of UC represented a far-reaching reform to working-age welfare. UC has replaced six different means-tested working-age benefits: tax credits (child and working), non-contributory jobseeker’s allowance (JSA), non-contributory employment and support allowance (ESA), incapacity benefit (IB), housing benefit (HB), and income support (IS). While the initial legislation underpinning UC was passed in 2012, rolling out the new system has taken over a decade.^[14] The rollout of UC can be broadly grouped into three phases:

• Pilot stage (2013-2018): UC was initially rolled out via a system of pilots, beginning in the North West of England in 2013. By November 2018, the month before UC became available across the UK, the claimant caseload had reached 1.4 million people.
• Natural migration (2019-2022): From December 2018, legacy benefits closed to new claims and UC became available across the UK. This began a process of natural migration as not only all new claims but all legacy claimants who experienced a change in circumstance were directed to claim UC.
• Move to UC (2023-): Under this final phase of UC rollout, previously known as ‘managed migration’, DWP has actively worked to move claimants of legacy benefits onto UC. This began with tax credit claimants in April 2023, enabling the tax credit system to be closed in April 2025, and has been followed by efforts to transition remaining legacy claims (mainly ESA).

2.21 Ahead of the transition to UC, modelling suggested that its design would reduce fraud and error relative to the legacy system. These assumptions formed part of a broader costing and forecasting framework used to estimate the fiscal impact of UC and were explored in some detail in the OBR’s 2018 Welfare trends report. Taken together, these assumptions implied that, following its full implementation, UC would reduce measured fraud and error by approximately £1.6 billion a year.^[15] This reflected:

• £2.0 billion of estimated gross savings from fraud and error. Most estimated savings were expected to reflect the removal or simplification of features of the legacy system that were associated with higher fraud and error rates. The largest estimated saving was expected to come from the use of real-time earnings information (RTI), which replaced greater reliance on annual reporting and reconciliation processes, and was expected to reduce fraud and error by around £0.8 billion a year. The removal of hours rules was estimated to reduce fraud and error by a further £0.5 billion, with the rest explained by several smaller savings.
• These savings were expected to be partly offset by £0.4 billion of additional fraud and error. This was largely expected to reflect the policy decision to restrict eligibility and entitlement within UC through the application of capital (savings) rules to all claimants. This affected previous tax credit claimants who were not subject to capital rules under the legacy system, and thus created a new channel through which overpayments could arise. Estimates suggested that these rules would increase fraud and error costs by £0.4 billion relative to the legacy system in 2022-23. Other changes, including the move to monthly assessment and late RTI reporting, were expected to only have a very small impact.

Table 2.1: 2018 estimated impact of transition to universal credit on fraud and error

Differences in fraud and error rates between UC and legacy benefits

2.22 Across the set of legacy benefits that UC replaced, there was a wide variation in fraud and error rates (Chart 2.8). These differences largely reflect variation in claimant circumstances and benefit design, particularly the extent to which awards depended on volatile in-work earnings or required retrospective reconciliation of income:

• Tax credits: tax credits (working and child) consistently exhibited the highest rate of overpayments of any of the legacy benefits, with an initial peak of over 20 per cent of tax credit spending shortly after their introduction in April 2003 followed by a rate that remained consistently over 10 per cent throughout the 2010s.^[16] This decline during the 2000s is explained in paragraph 2.5. Ongoing high fraud and error rates in tax credits reflected both a largely in-work caseload with fluctuating earnings and household circumstances, and a design based on estimated annual income with end-year reconciliation, which created substantial scope for temporary, in-year overpayments.^[17]
• Working-age housing benefit: housing benefit tended to show the second-highest rate of fraud and error, averaging over 6 per cent across the 2010s. This was driven by the complexity of the means-test calculation, including detailed rules on earnings, capital and household composition, alongside frequent changes in rent and living arrangements.
• Income support, jobseeker’s allowance, incapacity benefit and employment and support allowance: These benefits generally exhibited lower rates of fraud and error. This largely reflected a more stable claimant base with limited or no labour market attachment, reducing exposure to fluctuating in-work earnings.^[18] These benefits were awarded fortnightly or monthly based on current circumstances rather than annually reconciled, and awards tended to be based on relatively static household situations. In addition, ongoing conditionality and regular claimant contact (particularly in JSA) supported more timely reporting of changes in circumstances, limiting the build-up of undetected overpayments.

Chart 2.8: Legacy benefit fraud and error rates

Note: The fraud and error rate for tax credits includes in-year overpayments, as explained in Box 2.1.

Source: DWP, HMRC, OBR

2.23 Chart 2.9 shows that by 2019-20, the year before the pandemic, and when UC had been fully rolled out for new claims across the UK, its overpayment rate had reached 9.4 per cent of spending. This compares with the legacy fraud and error rate of 8.6 per cent in 2017-18, which was the final year before UC rollout began to significantly impact the legacy caseload. Fraud and error rates under UC then increased sharply during the pandemic period before subsequently falling back to 8.5 per cent by 2025-26, due to the pandemic-related trends discussed in the previous section. This makes it difficult to disentangle the effects of the pandemic from the effects of the new system itself. However, both the initial and current UC overpayment rates remain towards the upper end of the range observed under the legacy system, which had not exceeded 9 per cent since the earliest years of tax credits in 2005-06. Some allowance should be made for teething issues in the early stages of UC rollout, the lingering effects of the pandemic that were discussed in detail in the previous section of this chapter, and also the possibility that there was an upward trend in the prevalence of fraud across society in this period. But, nevertheless, this could suggest that some features of UC are having an upward impact on fraud and error relative to the legacy system, rather than delivering the reductions that were originally expected. This is explored further in the next sections.

Chart 2.9: Legacy benefit and universal credit fraud and error rates

Note: Full rollout refers to the stage from 2019-20 onwards when UC had replaced legacy benefits for all new claimants, but some existing claimants continued to claim legacy benefits.

Source: DWP, HMRC, OBR

Universal credit fraud and error rates by conditionality group

2.24 The impact that the move to UC has had on fraud and error varies by the labour market conditionality groups within the UC and legacy system. UC conditionality (or, more properly, labour market regime) groups, which broadly reflect a claimant’s labour market status and expected level of work-related activity, can be broadly compared with legacy benefits that served claimants in similar circumstances, although the mapping is not perfect and so conclusions on its basis should be treated with a degree of uncertainty.^[19] Table 2.2 provides more detail on UC groups and sets out the broad correspondence between three groups that can be identified in both UC and the legacy system, which we describe as ‘unemployed’, ‘inactive’ and ‘in work’.

Table 2.2: Conditionality groups in universal credit and legacy benefits

2.25 Chart 2.10 shows that the impact of UC on fraud and error rates has varied across these three claimant groups. Among in-work claimants, the introduction of UC has been associated with a reduction in fraud and error (excluding the pandemic period). By contrast, rates have risen significantly for both inactive and unemployed claimants, with the increase for the latter particularly marked, lifting it well above the inactive rate with which it had previously moved closely, and above that for in-work claimants. This suggests that, while system-wide factors have contributed to higher fraud and error under UC relative to legacy benefits, these effects have been concentrated among unemployed and inactive claimants, whereas the transition from tax credits to UC appears to have reduced fraud and error among in-work claimants. We explore the potential reasons for these differences below.

Chart 2.10: Legacy benefit and universal credit fraud and error rates by conditionality group

Note: Legacy rates include both group-specific benefits for each group (e.g. JSA for the unemployed) and each group’s relevant share of both overpayments and expenditure from benefits that covered multiple groups, such as tax credits and housing benefit.

Source: DWP, HMRC, OBR

The role of the design of universal credit

2.26 Alongside the overall rates of fraud and error by benefit, sampling exercise data also shows overpayments decomposed by their reason (Chart 2.11). This exercise suggests that the largest overpayment reasons under the legacy system – household composition and income – have respectively remained broadly stable and modestly fallen with the move to UC. The fall in income-related overpayments corresponds with lower fraud and error rates in the in-work group in UC than legacy benefits, discussed above. By contrast, increases in overpayments have been concentrated in reasons that were relatively minor under the legacy system, namely capital, housing, and unknown reasons. We explore these in more detail below.

Chart 2.11: Legacy benefit and universal credit fraud and error rates by reason

Note: ‘Household composition’ comprises an undeclared partner and other household composition reasons for fraud and error. ‘Other’ includes disability, living abroad and childcare costs, as well as all other smaller reasons for fraud and error.

Source: DWP, HMRC, OBR

2.27 The growth in capital, housing, and unknown/uncategorisable reasons for fraud and error in UC relative to the legacy system can be explained by specific design elements of UC:

• Capital: Of these reasons, capital has experienced the largest growth, with overpayment rates rising from an average of 0.2 per cent of UC and legacy spending between 2011-12 and 2018-19 to an average of 1.4 per cent between 2019-20 and 2025-26. Under UC, all claimants are subject to capital requirements reflecting their cash savings and investments. UC entitlement is reduced for those with between £6,000 and £16,000 in capital, while those with over £16,000 are not eligible. While DWP-administered benefits in the legacy system were also subject to capital limits, HMRC-administered tax credits did not include the same restrictions.
• Housing: Overpayments relating to housing reasons, reflecting incorrect information around housing costs or residency, grew from representing 0.2 per cent of UC and legacy spending in the years between 2011-12 and 2018-19 to an average of 0.7 per cent since 2019-20. Under UC, most claimants receive the housing element as part of a single monthly payment, based on information provided to DWP to verify residency and rent liability. This contrasts with the legacy system, where housing benefit was administered by local authorities rather than DWP, and payments were more commonly made directly to landlords. These differences may have contributed to higher housing-related fraud and error in UC through two channels. First, direct payment to claimants may increase the risk that changes in housing circumstances are not as accurately or promptly reported. Second, the shift from local to central administration may have reduced the role of local knowledge in identifying implausible claims at the point of award.
• Unknown: Overpayments for which DWP was unable to attribute an error reason likewise rose significantly in the transition to UC, from 0.2 per cent of UC and legacy spending between 2011-12 and 2018-19 to 1.0 per cent of spending since 2019-20. The increased prevalence of ‘unknown’ overpayments partly reflects methodological changes in how DWP classifies fraud and error made since UC was fully rolled out, rather than reflecting a ‘genuine’ underlying increase in fraud or error activity.^[20]

2.28 The reduction in overpayments related to income in the transition from the legacy system to UC has been modest, but the move to UC has had a clear impact in changing their composition. This has been driven by the following factors (Chart 2.12):

• Removal of in-year overpayments.^[21] Compared in particular to tax credits, UC was designed to create a more responsive system that reflects changes in circumstances in real time. The shift to monthly assessment replaced the annualised structure of tax credits, meaning entitlements are decided at the point of payment, updated in line with current earnings and household circumstances, rather than being reconciled retrospectively as in the tax credits system. This effectively removed the possibility for ‘in-year’ overpayments (blue bars in Chart 2.12), which had averaged 1.7 per cent of UC and legacy spending in the 2010s (equivalent to an average of over 20 per cent of all UC and legacy fraud and error).
• Rise in self-employed income-related fraud and error. Partly offsetting this is the rise of self-employed and self-reported income-related fraud and error in UC (yellow bars in Chart 2.12) – although it is not possible to separately identify the equivalent group in the legacy system in Chart 2.12, so these conclusions are somewhat tentative. Since full rollout in 2019-20, overpayments related to self-employed income in the UC system have averaged 1.9 per cent of UC and legacy spending, roughly in line with the amounts previously lost to in-year overpayments. While a large part of this increase was driven at least initially by the pandemic (see paragraph 2.15), the persistence of high self-employed overpayments as overall UC fraud and error rates have declined suggests the increase may also be linked to the design of UC. This is likely to reflect the shift to monthly income assessment under UC, which, alongside the MIF, places a higher reporting requirement on self-employed claimants than the legacy tax credits system. Additionally, the removal of hours rules, income-related cliff edges and pound-for-pound withdrawal of benefits at low levels of earnings that existed for ‘unemployed’ and ‘inactive’ benefits in the legacy system (JSA, ESA and IS) has increased the prevalence of work and income among claimants in these conditionality groups in UC.^[22] DWP evidence suggests this is very largely driven by self-employed income. Income-related overpayments in these conditionality groups have risen from around 1 per cent of spending in 2017-18 to nearly 3 per cent in 2025-26.
• The fall in other income-related fraud and error, shown in the purple bars in Chart 2.12, is mostly driven by the use of real-time information on employee earnings in UC, reducing reporting requirements and therefore the scope for overpayments. In addition, the decline will be partly driven by the reduction in legacy self-employed fraud and error as tax credits have been progressively replaced by UC (as mentioned above, it is not possible to separate out self-employed income-related overpayments for the legacy system in Chart 2.12).

Chart 2.12: Income-related fraud and error rates in legacy benefits and universal credit

Note: Income (other) includes all sources of income (including both earnings and employment) related fraud and error overpayments other than in-year overpayments, and self-employed and self-reported income within UC. We are not able to separate out legacy self-employed fraud and error from other income-related fraud and error, meaning there is not a consistent time series for self-employed fraud and error between UC and legacy benefits.

Source: DWP, HMRC, OBR

Conclusion: the effect of the move to universal credit on fraud and error

2.29 Quantifying the net impact of the move to UC on fraud and error is complicated by the concurrent impact of the pandemic, as well as smaller factors such as sampling exercise methodology changes, the gradual process of migration from legacy benefits, and the impact of new fraud and error policy initiatives. This all limits our ability to precisely isolate the impact of the move from the legacy system to UC on rates of fraud and error.

2.30 But while very uncertain, it appears the various impacts of the transition from legacy benefits to universal credit have been broadly offsetting for fraud and error, in contrast to the modest reduction in fraud and error originally anticipated as a result of the move to UC:

• UC design features appear to have increased the prevalence of capital, housing, and self-employed income-related fraud and error. The extension of capital rules and changes to the administration of the housing element have increased opportunities for fraud and error. Self-employed fraud and error is likely to be higher in UC than in the legacy system in part because of the increased prevalence of self-employment income in the caseload due to UC encouraging work at low hours and earnings much more than the legacy system did, and in part because monthly assessment and other changes have increased reporting requirements for self-employed claimants.
• Offsetting these features, UC has reduced other income-related fraud and error, particularly due to the shift from annual to monthly assessment for in-work claimants, and the use of real-time earnings information for those earning via PAYE.

2.31 Alongside the evidence on the impact of the pandemic on fraud and error in the preceding section of this chapter, these findings allow us to reflect on our approach to forecasting the evolution of fraud and error rates in the years ahead. This assessment is provided in the following chapter.

Chapter 3: Fraud and error in the forecast

3.1 This chapter explores how fraud and error affects the public finances and our medium-term forecasts of welfare spending. We also set out how the findings in the previous chapter, alongside ongoing forecast development work with the Department for Work and Pensions (DWP), have led us to make changes to our fraud and error forecast methodology, and the wider lessons and risks this analysis highlights for our forecast.

How fraud and error affects our welfare spending forecasts

3.2 There are four separate ways that fraud and error feeds into our welfare spending forecasts. Chapter 2 explains that welfare overpayments each year are estimated based on an annual DWP sampling exercise. To capture the effect of overpayments on the public finances we need to consider how many of the total estimated overpayments derived from the sampling exercise evidence will actually be identified in future and then subsequently recovered. Figure 3.1 below sets out these stages for universal credit (UC). We therefore forecast how many overpayments in the forecast are: (1) made; (2) identified; (3) recovered; and (4) written off. Details on how fraud and error affects public sector net borrowing are presented in Box 3.1 later in this chapter.

Figure 3.1: The four stages of fraud and error in universal credit

Note: The identification and recovery rates presented in this figure are DWP estimates based on identification and recovery trends in UC so far, and assumptions about how these trends may persist or change in the future; the methodology to produce these estimates is being refined by DWP and may be subject to change in the future. This figure is illustrative of how overpayments made in a given year will eventually affect identifications, recoveries, and write-offs; this is different to our medium-term forecast processes for identifications, recoveries, and write-offs, which are described in paragraph 3.12 and the latest forecasts presented in Table 3.1.

Source: DWP, OBR

How we forecast fraud and error

3.3 For all benefits aside from universal credit, fraud and error rates have been largely stable over time (Chart 2.1 in Chapter 2), and therefore our forecasts will generally simply assume the fraud and error rates seen in latest outturn will persist through the forecast period.^[23] For UC, as discussed in Chapter 2, the fraud and error rate spiked in 2020-21 before falling back significantly over the following five years. In our March 2026 forecast,^[24] we expected UC fraud and error rates to continue to fall over the forecast period (Chart 3.1). This forecast is developed through the following key steps:

• accounting for the expected changing composition of UC spending by conditionality group in the forecast;
• making an assumption, based on available evidence, on trends in the underlying propensity to commit fraud; and
• accounting for the impact on fraud and error of DWP activity and policy initiatives over the forecast period.

3.4 The subsequent sections describe each of these steps in more detail.

Chart 3.1: Universal credit fraud and error rate: outturn and March 2026 forecast

Source: DWP, OBR

Universal credit spending composition

3.5 The composition of UC spending is expected to shift slightly over the forecast period towards a higher share of spending on the inactive group (see Table 2.2 in Chapter 2 for details on UC conditionality groups). Given inactive cases tend to have a much lower overpayment rate (7.7 per cent of spending in 2024-25) compared to in-work and unemployed cases (13.2 and 13.7 per cent of spending in 2024-25 respectively), this change in composition is expected to reduce the overall UC overpayment rate by 0.6 percentage points at the forecast horizon (Chart 3.2).

Chart 3.2: Composition of universal credit spending by conditionality group and impact on March 2026 fraud and error rate forecast

Note: ‘In work’ refers to cases in the light touch or earning enough conditionality groups, unemployed to cases in the intensive work search group, and inactive to cases in the no work-related requirements, work preparation, or work-focused interview groups. See Table 2.2 in the previous chapter for further details.

Source: DWP, OBR

Underlying propensity for fraud and baseline departmental activity

3.6 Recent UC forecasts have incorporated an assumption on the underlying propensity to commit fraud over the forecast period. This is intended to capture the overall impact of a range of potential factors that could lead to increases or decreases in the future underlying fraud and error rate. These include factors such as changing economic circumstances, developments in the design of the welfare system, and changing societal views in relation to fraud.

3.7 At previous forecasts, we have assumed a net increasing propensity for fraud in UC across the forecast period of 5 per cent each year (dashed yellow line in Chart 3.3). This reflected evidence collected by DWP of increasing fraud across society and increased opportunity for fraud due to the digitalisation of the claims process and to greater information sharing on opportunities for fraud via social media.^[25] This upwards pressure was assumed to be partially offset by operational improvements in DWP’s business-as-usual baseline fraud and error prevention activity across the forecast period (solid yellow line in Chart 3.3).

Chart 3.3: March 2026 universal credit fraud and error rate forecast: capturing underlying propensity and baseline activity

Note: The 5 per cent rising underlying propensity assumption refers to an annual 5 per cent increase in pre-measures fraud. Baseline departmental activity refers to additional fraud and error savings from UC ‘continuous improvement’ initiatives in the forecast period.

Source: DWP, OBR

Additional policy measures

3.8 In addition to business-as-usual preventative activity, DWP has made significant investment at recent fiscal events in initiatives to further reduce fraud and error. In the forecast, we make an additional adjustment to reflect the estimated impact of this on fraud and error. This additional resource includes:

• Targeted case reviews (TCRs). Announced in 2022, by March 2025 DWP had an additional 6,900 FTE staff undertaking reviews of UC cases at highest risk of fraud and error to identify and amend incorrect claims. Despite all additional staff now being in place, TCR savings are estimated to continue to increase across the forecast period on the assumption that it takes several years to reach steady-state savings. By 2030-31, DWP expects TCRs to reduce UC overpayments by £3.2 billion, a saving £3.0 billion larger than the 2024-25 impact.^[26]
• Additional core counter-fraud staff. Announced initially in 2021 and extended indefinitely in 2022, DWP increased its core counter-fraud staff from 4,700 full-time equivalent (FTE) staff in 2020-21 to 6,600 FTE in 2024-25, and intends to further increase staff to 9,700 FTE by 2027-28, more than double 2020-21 levels.^[27] The total impact of all core counter-fraud activity is expected to reduce overpayments by £1.0 billion in 2030-31, though most of this annual saving is already seen in 2024-25 outturn (£0.9 billion).
• Other policy measures. The main measures in this group are DWP’s eligibility verification measure (EVM) and periodic redeclaration of earnings (PRE). EVM, announced in 2023 for implementation in 2026, increased DWP’s legislative power to check claimant eligibility against capital rules with banks, and is expected to reduce fraud and error by £0.3 billion in 2030-31. PRE, announced in 2024 and introduced in 2025, asks claimants to periodically redeclare their circumstances and is expected to reduce overpayments by £0.2 billion in 2030-31.

3.9 Overall, DWP has committed very significant additional resource to combatting fraud and error following the pandemic. Across the 2020s, DWP spending on combatting fraud and error is forecast to rise from £150 million in 2020-21 to £940 million in 2029-30 (Chart 3.4). Dedicated FTE staffing is forecast to increase (from 4,700 in 2020-21 to 16,600 in 2029-30). This significant expansion of fraud and error initiatives by DWP is the main factor explaining the substantial forecast reduction in the UC fraud and error rate shown in Chart 3.5, to a low of 7.5 per cent of spending in 2028-29 and 2029-30, which would be 2.2 percentage points below 2024-25 outturn.

Chart 3.4: DWP spending and staffing on combatting fraud and error: outturn and March 2026 forecast

Source: DWP, OBR

Chart 3.5: March 2026 universal credit fraud and error rate forecast: capturing additional policy

Source: DWP, OBR

Changes to methodology

3.10 Following the review of recent trends in Chapter 2 of this report, we have decided to change the methodology we will use at future forecasts regarding the underlying propensity for fraud. Chapter 2 shows that the main cause of higher UC fraud and error rates in outturn after 2019-20 was likely the higher rates of fraud and error in the Covid cohort. The analysis in Chapter 2 suggests that this effect has now largely dissipated, leaving overall fraud and error rates at a similar level to before the pandemic. This does not suggest there is a rising underlying propensity for fraud and error in the UC caseload.

3.11 Considering this, we intend to change our methodology to assume a flat propensity for fraud and error in the future. In practice, this means removing the two stages of forecast development shown in Chart 3.3 above, which accounted for the underlying propensity assumption partly offset by baseline prevention activity. Given the uncertainty around this issue, and in the absence of clear evidence of an upward or downward underlying trend, we judge it is reasonable to assume any upwards and downwards pressures will broadly net out over the medium term. This also aligns with our approach to forecasting HM Revenue and Customs’s tax gap – which similarly measures the overall rate of compliance within the tax system.^[28] We expect this methodology change to reduce UC spending in the forecast period. This new assumption is highly uncertain and we will keep it under review as more outturn data becomes available.

Overpayment identifications, recoveries, and write-offs

3.12 As set out in paragraph 3.2, there are four stages which determine the effect of overpayments on the public finances. Having produced a forecast for fraud and error, i.e. the monetary value of overpayments, in each year we need to forecast how much will be identified, recovered, and written off to determine fraud and error’s full fiscal effects. This section details how DWP’s identifications, recoveries, and write-offs of overpayments are captured in our forecast (Table 3.1), and their effects on public sector net borrowing (Box 3.1):

• Identifications occur when DWP identifies that it has overpaid a current or previous claimant and subsequently refers the overpayment for debt recovery. Identifications totalled £1.9 billion in 2024-25, with £1.4 billion (70 per cent) of this relating to UC. Our forecast for identifications is driven by overpayments identified in the latest year of outturn as a share of the total overpayments estimated to have been made in that same year.^[29] To forecast identifications in future years, this share is adjusted to reflect the expected impact of any relevant policies due to be implemented over the course of the forecast, and then applied to our forecast for overpayments made in each year. In our March 2026 forecast, identifications across welfare are projected to increase slightly from £2.4 billion in 2025-26 to £2.7 billion in 2030-31, driven mainly by additional initiatives to identify fraud and error within UC.
• Recoveries occur when DWP recovers a previously identified overpayment. These were estimated to be £1.3 billion across welfare in 2024-25. Our forecast for recoveries is largely driven by assumed recovery rates based on recoveries observed in recent years, the duration composition of the stock of identified overpayments, and the expected total stock of overpayments identified across the forecast. In our March 2026 forecast, recoveries are expected to increase from £1.3 billion in 2025-26 to £1.7 billion in 2030-31, driven by the rising value of the total stock of overpayments identified.
• Write-offs occur when DWP decides it will no longer try to recover previously identified overpayments. Total write-offs across welfare are expected to increase slightly over the forecast, from £0.4 billion in 2025-26 to £0.5 billion in 2030-31, driven by the rising stock of overpayments identified.^[30] Write-offs for UC overpayments are significantly lower, averaging less than £0.1 billion a year over the forecast. This is mainly because write-offs are skewed heavily towards older-duration overpayments, and given UC has only been rolled out over recent years the stock of UC overpayments, compared to overpayments of other benefits, relates mostly to more recent overpayments.^[31]

Table 3.1: Identifications, recoveries and write-offs in our March 2026 forecast

Lessons for, and risks to, our forecasts

3.13 As set out above, the findings in Chapter 2 regarding the drivers of recent trends in fraud and error have led us to change our forecast methodology for the underlying rate of fraud and error in UC. The evidence suggests that the sharp rise and subsequent fall in fraud and error across all of welfare after 2019-20 was primarily driven by the Covid cohort on UC, rather than a widespread increase in fraud and error across UC as a whole. As a result, we have decided that in our next forecast we will remove our previous assumption that there is an ongoing underlying increase in the propensity for fraud and error over the forecast period.

3.14 The analysis in Chapter 2 also highlighted that fraud and error rates in the past decade have been quite volatile. This has been driven by several concurrent underlying trends which we will continue to monitor closely as we develop and produce future forecasts, including:

• How the completion of the move to UC will affect steady-state fraud and error. All outturn trends in UC fraud and error are taken from a ‘partial’ UC caseload, where some but not all of the move to UC has taken place and therefore the UC caseload composition is different from what we would expect in a post-move-to-UC world.
• How wider societal trends and changing attitudes to and opportunities for fraud may affect trends in the underlying propensity for fraud. Some wider evidence indicates changing societal views towards fraud. The digitalisation of benefit administration through UC provides new potential routes for fraud never observed before. These factors may contribute to significantly different trends in the underlying propensity for committing welfare fraud to those seen in previous decades, and we will continue to review evidence on this as it becomes available.
• How effective the additional policy initiatives introduced by DWP to address fraud and error will be in the forecast period. The ever-changing nature and composition of fraud within UC, along with potential diminishing returns from further initiatives acting on a smaller pool of fraud and error than seen in the early 2020s, means the extent of effectiveness of new initiatives is highly uncertain.

3.15 The treatment of overpayment identifications in the current National Accounts framework, with the implicit assumption that they will be recovered in full once identified, creates a fiscal illusion that underestimates the cost of fraud and error in our medium-term forecast. We have raised concerns about this accounting treatment with the ONS which has recently stated that it will assess it.

Index of charts and tables

• Chapter 1: Executive summary
  • Chart 1.1: Fraud and error rate across welfare by benefit group
  • Chart 1.2: Universal credit fraud and error rate in the Covid and non-Covid cohorts
• Chapter 2: Recent trends in fraud and error
  • Chart A: Universal credit fraud and error rates and confidence intervals
  • Chart 2.1: Fraud and error rates across welfare by benefit group
  • Chart 2.2: Universal credit fraud and error rates in the Covid and non-Covid cohorts
  • Chart 2.3: Universal credit fraud and error rates by cohort and duration
  • Chart 2.4: Decomposition of change in universal credit fraud and error rate by cohort and conditionality groups, 2019-20–2020-21
  • Chart 2.5: Composition of universal credit fraud and error by reason and cohort
  • Chart 2.6: Change in universal credit fraud and error rate, 2020-21–2025-26
  • Chart 2.7: Universal credit fraud and error rates by cohort and conditionality group, 2020-21–2025-26
  • Table 2.1: 2018 estimated impact of transition to universal credit on fraud and error
  • Chart 2.8: Legacy benefit fraud and error rates
  • Chart 2.9: Legacy benefit and universal credit fraud and error rates
  • Table 2.2: Conditionality groups in universal credit and legacy benefits
  • Chart 2.10: Legacy benefit and universal credit fraud and error rates by conditionality group
  • Chart 2.11: Legacy benefit and universal credit fraud and error rates by reason
  • Chart 2.12: Income-related fraud and error rates in legacy benefits and universal credit
• Chapter 3: Fraud and error in the forecast
  • Figure 3.1: The four stages of fraud and error in universal credit
  • Chart 3.1: Universal credit fraud and error rate: outturn and March 2026 forecast
  • Chart 3.2: Composition of universal credit spending by conditionality group and impact on March 2026 fraud and error rate forecast
  • Chart 3.3: March 2026 universal credit fraud and error rate forecast: capturing underlying propensity and baseline activity
  • Chart 3.4: DWP spending and staffing on combatting fraud and error: outturn and March 2026 forecast
  • Chart 3.5: March 2026 universal credit fraud and error rate forecast: capturing additional policy
  • Table 3.1: Identifications, recoveries and write-offs in our March 2026 forecast