The Fed's $56 billion fraud-loss estimate fell, but its reports do not establish a decline

By Cumulant Research | 2 September 2026

An [Associated Press and FRONTLINE investigation](https://apnews.com/article/scams-online-scams-ai-internet-safety-phishing-fraud-34517de3546d604e20c9d78d3dc5af4f) published on September 2 interviewed 58 people in the United States who said they had lost money to cyberscams. They ranged in age from 32 to 90 and reported losses from several thousand dollars to $4 million. Only Alice Lin, who sued JPMorgan Chase, received any money back. Her lawyers said the bank settlement was for an undisclosed amount.

That is strong evidence about what happened to these interviewees, but it is not a national recovery rate. AP describes an investigative interview group, not a random or probability-based sample of U.S. fraud victims. The finding shows that serious recovery failures occurred among the people AP and FRONTLINE examined. It cannot tell us what share of all victims recover money.

For a broad national measure, the latest available evidence is the Federal Reserve's [2025 Survey of Household Economics and Decisionmaking](https://www.federalreserve.gov/publications/2026-economic-well-being-of-us-households-in-2025-banking.htm). That survey appears to point in two directions at once: estimated gross losses rose, but the dollars estimated to remain unrecovered fell.

Yes, the Fed's published point estimatepoint estimateA point estimate is a survey's single best numerical estimate without a range showing its uncertainty. fell. The [2024 SHEDSHEDThe Survey of Household Economics and Decisionmaking is an annual Federal Reserve survey about the financial lives of U.S. adults.](https://www.federalreserve.gov/publications/2025-economic-well-being-of-us-households-in-2024-banking-and-credit.htm) estimated $84 billion in gross non-credit-card fraudnon-credit-card fraudThis is the Fed's category for financial fraud or scams involving a respondent's money other than credit-card fraud. losses, $21 billion recovered or refunded and $63 billion left unrecovered. The [2025 SHED](https://www.federalreserve.gov/publications/2026-economic-well-being-of-us-households-in-2025-banking.htm) estimated $100 billion gross, $44 billion recovered or refunded and $56 billion unrecovered.

The defensible finding is that the published point estimate fell. The report tables do not establish that the underlying national total fell.

The unrecovered estimate therefore moved down by $7 billion, while gross lossgross lossGross loss is the amount reported lost before any money was recovered or refunded. increased by $16 billion and recovered or refunded moneyrecovered or refunded moneyThis is money a respondent said was restored, although the survey did not ask who returned it or ultimately absorbed the loss. increased by $23 billion. Those are Cumulant Research calculations using the Fed's published whole-billion values.

The catch is uncertainty. The linked Fed tables provide point estimates but do not show confidence intervals for the dollar totals or for the differences between waves. The [U.S. Census Bureau explains](https://www.census.gov/programs-surveys/saipe/guidance/confidence-intervals.html) that a confidence intervalconfidence intervalA confidence interval is a calculated range that describes uncertainty around an estimate made from a sample. describes uncertainty around an estimate made from a sample. Without an interval or a comparable statistical test for the $7 billion difference, the reports do not show whether the movement is large relative to ordinary survey variation.

The arithmetic is gross loss minus recovered or refunded money. For the 2024 wave, $84 billion minus $21 billion gives $63 billion. For the 2025 wave, $100 billion minus $44 billion gives $56 billion. Because the inputs are reported in whole billions, the calculated changes have the same limited precision.

The word "recovered" also has a narrower meaning than readers may assume. The [2024 questionnaire](https://www.federalreserve.gov/publications/2025-supplemental-appendixes-report-economic-well-being-us-households-2024-appendix-a.htm) and [2025 questionnaire](https://www.federalreserve.gov/publications/2026-supplemental-appendixes-report-economic-well-being-us-households-2025-appendix-a.htm) asked how much money respondents recovered or got refunded. They did not ask whether the money came from a bank, another financial company, law enforcement, an insurer, the perpetrator or another source.

The [2024 Fed report](https://www.federalreserve.gov/publications/2025-economic-well-being-of-us-households-in-2024-banking-and-credit.htm) says 63 percent of affected adults who knew the outcome lost money and 32 percent had at least some loss that was not recovered. In the [2025 report](https://www.federalreserve.gov/publications/2026-economic-well-being-of-us-households-in-2025-banking.htm), those shares were 65 percent and 35 percent.

There is no contradiction between a lower total and a higher share of people left short. A small number of high-value recoveries can move the dollar total even while more affected people retain some loss. Conversely, changes in the size of unrecovered cases can move the total without changing many people's outcomes. The published aggregates do not reveal which explanation dominates.

The percentage changes should not be treated as established increases either. The same uncertainty problem applies: the report gives the point estimates but does not provide a confidence interval for the difference between 32 percent and 35 percent.

The Fed's [2024 survey description](https://www.federalreserve.gov/publications/2025-economic-well-being-of-us-households-in-2024-description-survey.htm) reports a final sample of 12,295 adults. The [2025 description](https://www.federalreserve.gov/publications/2026-economic-well-being-of-us-households-in-2025-description.htm) reports 12,934. But most respondents did not answer the detailed fraud-loss questions because those questions were shown only after a person reported non-credit-card fraud.

Appendix B records 624 gross-loss responses in the [2024 wave](https://www.federalreserve.gov/publications/2025-supplemental-appendixes-report-economic-well-being-us-households-2024-appendix-b.htm) and 672 in the [2025 wave](https://www.federalreserve.gov/publications/2026-supplemental-appendixes-report-economic-well-being-us-households-2025-appendix-b.htm). The separate question asking how much was recovered after a partial recovery had 98 and 111 respondents.

Those smaller counts do not make the estimates invalid. They do make uncertainty especially important because fraud losses are heavy-tailed: many losses are relatively modest while a few can be very large. The 2025 Fed distribution table, for example, notes that one reported $5 million loss was removed from that particular distribution table. A handful of large observations can therefore move a nationwide dollar estimate substantially.

The [2025 SHED was fielded from October 17 through October 28, 2025](https://www.federalreserve.gov/publications/2026-economic-well-being-of-us-households-in-2025-description.htm). The [2024 wave ran from October 18 through October 31, 2024](https://www.federalreserve.gov/publications/2025-economic-well-being-of-us-households-in-2024-description-survey.htm). Respondents were asked about fraud experienced in the past year, so each survey looks backward from its field period rather than measuring a clean January-to-December calendar year.

The questionnaires also instructed affected respondents to answer the dollar questions for their [most recent incident](https://www.federalreserve.gov/publications/2026-supplemental-appendixes-report-economic-well-being-us-households-2025-appendix-a.htm). Someone defrauded several times could therefore report the amount from one incident rather than the sum of every loss during the past year. The estimate is an extrapolation from most-recent incidents, not a transaction-by-transaction national ledger.

The two samples are not completely separate. The [2025 survey description](https://www.federalreserve.gov/publications/2026-economic-well-being-of-us-households-in-2025-description.htm) says about one-third of respondents had also participated in the fall 2024 SHED. That overlap matters when measuring uncertainty around a change because responses across waves may be related.

The population benchmarks changed as well. The [2024 weighted sample represented an estimated 259 million U.S. adults](https://www.federalreserve.gov/publications/2025-economic-well-being-of-us-households-in-2024-description-survey.htm), while the [2025 weighted sample represented 265 million](https://www.federalreserve.gov/publications/2026-economic-well-being-of-us-households-in-2025-description.htm). Each wave was adjusted to its own current population benchmark.

Finally, this is a two-point comparison, not a long trend. The [2024 report states that 2024 was the first SHED wave to ask about financial fraud](https://www.federalreserve.gov/publications/2025-economic-well-being-of-us-households-in-2024-banking-and-credit.htm). Two observations can show that one published estimate is lower than another, but they cannot establish a durable direction.

The published totals are consistent with several explanations, and the aggregate tables do not choose among them.

• Recovered cases may have involved larger dollar amounts in the 2025 sample, lifting the recovered total without increasing the share of people made whole.
• Unrecovered cases may have involved smaller amounts on average, lowering the nationwide estimate even as more affected adults remained short.
• The mix of victims, incidents and loss amounts may have changed between waves.
• Different respondents, changed population weights, non-response and ordinary sampling variation may have moved the estimates.
• Several mechanisms may have operated at the same time.

These are possible interpretations, not findings about causationcausationCausation means that one factor produced a change, which cannot be established by observing two survey estimates alone.. Establishing why the estimate moved would require analysis of the respondent-level data, a defensible uncertainty calculation and, for policy claims, evidence linking recoveries to particular institutions or interventions.

The answer has two parts. As a description of the published point estimates, yes: unrecovered non-credit-card fraud fell from $63 billion in the 2024 SHED to $56 billion in the 2025 SHED. As evidence of a real national decline, no: the linked reports do not provide the uncertainty analysis needed to distinguish that $7 billion movement from survey variation.

The safer reading is therefore narrower. The Fed estimated more gross loss, substantially more recovered or refunded money and fewer unrecovered dollars. At the same time, it estimated that a larger share of affected adults was left with some unrecovered lossunrecovered lossUnrecovered loss is another name for the money still missing after any recovery or refund.. That is a changed survey pattern, not proof that America's fraud-recovery problem improved.