Improper and Phantom Enrollment Predict Exchange Attrition

From February 2025 to February 2026, exchange enrollment declined by 2.6 million people, or 12 percent, according to monthly effectuated enrollment data from the Centers for Medicare and Medicaid Services. Enrollment has declined for three reasons: 1) removal of phantom enrollees for non-payment of premiums, 2) removal of improper enrollees from program integrity efforts, and 3) proper and improper enrollees deciding that the coverage is no longer worth a higher premium, which is often any premium since the enrollees in zero-premium plans dropped by about 3 million.

Because prior Paragon research concluded that improper and phantom enrollment substantially inflated ACA exchange enrollment, we would expect states with the greatest improper and phantom enrollment to experience the largest enrollment declines as eligibility verification tightened and zero-premium coverage became less common. This Prognosis presents four tests of that hypothesis.

The results reveal a consistent pattern—states with higher percentages of improper enrollees experienced much greater enrollment declines and states with higher percentages of zero-claim enrollees experienced much greater enrollment declines. The evidence is consistent with the hypothesis that improper and phantom enrollees have been disproportionately removed from the exchanges. That conclusion is also consistent with a recent Health and Human Services report estimating that the entire net decline in enrollment resulted from the removal of improper and phantom enrollees. In other words, we find that the states whose enrollment was most inflated by improper and phantom enrollment have predictably had much greater enrollment declines as verification tightened and zero-premium plans became less common following the expiration of the COVID-era subsidy boost. Phantoms cannot pay premiums, so a positive premium payment will result in insurers cancelling their coverage.

Paragon defines improper enrollment as the number of enrollees claiming income between 100 and 150 percent of the federal poverty level (FPL) in excess of the number of people eligible for subsidized exchange plans who plausibly have income in that range. Enrollees in this category qualify for the largest subsidies—frequently zero-premium plans. This creates a strong incentive for enrollees, brokers, and insurers to misstate enrollee income to maximize respective subsidies, commissions, and revenues.

A large share of improper enrollees are phantom enrollees: individuals who appear in enrollment counts but are fictitious, unaware they were enrolled, or are already covered elsewhere. Because phantom enrollees do not use any coverage, they are a subset of zero-claim enrollees — those who generate no medical claims during the coverage period. In 2024, 35 percent of all exchange enrollees, and 40 percent of fully subsidized enrollees, filed no claims, more than double the rate expected in a normal health insurance market.

The phantom share of zero-claim enrollment rises as the overall zero-claim rate rises. In states with roughly 35 percent of zero-claim enrollees, we assume that about 15 percent of enrollees are phantoms.

Figure 1 shows that states with larger shares of zero-claim enrollees in 2024 experienced much larger declines in effectuated enrollment between February 2024 and February 2026. (We showed the result in Figure 1 for zero-claim enrollees in 2024 because that is the most recent year zero-claim data is available.) This is exactly what we would expect if many zero-claim enrollees were improper or phantom enrollees who were later removed from the program. The slope of the regression line is −1.03, meaning each additional percentage point in a state’s zero-claim share is associated with about a 1.03-percentage-point decline in effectuated enrollment over the two-year window. The correlation coefficient (r = -0.56) indicates a moderately strong inverse relationship between these two measures.

HealthCare.gov states are shown in blue, states that transitioned from HealthCare.gov to state-based exchanges at some point are shown in orange, and states that have been state-based exchanges since the exchanges launched in 2014 are in red. As previous Paragon research has documented, enrollment fraud and resulting improper enrollment has been much more severe in HealthCare.gov states. This visual pattern is striking and, as we will show below, remains consistent across several different ways of analyzing the data. Most state-based exchanges and states that have transitioned to state-based exchanges cluster in the upper-left portion of the figure with relatively low zero-claim rates and effectuated enrollment increases, while many HealthCare.gov states cluster in the lower-right portion with both high zero-claim rates and large enrollment declines.

One of the most striking features is the clustering of states with between 35 and 40 percent of enrollees having no claims (Arizona, Indiana, Mississippi, Missouri, North Carolina, Oklahoma, and South Carolina) with an average 25 percent decline in effectuated enrollment between February 2024 and 2026. The figure also shows that significant phantom enrollment has likely persisted in Georgia, Florida, Louisiana, and Texas—states with high rates of zero-claim enrollees but little change in effectuated enrollment from February 2024 to 2026.

Figure 2 shows a clear relationship between estimated improper enrollment and subsequent enrollment declines. States with higher estimated improper enrollment generally experienced larger declines in effectuated enrollment between February 2025 and February 2026, consistent with the hypothesis that program integrity efforts and the reduction of zero-premium plans disproportionately removed improper enrollees. The slope is −0.31: each additional percentage point of improper enrollment as a share of total exchange enrollment is associated with a 0.31-percentage-point decline in effectuated enrollment from 2025 to 2026. Once again, a relatively high correlation coefficient (r = -0.47) indicates a close relationship between the two variables.

Figure 3 provides especially strong evidence because it uses our 2025 estimates of improper enrollment—developed before the subsequent enrollment decline—to predict the enrollment decline between 2025 and 2026. Since the improper enrollment estimates predate the decline, the relationship cannot simply reflect the two measures relying on the same underlying enrollment data. It holds regardless: the slope is −0.36 (r = -0.53), close to Figure 2 in both direction and strength. An estimate developed a full year earlier still predicts the following year’s attrition, suggesting the relationship is stable rather than an artifact of one year’s estimate of improper enrollment.

One alternative explanation is that states with more healthy enrollees simply experienced larger voluntary losses of legitimate coverage. But that explanation does not fit the evidence particularly well. There is no obvious reason why ordinary healthy enrollees would be disproportionately concentrated in the states with the highest estimated improper enrollment. By contrast, if our improper enrollment estimates capture real differences in program integrity across states, we would expect them to predict subsequent enrollment declines once eligibility verification strengthened. That is precisely what Figure 3 shows.

Finally, Figure 4 examines the change in enrollment between 2026 open enrollment sign-ups and February 2026 effectuated enrollment. States with higher estimated improper enrollment after the most recent sign-up period experienced much larger declines, suggesting that many improper enrollments never translated into paying enrollees. The slope is −0.44, meaning that each additional percentage-point of improper enrollment is associated with a 0.44 percentage-point decline in effectuated enrollment relative to initial sign-ups. The relationship is particularly strong (r = -0.68). Once again, states with state-based exchanges are clustered on the top left, showing little change between open enrollment and February effectuated enrollment. In contrast, in states with a high percentage of improper enrollees—which is strongly associated with phantom enrollment—the drop-off was much more significant.

Florida, Georgia, and Texas

According to our estimates in The Persistent Obamacare Enrollment Fraud, there were nearly 4.5 million improper enrollees in the states of Florida, Georgia, and Texas after the 2026 open enrollment period. From 2026 open enrollment to February 2026 effectuated enrollment, enrollment declined by 1.7 million people in these states. These states all appear on the right side of the four figures and the decline in effectuated enrollment is well above the regression line—indicating that enrollment has declined less in those states than is predicted by the share of improper and zero-claim enrollees. If 75 percent of the enrollment decline in these states reflected the removal of improper enrollees, roughly 3.2 million improper enrollees would still have remained in February 2026.

Conclusion

Taken together, the four figures tell the same story despite measuring different aspects of enrollment. States with the greatest amounts of improper and phantom enrollment consistently experienced the largest declines in effectuated enrollment after program integrity efforts intensified and zero-premium plans became less prevalent. That is what we would expect if much of the observed decline reflects the removal of duplicate, improper, and phantom enrollment rather than the loss of legitimate coverage.