Defending the Paragon ACA Improper Enrollment Analysis

The Simple and Transparent Methodology for Estimating Improper ACA Exchange Enrollment

The analysis in The Persistent Obamacare Enrollment Fraud replicates and extends our prior work comparing open enrollment sign-up levels with estimates of the plausibly eligible population derived from the Census Bureau’s American Community Survey (ACS).

The central question is the degree to which ACA exchange enrollment in the 100–150 percent FPL income band exceeds the plausible number of individuals who could legitimately be enrolled—indicating improper enrollment. Our analysis focuses on the 100–150 percent FPL group because enrollees in this income range receive the largest subsidies, driving net premiums close to or equal to zero and creating strong incentives for enrollees, unscrupulous brokers, other dishonest insurance intermediaries, and health insurers to misreport income to maximize subsidies and commissions.

Data Sources

The analysis draws on three distinct data sources, each contributing a specific element to the final estimates:

1. The 2026 State-Level Open Enrollment Period Public Use File. This data is reported by CMS and details the observed exchange enrollment during the open enrollment period by income band and state. This dataset also supplies the CMS designation of whether the state uses a state-based exchange (SBE) or the federal exchange hosted by HealthCare.gov.
2. 2024 ACS Integrated Public Use Microdata Series (IPUMS) Extract. This dataset utilizes the ACS 1-Year Estimates to derive our population-level eligibility estimates for the 100–150 percent FPL income band. It supplies information on age; poverty level; and Medicaid, Medicare, or other government coverage enrollment by state.
3. Census Bureau Annual Estimates of Resident Population. We use the Census Bureau’s NST-EST2025-POP file to adjust the ACS estimates forward one year to match the open enrollment data year. We gather the estimates of total population for each state for years 2021, 2024, and 2025.

Preparatory Data Steps

Before conducting the main analysis, we assembled several supporting datasets.

State Exchange Platform Classification

We imported and cleaned the 2026 OEP state-level file (available on the CMS website) to retain one record per state, identifying whether each state uses the federal platform (HealthCare.gov) or an SBE. When comparing aggregate improper enrollment levels between HealthCare.gov and SBE states, we created a third category for the nine states—Georgia, Illinois, Kentucky, Maine, Nevada, New Jersey, New Mexico, Pennsylvania, and Virginia—that have recently switched from HealthCare.gov to SBEs.

Population Growth Factors

We imported U.S. Census Bureau Annual Estimates of Resident Population and computed a growth factor as:

We used this factor in the 2025 estimates to project the 2024 ACS-derived counts forward by one year. This updates the 2025 numbers from last year’s “Greater Obamacare Enrollment Fraud” report to ensure we present the latest information.

We used a second growth factor, defined below, to obtain 2026 estimates from the 2025 estimates. Because the Census Bureau has not released state-level population estimates for 2026 at the time of publication, we used the state-level average annual growth rate from 2021 to 2025.

Eligibility Estimation from the ACS

ACS Sample Selection

We loaded the raw ACS IPUMS extract and restricted it to the 2024 ACS 1-year sample. Minnesota, New York, and Oregon were excluded from our analysis, because they each operated Basic Health Programs as of January 1, 2025, routing low-income individuals outside the standard exchanges. We also excluded the District of Columbia due to missing income information among the majority of plan selections reported in CMS documentation.

Target Population Definition

We classified an individual as part of the target population if all of the following conditions were met:

• Ages 19–64 (inclusive)
• Not covered by Medicaid (hinscaid = 1 in IPUMS coding, indicating no Medicaid or other government coverage for those with low incomes or disabilities)
• Not covered by Medicare (hinscare = 1 in IPUMS coding, indicating no Medicare coverage)
• Income between 100 percent and 149.9 percent FPL (cbpoverty > 100 and < 150)

Importantly, we included people who reported enrollment in an employer-based plan, one key reason why we expect the potential eligible enrollment category is somewhat high (and our reported improper enrollment rate thus somewhat low). Survey person-weights are summed within each state to produce a weighted count of the eligible population.

Medicaid Expansion Adjustment

In states that expanded Medicaid, many adults with incomes between 100 and 138 percent FPL are generally eligible for Medicaid rather than subsidized exchange coverage. To avoid overstating the population plausibly eligible for exchange enrollment, we apply a conservative adjustment in expansion states by reducing the estimated eligible population by 50 percent. This adjustment approximates the share of individuals within the 100–150 percent FPL range who would generally fall below the 138 percent Medicaid eligibility threshold in expansion states. Non-expansion states did not receive this adjustment.

This adjustment, which was also used in our 2024 and 2025 reports, likely overstates exchange eligibility (assuming a uniform income distribution in the 100–150 percent FPL range) and thus is another reason why our estimate of improper enrollment is likely conservative. We rounded to the nearest whole person.

Population Growth Adjustment

Because the 2026 plan year OEP data reflects a population two years more recent than the 2024 ACS, we scaled the Preliminary Eligible Pop count upward before applying the Medicaid expansion adjustment:

Raw Eligible Pop = Preliminary Eligible Pop
× (1 + Growth Factor 2025) × (1 + Growth Factor 2026)

The growth factor is derived from the U.S. Census Bureau NST-EST2025 estimates and is applied at the state level. This adjusted figure then undergoes the same Medicaid expansion halving rule as described above to obtain the final estimates of eligible enrollees.

Improper Enrollment Calculation

For each state, the number of improper enrollees is calculated as:

Improper Enrollees = OEP Enrollment (100–150% FPL) – Eligible Enrollees

OEP enrollment is the observed exchange sign-up count from the CMS OEP state-level files, and eligible enrollees reflect the ACS-derived estimate described above.

States where this difference is negative (i.e., observed enrollment does not exceed the eligibility estimate) are set to zero, indicating no detectable excess enrollment. Only positive values—where sign-ups exceed the estimated eligible population—are reported as improper enrollment.

We also express improper enrollment as the ratio of observed sign-ups to estimated eligible enrollees, providing a relative measure of improper enrollment intensity by state.

Methodological Notes and Limitations

• Eligibility estimates are conservative. The halving rule applied to Medicaid expansion states (Section 3) produces an upper-bound estimate of exchange eligibility. The share of 100–150 percent FPL residents eligible for exchange coverage in expansion states is likely smaller, because the number of people in the 100–138 percent FPL range likely exceeds the number of people in the 138–150 percent FPL range.
• Population projection for 2026. The 2026 estimates adjust the 2024 ACS counts using state-level population growth factors derived from Census Bureau estimates. This approximation assumes uniform compositional changes within the 100–150 percent FPL group relative to the general population.
• Reasons for improper enrollment. The improper enrollee estimates capture any enrollment in excess of the estimated eligible population. This includes fraudulent enrollment by brokers or third parties, administrative errors, income misreporting, and legitimate enrollment by individuals whose circumstances the ACS does not fully capture. The analysis cannot distinguish among these cases.

Methodology for Estimating Improper Spending

To obtain an estimate of the fiscal cost of improper enrollment in the 100–150 percent FPL category, it is necessary to distinguish between enrollees below the income threshold (100 percent FPL) and enrollees above the income threshold (150 percent FPL), as each type of improper enrollee imposes different costs on the federal government.

We estimate that in non-Medicaid-expansion states, two-thirds of improper enrollees (3.42 million people) have income below 100 percent FPL. We assume that the remaining improper enrollees (2.79 million people)—in both Medicaid expansion and non-expansion states—have income above 150 percent FPL. The average subsidy for a 45-year-old with income between 100 and 150 percent FPL is $8,055—this is the cost of improper enrollment for people with income below 100 percent FPL. For people above 150 percent FPL who underestimate their income, we assume an average loss to the taxpayer of $1,800, which is approximately the average difference in subsidy level between the 100–150 percent FPL category and enrollees above 150 percent FPL.

Two additional factors influence the federal costs of improper enrollment. First, subsidy recapture reforms instituted for the 2026 plan year under the OBBB require enrollees above 100 percent FPL who receive subsidies based on incorrect estimated income to reconcile those amounts with their actual year-end earnings and repay any excess subsidies received. We estimate that this recapture rule will affect half of improper enrollees with incomes above 150 percent FPL, while half of enrollees will evade the recapture requirement, possibly failing to file taxes.

Second, because exchange enrollment declines over the course of the year, open enrollment sign-ups overstate average monthly enrollment. Based on historical patterns, we assume that average monthly enrollment in the 100–150 percent FPL category will be 18 percent lower than initial open enrollment sign-ups. Combining this attrition adjustment with the partial-recapture scenario described above, we estimate that improper enrollment will cost approximately $24.7 billion in 2026.

To illustrate the sensitivity of this estimate, the projected cost would decline to $22.0 billion if none of the improper enrollees above 150 percent FPL evade subsidy recapture requirements and attrition reaches 20 percent. Conversely, the estimated cost would increase to $27.7 billion if all improper enrollees above 150 percent FPL evade recapture requirements and attrition is only 15 percent.

Answering Potential Criticisms of Our Methodology

1. “Improper” is a loaded label for a statistical residual

The charge: Paragon takes a gap between two imperfect numbers and brands it “fraud,” even though its own methodology appendix concedes the gap also contains survey error, administrative noise, and legitimate enrollment.

Response: We use the term “improper enrollment,” not “fraud,” precisely because our empirical method – which calculates a residual, or difference, between the number of exchange sign-ups and the number of plausibly eligible people – is broader than fraud. Improper enrollment is enrollment that the eligible population cannot account for. The relevant question is arithmetic rather than semantic: if roughly 6 million sign-ups in the 100–150 percent FPL band exceed the number of people plausibly eligible to be there, those sign-ups do not meet the legal requirements for the program. We characterize that gap using independent evidence of abuse—the concentration of enrollees at the income level that maximizes subsidies, plan selections inconsistent with enrollees’ financial interest, a surge in enrollees with unknown race or ethnicity, high no-claims rates, CMS-documented duplicate Medicaid-exchange enrollment, the GAO’s secret-shopper findings, and DOJ prosecutions of major fraud schemes. Combined, this evidence points to pervasive program integrity failures in the exchanges that have allowed bad actors—including financially motivated enrollment intermediaries—to exploit the program at both taxpayer expense and at the expense of enrollees and brokers who are playing by the rules.

2. The comparison pits administrative counts against a survey that undercounts the poor

The charge: The ACS undercounts low-income and hard-to-reach populations and measures income imprecisely, so comparing an enrollment count from administrative data—which captures virtually all sign-ups—to a survey-based eligibility estimate is apples to oranges. Moreover, the two data sets use fundamentally different measures of income.

Response: We use the best, most complete, and authoritative source for annual information about the income distribution of the US population: the Census Bureau’s American Community Survey (ACS). The income data collected in the ACS has been used in thousands of academic and scholarly studies in health policy. Of course, we acknowledge that the ACS—like any dataset derived from a survey methodology—is imperfect. The definitions of income used by the Census Bureau and used to determine exchange eligibility are not identical. But the differences run in both directions, so it is not clear how this affects the result. Some income (such as child support received) is included in Census money income but does not count toward the modified adjusted gross income (MAGI) used for the exchanges. Other income (such as realized capital gains) is omitted from Census money income but is included in MAGI.

It is also true that if the ACS undercounts low-income people, our eligibility estimate would be too low. But two points matter. First, we calibrate our estimates using ACS person-weights, which are designed by the Census Bureau to adjust for under-coverage (missing certain populations entirely) and household non-response. Second, our conclusion of pervasive improper enrollment does not rest on the Census comparison alone—the independent indicators noted throughout do not depend on the survey at all. And several features of our method point the other way, toward our estimate being too low rather than too high: 1) we assume a 100 percent take-up rate among the eligible, which is implausible; 2) we do not exclude undocumented immigrants, veterans’ health care recipients, or people with access to affordable employer coverage, none of whom qualify for exchange subsidies and all of whom would shrink the eligible population if removed; 3) we exclude four jurisdictions (the District of Columbia, Minnesota, New York, and Oregon) entirely. Finally, when the Census Bureau released revised population estimates, we updated our 2024 and 2025 results and discovered that our earlier estimates had slightly understated, not overstated, improper enrollment.

3. Point-in-time eligibility versus ever-enrolled sign-ups

The charge: Open-enrollment sign-ups count anyone who selected a plan at any point during the open enrollment period, while the ACS is a point-in-time snapshot. Over a year, more people cycle through the 100–150 percent FPL range than are in it on any given day, so a static eligibility count looks artificially small.

Response: The Census uses point-in-time measures of coverage, and we remove people enrolled in Medicaid and Medicare to isolate the population eligible for exchange coverage, which addresses churn into and out of public programs. More fundamentally, this asymmetry cannot explain residuals of the size we find: reported 100–150 percent FPL sign-ups run nearly five times the eligible estimate in Florida and nearly two-and-a-half times the eligible estimate in Texas. A year of income churn does not multiply a state’s eligible population several-fold. And if churn were the driver, the effect would appear broadly; instead it is concentrated in HealthCare.gov, non-expansion states—exactly where the incentives and weak controls are strongest—and is largely absent in states that have consistently used an SBE. Moreover, in our estimates of the fiscal cost of improper enrollment, we account for gradual attrition in the number of enrollees over the course of the year.

4. Income volatility and projected versus actual income

The charge: Eligibility is based on projected annual income at enrollment, while the ACS measures actual income. Low-income earnings are volatile, so a mismatch between the two is expected and innocent.

Response: The amount of advanced ACA subsidies is largely a function of projected income. The ACS measures actual income, and someone who projected income between 100–150 percent FPL in good faith and whose actual income landed elsewhere would not appear in our eligible count even though their enrollment was legitimate. Income volatility and imprecision with estimated income versus actual income may produce some of this mismatch. But the disparities we observe are too large—and too consistent with other sources of evidence—to explain our core results. Volatility is also the mechanism of the misreporting we document: in non-expansion states, a person expecting income below 100 percent FPL has a direct incentive to report income just above the line to qualify for subsidies, and before 2026 faced little or no repayment if actual income came in lower. Brokers and insurers both benefit from more enrollees in fully subsidized plans because many people would drop out of the program with any personal premium obligation. And volatility cannot explain the magnitude or the geography. Residuals do not run 5X eligibility in Florida and 2.4X in Texas because earnings are intrinsically more volatile there than in Massachusetts; they run that high because the enrollment incentives, weak verification, and intermediary activity are concentrated there. Random income mismatch would be spread broadly and roughly symmetrically, causing some people to mistakenly enter the 100–150 percent FPL group and causing others to mistakenly exit the 100–150 percent FPL group.

5. Stale base year and the Medicaid unwinding

The charge: Using a 2024 ACS with growth factors to judge 2026 enrollment ignores that the unwinding pushed millions off Medicaid and into exchange eligibility, expanding the legitimate pool in ways a 2024 snapshot misses. Combining data from different years invalidates the analysis.

Response: We compare 2026 open enrollment sign-ups to 2024 Census data, the most current available, and adjust for population change. The Medicaid unwinding occurred mostly in 2023 and early 2024, and because the Census uses point-in-time coverage measures, most 2024 respondents were surveyed after unwinding was complete or well underway in their state. Because we exclude people enrolled in Medicaid and Medicare when estimating the exchange-eligible population, anyone who transitioned off Medicaid and became exchange-eligible by the 2024 survey is captured as eligible, not counted as improper. Combining data from different years to bridge publication lags is standard empirical practice. The organization that pressed this objection did exactly the same thing in its own work, pairing 2023 Census data with 2025 CMS data in a research analysis. The adjustment is also extremely small and unlikely to make much difference in the overall results as the number of people in an income category changes slowly; our population-growth adjustment changed the estimated eligible count in most states by about 1 percent or less.

6. Plan-switching has innocent explanations

The charge: The silver-to-bronze/gold migration is explained by the expiration of enhanced subsidies plus silver-loading making bronze and gold plans zero-premium—rational consumer and broker behavior, not phantom enrollment.

Response: Silver-loading and the expiration of the enhanced subsidies did make bronze and gold plans relatively cheaper, and we explain that mechanism in the report. For some enrollees a non-silver plan can be a reasonable choice based on premium, provider network, or formulary. But these factors do not explain the clear trends we observe and document in the data. For enrollees between 100 and 150 percent FPL, silver plans still offered dramatically better protection for little additional premium: in 2026 the average bronze deductible was about $7,476, against roughly $80 for a 94 percent actuarial-value silver plan. Trading an $80 deductible for a $7,476 one to save $20–$40 a month is not rational for many of these enrollees. And consumer preferences do not change at a state line or in a single year, yet the shift is concentrated in HealthCare.gov states, largely absent in state-based exchanges, and coincides with much higher active re-enrollment (52 percent of sign-ups in HealthCare.gov states versus 31 percent in SBEs), which is how an intermediary moves someone into a commission-preserving zero-premium plan. Andrew Sprung, an ACA supporter and close tracker of the exchanges, attributes much of the bronze shift to unauthorized broker plan-switching.

7. Unknown race/ethnicity is a data-quality artifact

The charge: Missing demographic fields rose because of optional self-identification, automatic re-enrollment, and broker workflows—not because enrollees are fictitious.

Response: We present unknown race or ethnicity as one corroborating indicator among several, not as direct proof of fictitious enrollees, and automatic re-enrollment may indeed be part of why the share stays elevated. But the trajectory is virtually impossible to explain without the rise of phantom enrollees: the share with unknown race or ethnicity rose from 28 percent in 2020 to 50 percent by 2024, closely tracking the enrollment surge, and reached 56 percent in HealthCare.gov states versus 33 percent in states with SBEs in 2026. An optional field and broker-driven workflows would not produce a HealthCare.gov rate roughly 70 percent higher than the SBE rate unless intermediaries were enrolling people while collecting only the minimum information needed to complete a sign-up—which is what the DOJ cases describe.

8. No-claims rates do not prove phantom enrollment

The charge: Zero-claims enrollees include healthy people, partial-year enrollees, and people who simply did not use care, so referring to them as phantoms overstates the case.

Response: We use these data carefully and have been clear that phantom enrollees are a subset of zero-claim enrollees. Phantom enrollees are enrollees who are unaware of their enrollment, have other coverage, or are fictional—and we estimate that roughly 3 to 4 million annualized enrollees in 2024 were phantoms. Because CMS does not publish claims data limited to full-year enrollees, the figures count anyone enrolled at any point during the year, including partial-year enrollees who had little opportunity to file a claim—so the raw rate overstates phantom enrollment, and we do not treat the headline number as a clean count. What is telling is the trend and the corroboration. The no-claims share roughly doubled, from 20 percent in 2021 to 35 percent in 2024 (40 percent among enrollees in 94 percent actuarial-value plans, also up from 20 percent in 2021), and sits far above the broader private market (about 15 percent for private insurance). This is too large a move to be explained by a stable mix of healthy or partial-year enrollees, and is unique to the ACA exchanges; the small group market did not exhibit similar trends over this period. The strongest corroboration is administrative: CMS found an average of 1.6 million people per month in 2024 simultaneously enrolled in Medicaid/CHIP and subsidized exchange coverage, and in 2025 ended subsidies for roughly 1.5 million people found ineligible or enrolled without authorization. Duplicate and unauthorized enrollment at that scale is exactly what produces enrollees who never file a claim.

9. The cost figure is a tower of assumptions

The charge: The improper-spending estimate stacks assumptions about the income split of improper enrollees, average subsidies, recapture evasion, and attrition, compounding into a number with a wide error band.

Response: The fiscal figure does rest on more assumptions than the enrollment count, which is why we published both the assumptions and a sensitivity range: the estimate runs from about $22.0 billion (no recapture evasion, 20 percent attrition) to about $27.7 billion (full evasion, 15 percent attrition), with a central estimate near $24.7 billion. The figure is also anchored to an independent benchmark (CBO projects $105 billion in ACA subsidies for 2026), so our estimate is roughly a quarter of program spending, consistent with the improper-enrollment share of about a quarter. It is a bounded translation of the enrollment finding, and we acknowledge uncertainty based on excess PTC recovery and the pace of attrition in enrollment.

10. You are calling victims fraudsters

The charge: The report says many enrollees are unwitting victims of broker schemes, so labeling their enrollment “improper” conflates broker misconduct with consumer wrongdoing and tars low-income people.

Response: “Improper enrollment” describes the enrollment, not the enrollee. The report repeatedly identifies many enrollees as victims—people enrolled or switched without their knowledge or consent. Our methodology cannot distinguish between enrollees intentionally submitting false information and enrollment intermediaries exploiting unaware, disengaged, or manipulated enrollees (which we suspect is a much bigger problem), and our report does not level any accusations against individual enrollees. Based on corroborating evidence from GAO and DOJ, the actors primarily responsible for fraudulent behavior are unscrupulous brokers, call centers, EDE platforms, and the insurers who benefit. The reforms we recommend target verification, authentication, and intermediary oversight, not penalties on consumers. Reading “improper enrollment” as an accusation against the enrolled person misreads a structural finding as a personal one.

11. Motivated reasoning and the court ruling

The charge: A federal court in City of Columbus v. Kennedy blocked the CMS rule that relied on this kind of analysis.

Response: Every input in our analysis is public and replicable, and the same conclusion is corroborated by parties with no connection to us—CMS’s own duplicate-enrollment and ineligibility findings, the GAO secret-shopper audit (in which 23 of 24 fictitious applications were enrolled), and multiple DOJ prosecutions. Those independent records establish that large-scale improper and fraudulent enrollment is real. As for City of Columbus v. Kennedy, the decision is widely mischaracterized. The court did not evaluate the merits of our analysis; it concluded only that CMS had not adequately responded to comments criticizing its rule—a procedural finding about agency rulemaking, not a substantive rebuttal of our empirical work.

12. Other data sources show that fraud in the ACA exchanges is low

Response:

This criticism distorts analyses conducted by the Treasury Inspector General for Tax Administration (TIGTA) and CMS—analyses based on data from before the explosion of improper enrollment. In 2023, the TIGTA released a report on improper premium tax credit (PTC) payments through the ACA exchanges. It found:

“As of May 5, 2022, the Internal Revenue Service (IRS) processed 6.2 million Tax Year 2021 returns with $42.5 billion in PTCs that were either received in advance or claimed at the time of filing. A total of $3.6 billion in APTC reported by these taxpayers was in excess of the amount to which they were entitled…”

These results imply that 8.5 percent of PTCs were improperly distributed in tax year 2021. Strikingly, of the 6.2 million tax returns with PTCs in 2021, 2.0 million (32.3 percent) received excess advanced PTC payments.

But the TIGTA’s calculations understate the current scale of improper PTC payments for several reasons:

1. More than 30 million tax returns for the 2021 tax year processed after May 5, 2022, were not included in the TIGTA’s analysis.
2. The 2021 tax year was the first year the Biden administration implemented its expanded COVID credits. Enrollment in the ACA exchanges has since exploded, rising from 12 million in 2021 to 24 million in 2025. Much of this growth has occurred in the 100 to 150 percent FPL category, which is the focus of Paragon’s work.
3. The IG report does not disaggregate its results by income group. Given that the incentives for fraudulent enrollment are strongest for the 100 to 150 percent FPL group, it is likely the improper payments reported by the IG came disproportionately from this group.
4. The IG report does not capture the main group of improper enrollees: individuals who likely didn’t file tax returns. As our papers explain, much of the fraud in the ACA exchanges is driven by unscrupulous brokers and insurance agents who enrolled people without their knowledge or consent. It is likely that many of the people who fell victim to these schemes had such low income that they did not file a federal tax return. (An estimated 10 million people do not file a federal income tax return each year.) These cases would not have been included in the IG’s analysis, which relied on taxpayers’ income reported on tax returns to determine APTC overpayments.

Critics also cite a 2024 CMS report in an attempt to undermine Paragon’s analysis. This analysis, based on 2022 tax year data, examined a random sample of 2,000 applications to the federal exchange. It found that the improper payment rate in the APTC program was 1.01 percent ($563 million) in tax year 2022. But this topline finding is highly misleading.

1. CMS’s reported error rate likely reflects only whether application paperwork was processed correctly—not whether applicants were actually eligible or whether the APTC amounts were accurate. This is a far more limited metric than what Paragon analyzed and does not address the broader and more serious problem of improper enrollment and subsidy overpayments.
2. CMS’s analysis is outdated. Enrollment in the ACA exchanges has increased substantially in the last few years—rising from 15 million in 2022 to 24 million in 2025—especially among lower-income groups.
3. CMS did not disaggregate the results by income and may not have had adequate statistical precision to capture the level of fraudulent enrollment in the 100 to 150 percent FPL group.

A subsequent investigation by GAO, published in December 2025, found large-scale non-reconciliation of APTCs with actual income. The agency reports: “Based on our preliminary analysis, we could not identify evidence of reconciliation in IRS tax data, as of April 2025, for over $21 billion in APTC for enrollees that provided an SSN [social security number] through the federal Marketplace in plan year 2023. This represents approximately 32 percent of APTC paid on behalf of enrollees that provided an SSN to the federal Marketplace for that plan year.”

13. Existing safeguards preclude large-scale fraud

The charge: The exchanges already have identity-proofing, income verification, and broker-conduct rules that make fraud at the scale Paragon claims implausible.

Response:

To maximize enrollment at any cost, the Biden administration systematically weakened or dismantled program integrity measures, which were meant to prevent fraudulent enrollment in the ACA exchanges. For example, the Biden administration:

1. Required all exchanges to accept enrollees’ self-attestation of income if reported income could not be verified through alternate data sources, such as tax filings. This made it particularly easy for individuals with incomes below 100 percent of FPL (who were already eligible for Medicaid) to sign up for fully subsidized ACA plans, since people in this income group often do not file taxes.
2. Sharply limited the amount the federal government can recover if APTCs are paid on behalf of ineligible enrollees.
3. Opened up a special enrollment period (SEP) based on a claim of low income, and eliminated the need for applicants on federal exchanges to submit proof that they qualify for that SEP and other SEPs, essentially permitting year-round open enrollment.
4. Implemented more aggressive automatic re-enrollment procedures, which increased the number of people who never receive a bill and may have remained enrolled for multiple years without their knowledge.

For a more in-depth summary of these changes, see the section “Biden Policies That Exacerbated Improper Enrollment and Fraud on Federal Exchanges” in our report.

GAO’s ability to enroll 23 of 24 fictitious individuals in the exchanges, despite lacking eligibility, demonstrates that the purported safeguards are inadequate.

14. Children counted as applicants but not as eligibles

The charge: Paragon inflated the residual by including children among exchange applicants while excluding them from the eligible-population count—an “apples to oranges” comparison.

Response: Our analysis focuses on the 100–150 percent FPL category, and at that income level children qualify for Medicaid or CHIP in every state, so they are ineligible for ACA subsidies. Thus children at this income level should not be enrolled in an exchange plan. We exclude children from the eligible-population count for that reason, and we define the target population throughout as non-elderly adults ages 19–64 (we likewise exclude seniors, who qualify for Medicare). Children are excluded consistently from both sides of the comparison, not counted as applicants on one side and dropped from eligibles on the other. The “apples to oranges” characterization misdescribes a deliberate and correct eligibility exclusion.