What the 2026 OEP Public Use Files Actually Show
The First ACA Enrollment Decline in Years: What the 2026 OEP Public Use Files Actually Show
CMS just dropped the 2026 Open Enrollment Period Public Use Files, and for the first time since the ARP-era enhanced subsidies began, ACA marketplace enrollment fell. Plan selections dropped 4.9%, from 24.3 million to 23.1 million.
But the headline number barely scratches the surface. CMS released the state-level, county-level, ZIP-level, and state-metal-enrollment status files today, and cross-referencing them with the CMS Exchange PUFs (data from rate filings that includes rates, service areas, plan designs, etc. that CMS publishes each fall), the story underneath is far more interesting than “enrollment went down.”
Here’s some of what I found. Fair warning: I went deep into the weeds on this, so buckle up.
The Big Picture: 4.9% Down, But Wildly Uneven
The national 4.9% decline masks extraordinary state-level variation. Only 10 of 51 states/territories actually grew enrollment, and 8 of those 10 are state-based marketplaces. North Carolina lost 22% of its consumers. Texas grew 5.2%. New Mexico grew 18%. Charles Gaba at ACASignups.net has some great analysis already (and H/T to Charles for pointing to these files before CMS had even publicly posted any links), and he’s written about some of the things that likely explain state-level variation like New Mexico’s successful efforts to back-fill lost ARPA subsidies, but I’ll try to break down some of the other factors at play here.
HC.gov (FFM) states declined 5.4% while SBE states declined 3.9% (treating Illinois as SBE in both years, since it became an SBE in 2026). Both platforms lost consumers, but the gap is real: 1.5 percentage points, driven largely by Georgia’s 12.3% decline dragging the SBE average down. Exclude Georgia and SBE states declined just 1.8%. That differential still raises questions about whether the federal platform’s consumer experience, outreach infrastructure, or regulatory environment contributed to worse outcomes, or whether it’s compositional (more on that below).
The Great Silver Exodus
The most dramatic structural shift is in metal level mix. Silver collapsed from 56.2% of selections to 42.6%, while Bronze increased from 29.9% to 39.6%. Nearly 4 million consumers left Silver plans. There was also a meaningful move from silver to gold. A lot of these moves make perfect sense, but the reason may not be immediately evident.
The SLCSP Benchmark Increased ~30%, but not All Metals Moved Equally
The second-lowest-cost Silver plan (SLCSP: the benchmark that determines APTC subsidy amounts) increased roughly 30% on an enrollment-weighted basis across HC.gov and Georgia counties, climbing from $655/month to $854/month. But the SLCSP didn’t move in isolation, and comparing it to other metal levels reveals exactly why the subsidy math shifted so dramatically.
Silver and the SLCSP outpaced the cheapest Bronze option by nearly 8 percentage points, and 86% of marketplace enrollment sits in counties where the SLCSP outpaced Bronze. This isn’t just general rate inflation; Silver moved disproportionately.
Why? First, because of some regulatory action by Texas (a huge Healthcare.gov state) that increased the mandated “silver load” that issuers need to apply due to CSR defunding to 1.45, meaning issuers had to raise their silver rates relative to their bronze and gold rates a lot more. Second, because competitive reshuffling hit Silver hardest while actually suppressing Bronze increases. In counties where the cheapest Silver issuer changed (~7.8 million consumers), Bronze increased just 20.9% while Silver jumped 31.5%, a 10.6 percentage point gap. In stable counties (~7.9 million consumers), the gap was much smaller: Bronze +24.4%, Silver +28.6%. The SLCSP itself landed at +30.5% in both groups, but the Bronze floor was 3.5 percentage points lower in disrupted markets. Carriers exiting or repricing Silver (where benchmark dynamics create unique strategic pressure) left a gap at that metal level, while Bronze (where the competitive logic is simpler, and where new entrants and repositioning carriers aggressively competed) saw prices held down.
The result: in disrupted markets, the spread between what consumers pay for the cheapest Bronze plan and what determines their subsidy widened more than anywhere else. The subsidy math moved most aggressively in consumers’ favor in exactly the counties where they had the most reason to switch.
(Note: All competitive data in this section: issuer positions, rate changes, carrier entries and exits, county-level plan pricing are based on the CMS Exchange PUFs, which cover Healthcare.gov states plus Georgia (which publishes Exchange PUFs in the same format). Other SBE states’ plan-level data isn’t available yet. County-count metrics (like “46% of counties saw a cheapest-Silver change”) include Georgia, but enrollment-weighted figures use HC.gov county enrollment data only, since CMS doesn’t publish county-level OEP PUFs for SBE states.)
46% of HC.gov/Georgia counties saw a change in the cheapest Silver issuer, and those counties represent nearly half of all marketplace enrollment on these platforms. The number of issuer-county combinations in the market dropped 6% (9,118 to 8,568), with average carriers per county declining from 4.18 to 3.92. Aetna / CVS Health’s wholesale exit alone removed 479 county-level positions (the largest carrier withdrawal in recent ACA history). Meanwhile, 351 counties are now duopolies and 40 are outright monopolies.
Centene’s story is the most nuanced. They remain the dominant carrier on the federal platform (1,394 counties across 21 HC.gov states), but their strategy shifted dramatically. Their weighted-average rate increase across all metals was +33.7%, nearly 8 percentage points above the market average of +25.9%. However, Centene’s county footprint barely changed: they entered 34 new counties (25 in Iowa, 7 in Oklahoma, and 1 each in Texas and Alabama) while exiting 41 (37 in North Carolina, 4 in Michigan), a net loss of just 7 counties. They didn’t leave most markets; they repriced within them. By county count they lost roughly 15% of their cheapest-Silver positions (928 to 785), but the enrollment picture is more dramatic: across HC.gov states (where we have county-level enrollment data), the counties they lost represent about 38% of the enrollment in their cheapest-Silver footprint, because the losses were concentrated in high-enrollment counties while the positions they retained skew smaller and more rural.
Georgia tells the story most clearly: Centene lost cheapest Silver in 96 of 159 counties, including all 13 core metro Atlanta counties where Kaiser displaced them, plus 83 rural counties that went to Alliant and Oscar. Metro Atlanta alone represents about half of Georgia’s enrollment, so those 13 counties matter far more than the 83 rural ones. Across HC.gov states, Blues affiliates in Arkansas and Tennessee, Elevance, and Oscar picked up the bulk of the displaced positions. Centene appears to be trading market share for margin (although there’s always the risk this backfires if they raise rates too high, there could be an adverse selection problem because the only people who stick around are those with significant health problems).
When the SLCSP rises, APTC subsidies increase. This means Gold plans, which don’t move in lockstep with Silver, can often be cheaper than silver. And that’s exactly what happened in several states:
Arkansas: Gold went from $98/mo more than Silver to $136/mo less. Gold increasedfrom 4.7% to 31.5%. APTC per subsidized consumer jumped 54%.
Texas: Already Gold-dominant due to silver loading, Gold share rose further from 35.1% to 41.3% as the Gold-Silver inversion deepened (from -$77 to -$127/mo).
Gold share also increased in several SBE states: Washington (18.5%→52.5%) and Illinois (6.7%→31.0%), though we don’t have the plan-level pricing data to confirm the spread dynamics there.
In high-disruption states (where lowest-cost issuers reshuffled), the Gold-Silver spread narrowed by $34/month on average. In stable markets, it actually widened by $10/month. Consumers followed the subsidy math.
The Bronze increase tells the opposite story: states like NC (+18pp), AZ (+20pp), TN (+20pp) saw price-sensitive consumers who lost generous subsidies choose minimal coverage over no coverage.
Who Felt the Pain Depends Entirely on Whether They Shopped
The metal-level shifts above are the aggregate view. The consumer-level view is just as stark. The state-metal-enrollment status PUF breaks out net premiums (after APTC) by enrollment type, and the gradient is enormous:
New consumers: net premiums rose just 5% ($89→$94/month). They picked the cheapest available options (overwhelmingly Bronze and Gold) and largely sidestepped the APTC erosion.
Active renewals: net premiums rose 32% ($89→$117/month). These consumers shopped and often switched metals, but still absorbed significant subsidy loss.
Auto-renewals: net premiums rose 144% ($79→$192/month). They stayed on whatever plan they had (often Silver, often no longer competitively positioned) and ate the full impact of both rate increases and APTC erosion. This may mean a lot of those individuals ultimately end up disenrolling through non-payment attrition.
The gap between active and auto-renewals is the cost of inaction, quantified. Auto-renewed consumers went from paying the least of any group ($79/month) to paying the most ($192/month) in a single year. Active renewals mitigated the damage; new consumers avoided it almost entirely. In 30 of 31 HC.gov states, new enrollees paid lower net premiums than renewals in 2026, with an average gap of $30/month, and in states like Ohio ($73/month), Delaware ($70), and Indiana ($65), the advantage of shopping was even larger.
This connects directly to the switching data: auto-renewal as a share of all consumers dropped from 44.6% to 38.1%, and among re-enrollees, 27.3% actively switched plans in 2026, up from 19.6%. More than 1 in 4 re-enrollees changed plans. The consumers who engaged with their options navigated the SLCSP shift; the ones who didn’t got crushed by it.
Brokers: The Dominant Channel, With Caveats
CMS added Agent/Broker fields to the PUF for the first time this year, and the picture is striking: 78% of all HC.gov enrollment is broker-assisted. This is the overwhelmingly dominant distribution channel on the federal platform.
Does broker penetration protect enrollment? County-level analysis across 2,013 HC.gov counties suggests yes, but with a big asterisk, and several layers worth unpacking.
The headline: High-broker counties (>77% AB share) were nearly flat at -1.9%, while low- and medium-broker counties lost 10%+. That’s an 8-9 percentage point protective effect. High-broker counties were also the only group with positive new consumer acquisition (+1.1%). A decile analysis sharpens the picture: enrollment change moves from -9% in the lowest-broker-penetration decile to roughly flat in the highest, with each step up in broker share corresponding to about 1 percentage point less enrollment loss.
The auto-renewal channel matters too. Among re-enrollees, broker-assisted consumers auto-renew at a 39% rate, compared to 32% for direct-channel consumers. That 7-point gap suggests brokers are placing consumers in plans they’re more likely to stick with, or that the broker relationship itself creates inertia: perhaps members assume their broker would have called them if they needed to change, and brokers were too busy (or too lazy) to engage with members who are in a plan that they didn’t complain about. There’s also the question of how fraud plays into this: if a member was fraudulently enrolled by an unscrupulous broker, the broker may not want to mess with their enrollment (although there’s a decent chance an auto-enrolled member would go from a zero to a non-zero premium, which would mean the end of the fraudulent enrollment and the broker’s commission). Regardless of the cause, though, broker-channel consumers are structurally stickier.
The asterisk: the broker effect is highly state-dependent. Within-state correlations (which control for state-level policy, Medicaid status, and regulatory environment) show broker penetration is strongly protective in Florida (r=0.37), Texas (r=0.37), and South Carolina (r=0.48), but nonexistent in North Carolina (r=0.01) and actually negative in Tennessee (r=-0.29). Across all HC.gov states, 18 of 25 show a positive within-state broker-enrollment correlation. The effect is real, but not universal.
North Carolina is the starkest example: it lost 22% uniformly across all counties regardless of broker penetration, suggesting that when competitive disruption is severe enough (NC had among the highest rates of lowest-cost issuer churn in the Exchange PUF data), even strong broker channels can’t overcome the structural shock. This potentially points to a potential interaction effect: brokers are most effective in moderate-disruption environments where they can guide consumers to alternatives, and least effective where the market itself has been destabilized beyond what any channel can compensate for. There’s also something that’s not accounted for in this data, which is potential impact of broker appointment processes within states. For example, Florida Blue requires its agents to be “captive”: if they want to sell Florida Blue, they can’t sell anyone else (Oscar sued them over this rule and lost.) This sort of thing is something that may be confounding these conclusions; brokers aren’t always appointed with every carrier, and some carriers make it easier than others to get appointed. Some take anyone licensed to sell health insurance, while others are pickier, and these rules are not something I accounted for in this analysis, and it may matter. In the Florida example, if a broker has a member with Florida Blue, they’re not going to be motivated to move them to another option, even if it may be better for the member (e.g. if the benchmark plan has changed dramatically), because they won’t get paid any commissions on that. There are some brokers who will do this because they value long-term relationships with their client and really want what’s best for them, but it would be naive to think the economic incentives don’t influence broker behavior.
County size is not a confounder. Broker penetration has essentially zero correlation with county enrollment size (r=0.03), ruling out the possibility that large urban counties are simply driving both higher broker share and better enrollment outcomes.
This has direct policy implications. CMS’s ongoing broker integrity efforts (which are warranted given documented abuses) need to be weighed against the channel’s clear role in both enrollment preservation and new consumer acquisition. The states with the steepest declines tend to have mid-range broker penetration (65-73%), not the highest, and the data consistently shows that the broker channel is the primary mechanism through which consumers navigate competitive disruption.
Competitive Disruption → Consumer Behavior: The County-Level Evidence
This is where cross-referencing the Exchange PUF data with the county-level enrollment PUF gets really interesting. At the county level, 77% of HC.gov/Georgia counties experienced a change in the cheapest issuer in at least one metal level, far more pervasive than the 46% Silver-specific figure suggests. The question is: does that disruption translate to measurable changes in consumer behavior?
The answer is yes: emphatically for switching, and with a surprising twist for enrollment.
Who Was Exposed: The Metal Composition of Auto-Renewals
Before getting into the county-level mechanics, it’s worth establishing who was sitting in the path of the disruption. The state-metal-enrollment status PUF lets us see the metal composition of each enrollment type, and the gradient is stark. Across HC.gov states, 50% of auto-renewed consumers were on Silver plans, compared to 38% of active renewals and just 27% of new enrollees. Bronze runs in the opposite direction: 51% of new enrollees, 43% of active renewals, but only 37% of auto-renewals. Gold: 22% new, 19% active, 13% auto.
In other words, the consumers who did nothing were disproportionately concentrated in the exact metal tier most affected by the SLCSP shock and competitive reshuffling. And those consumers paid for it: in 2025, 51% of auto-renewals were paying ≤$10/month. In 2026, that collapsed to 21%. By comparison, 53% of new enrollees are paying ≤$10: new consumers found the free plans; auto-renewals lost them.
Quantifying State-Level Disruption
To formalize this, I constructed a composite disruption score for each HC.gov state using three z-scored components: (1) auto-renewal Silver share (higher = more SLCSP exposure), (2) auto-renewal premium increase (higher = bigger shock), and (3) loss of ≤$10 premium share among auto-renewals (bigger loss = more consumers going from free to paying). Each component captures a different dimension of how hard a state’s passive enrollees were hit by the benchmark repricing.
Enrollment-weighted, the correlation jumps to r=-0.37: a meaningful relationship given the number of confounders.
Mississippi sits at the extreme, driven by 75% Silver auto-renewal share and a +481% auto-renewal premium increase. And yet MS only lost 7.3%, because surging new enrollment partially offset the renewal collapse (as discussed above). Oklahoma, Indiana, and Arizona scored high on exposure but didn’t have that new-enrollment offset, and lost 15-16% each.
Active re-enrollment rates are highly predictable from disruption metrics. Among HC.gov re-enrollees, the share who actively chose a plan (rather than being auto-renewed) rose from 54% to 62% at the county level. An OLS regression of this active re-enrollment rate on competitive disruption variables produces R²=0.24, meaning nearly a quarter of the county-level variation is explained by measurable market disruption. for every 1% increase in a county’s highest issuer rate shock, the active re-enrollment rate rises 0.46 percentage points; an issuer exit adds 4.9 percentage points; a new market entrant adds 3.7pp, and a change in the cheapest-Silver rank changes 1.5pp. These effects are all independently significant and additive.
The disruption gradient is clear. In undisrupted counties (no rank changes, no exits, no entries), 53.5% of re-enrollees actively chose a plan. In high-disruption counties (multiple simultaneous changes), it hit 59.8%. Where a new entrant displaced the incumbent as cheapest, it reached 65.7%. (Note: this is the active re-enrollment rate, the broader measure of consumers who engaged with their plan choice. The plan-switching rate, which counts only those who changed to a different plan, was 27.3% nationally in 2026, up from 19.6%.)
Enrollment tells a more complicated story. Competitive disruption doesn’t cleanly predict enrollment decline (R²=0.02); though I think it’s probably got some explanatory power if you separated it from other too many other factors are at play (Medicaid unwinding, local economic conditions, state policy, carrier networks, etc.). But the segmented analysis revealed something that surprised me:
Counties with moderate SLCSP increases (27-36%) had the best enrollment outcomes at -4.0%, while both low-change counties (-10.4%) and extreme-shock counties (-9.4%) did worse. A caveat: Texas counties are heavily represented in the moderate-increase quartile, and as mentioned before, Texas has some oddities given aggressive state-level premium alignment policy. The pattern persists when Texas is excluded, but weakens, so this should be read as suggestive, not definitive. One interpretation: moderate competitive repricing may energize the market as new entrants come in, consumers re-shop, and the benchmark resets to something rational, but extreme shocks (or no change at all in a declining-subsidy environment) produce worse outcomes.
Meanwhile, Gold share change correlates with the Gold-Silver spread at r=-0.52 (p<0.001), the single strongest relationship in the dataset. Where the spread compressed or inverted (Gold becoming cheaper than Silver after APTC), Gold share increased. Consumers are doing exactly what economic theory predicts.
Retention
I computed a simple retention metric: 2026 re-enrollees as a share of 2025 total consumers. Nationally it’s 80.3%: meaning about 1 in 5 consumers from last year didn’t come back.
The variation is enormous:
New Mexico: 99.8% retention (virtually everyone came back, plus 15% new consumer growth = 18% total growth)
Louisiana: 90.5% (sticky market despite biggest affordability hit)
Massachusetts: 89.1%
North Carolina: 65.2% (1 in 3 gone)
Ohio: 67.9%
Arizona: 71.0%
The Mississippi Paradox
Mississippi might be the single wonkiest state in this data. New consumers increased 30% while renewals collapsed 13.5%. Net premiums went from $41/month to $131/month: a 220% increase, the largest nationally. Existing enrollees fled the premium shock, but new entrants poured in, partially offsetting losses. The overall market shrank 7.3%, but the composition of who’s enrolled changed dramatically. I don’t have a good theory for this one yet, so I’m curious if anyone more familiar with policy and market dynamics in Mississippi might have an idea.
The Louisiana Puzzle: Growing Through the Largest Affordability Hit
If Mississippi is the wonkiest state, Louisiana is the most instructive. Louisiana grew 1.2% (one of only two HC.gov states to grow alongside Texas) despite seeing net premiums double from $72/month to $144/month.
How? The parish-level data tells a clear story of rational consumer downshifting in a stable competitive environment.
Louisiana’s competitive landscape barely changed. Only 12% of parishes saw a change in the cheapest Silver issuer (vs 46% across HC.gov/Georgia counties). Zero issuer exits. BCBS of Louisiana (the dominant carrier across all 64 parishes) raised Silver rates 26%, but CHRISTUS Health actually lowered Silver prices 2.7% in its 8 parishes, and UnitedHealth’s increase was a relatively modest 16%. One new entrant (AmeriHealth Caritas) entered 4 parishes. The market went from 4 to 5 issuers, the opposite of the national consolidation trend.
This stability enabled an almost perfectly efficient metal shift. Louisiana lost 38,382 Silver consumers and gained 42,626 Bronze consumers: nearly a 1:1 swap. Gold barely moved (-590). Consumers didn’t leave the market; they downshifted one tier. The Bronze share jumped from 43% to 57%, while Silver fell from 51% to 38%.
The retention numbers look impressive at first glance: Louisiana’s 90.5% retention rate (re-enrollees as a share of prior year’s consumers) is in the top quartile nationally. Renewals actually grew 8.8%, completely offsetting a 36% collapse in new consumer acquisition.
But here’s the problem: 61% of Louisiana’s re-enrollees were auto-renewals: one of the highest passive rates in the country. And those passive renewals are sitting on a ticking clock. In 2025, 70% of Louisiana’s auto-renewals were paying ≤$10/month. In 2026, that collapsed to 25%. That’s roughly 73,000 consumers who went from essentially free coverage to a real monthly bill, and they didn’t actively choose to stay; they just didn’t leave yet. Their average premium jumped from $58 to $151/month. These consumers are technically enrolled, but many of them may not yet realize what they owe, and under the ACA’s 3-month grace period rules, the non-payment attrition wave hasn’t hit yet. They have until tomorrow (March 31) to pay all three months’ worth of premiums, and chances are, if they haven’t paid yet, they’re not going to. Louisiana’s 90.5% retention may look very different by Q2.
Compare this to Mississippi, which had a similar affordability shock (+$90/month net premium increase) but opposite enrollment dynamics: MS saw new consumers increase 30%, while renewals collapsed 13.5%. Louisiana kept its existing consumers on paper; Mississippi replaced them. But Louisiana’s “retained” consumers include a large population that passively rolled over into plans they may not be able to afford.
What explains the point-in-time retention? The most likely explanation is competitive stability itself. When the market doesn’t reshuffle, the path of least resistance is to do nothing, and in Louisiana, doing nothing meant staying enrolled at a much higher price rather than navigating a disrupted landscape. Whether that inertia translates to durable enrollment or just a delayed exit remains to be seen, but I expect that they’ll go from a seeming success story (one of the few growing FFM states) to, at best, a middling decline..
Expansion vs. Non-Expansion: As Usual, it’s complicated
At the headline level, expansion states lost 7.1% while non-expansion states lost only 2.8%. But the aggregate is misleading.
Within HC.gov: expansion states lost 12.9% vs just 1.5% for non-expansion. That non-expansion number is massively skewed by Texas (+5.2%, 4M consumers). Exclude Texas and HC.gov non-expansion drops to -8.1%: still better than expansion’s -10.3%, but a much smaller gap (2.2pp) than the headline suggests.
Within SBE (treating Illinois as SBE in both years): expansion states (which are 20 of 21 SBE states) declined just 1.8%. Georgia is the only non-expansion SBE state, and its 12.3% decline is what drags the overall SBE average from -1.8% to -3.9%.
So why did expansion states fare worse? The FPL data in the state-level PUF tells us. The entire gap is a compositional artifact driven by what’s happening at the bottom of the income distribution.
Non-expansion states have 43% of their HC.gov enrollment in the 100-138% FPL bracket: people who earn too little to qualify for Medicaid (because there’s no expansion) but qualify for very generous Exchange subsidies. That population grew 1.1% in 2026. In expansion states, by contrast, anyone below 138% FPL is on Medicaid, so the Exchange doesn’t have that captive base. Expansion states have just 9.6% of their HC.gov enrollment in that bracket.
The 139-150% FPL group sharpens this further. This is the population just above the Medicaid line in expansion states, and still part of the no-Medicaid-option group in non-expansion states. In non-expansion states, this group declined 21.3%, which is far worse than the 13.1% decline in expansion states. Georgia is the most extreme case: the 100-138% FPL group grew 7.8% while the 139-150% group collapsed 40.3%, evne though those groups are separated by just a few thousand dollars of annual income, in the same state and same competitive environment.
Above 150% FPL, expansion and non-expansion states performed similarly. The 150-200% bracket declined 11.9% in expansion vs 10.8% in non-expansion. At 400%+ FPL, non-expansion states actually fared worse (-51.4% vs -38.5%).
The story isn’t that expansion states have a less resilient Exchange population. It’s that non-expansion states have a captive population below 138% FPL with no Medicaid alternative and nowhere else to go, and that population barely moved, buffering the overall decline. Expansion states don’t have that anchor: their low-income population is on Medicaid, and their Exchange enrollment starts at 139% FPL, where consumers are more price-sensitive and more likely to leave when subsidies erode.
The metal-level data sharpens this further. The state-metal-enrollment status PUF lets us see how the 100-138% FPL population moved across metal levels, and the pattern is striking. In non-expansion HC.gov states (ex-TX), Silver enrollment among the 100-138% group fell 18% (3.48M→2.84M), but Bronze grew 52% (815K→1.24M) and Gold exploded up 700% (41K→326K). Half of all Gold consumers in non-expansion states are now paying ≤$10/month. These consumers followed the subsidy math perfectly: the SLCSP growth made Gold free or near-free for low-income consumers, and many jumped on it. The ones who stayed on Silver are paying dramatically more: the ≤$10 share on Silver collapsed from 41% to 9%; getting that low price of a silver plan is pretty hard unless a carrier is aggressively silver-gapping.
This finding matters for what comes next. The 100-138% FPL consumers who actively shopped and switched to Bronze or Gold are probably fine: they found affordable coverage and will pay their premiums. But the ones who auto-renewed on Silver may have been paying $0 for years, and it’s a big question whether they’ll stick around or now. Auto-renewals among the 100-138% group dropped 34% year-over-year in both expansion and non-expansion states, suggesting a large share of last year’s passive Silver enrollees simply didn’t come back. The ones who did auto-renew are now paying real money for a plan that may no longer be competitively positioned, and they’re disproportionately concentrated in non-expansion states, where the 100-138% Silver population was enormous. This is the population most likely to drive non-effectuation (more on that in the postscript below), and when it happens, a significant chunk of non-expansion states’ enrollment buffer will evaporate.
What It All Means
This OEP tells a story of a marketplace in structural transition. The enhanced APTC expiration was the proximate cause of the enrollment decline, but the downstream effects, such as competitive disruption from issuer repositioning and exits, SLCSP benchmark increases, Gold-Silver pricing inversions, market consolidation (4.18→3.92 carriers per county), and massive variation in broker effectiveness are what actually shaped outcomes at the state and county level.
A few takeaways for anyone working in this space:
The SLCSP is one of the most important numbers in the ACA marketplace and it moved ~30%. Everything downstream follows from the benchmark. The Gold-Silver spread is the clearest proof: where it compressed or inverted, Gold share increased. Consumers follow the math.
Competitive disruption drives active engagement, not exit. Nearly a quarter of county-level variation in active re-enrollment rates is explained by measurable competitive disruption. Rate shocks, issuer exits, and new entry all independently push consumers to actively choose plans rather than auto-renew. A composite disruption score (Silver auto-renewal share, premium shock, and ≤$10 share loss) is enrollment-weighted correlated with enrollment change at r=-0.42, and states in the top disruption tercile lost 10.2% vs 3.4% growth in the bottom tercile.
Moderate disruption is better than none. The “Goldilocks” finding: counties with moderate SLCSP increases (27-36%) had the best enrollment outcomes (-4.0%), while both stable markets and extreme-shock markets did worse. Some competitive churn is healthy for the market.
Broker channels matter enormously but they’re not a silver bullet. The 8-9pp protective effect is real, broker-assisted re-enrollees auto-renew at higher rates (39% vs 32%), and 75% of new HC.gov consumers came through brokers. But the effect is state-dependent and breaks down under extreme competitive disruption; e.g. where NC lost 22% uniformly regardless of broker share. CMS’s broker integrity efforts need to account for the channel’s demonstrated role in enrollment preservation.
SBE states fared better, but not by as much as you’d think. SBE states declined 3.9% vs 5.4% for HC.gov, a 1.5-point gap, not the 8-point chasm the unadjusted CMS totals suggest (Illinois’s platform switch inflates the raw numbers). Exclude Georgia and SBE states declined just 1.8%. Whether the remaining gap reflects platform advantages, state policy environments, expansion status, or compositional factors deserves further investigation. I admit to being an SBE skeptic; this data gives me pause, but I generally think that unless a state is planning to implement its own subsidy program that HealthCare.gov can’t administer, having an SBE actually doesn’t give you much of an advantage vs. sticking with HealthCare.gov. As we’ve seen, states still can drive a lot of decisions through the rate-setting process, and in the example of Texas’s initiatives, those decisions can be quite impactful.
Watch the competitive landscape. Centene’s margin-over-market-share repositioning, Aetna / CVS Health’s wholesale exit, and regional Blues affiliates aggressively picking up cheapest positions reshaped 46% of Silver markets and 77% of HC.gov/Georgia counties across all metals. The next OEP will tell us whether this wave of repositioning stabilizes or creates another round of disruption. But disruptive policy dynamics aren’t the only thing people should pay attention to; the behavior of firms in the market reshapes the market as a whole rather than just reshuffling members.
Postscript: These Numbers Will Get Worse
Everything above is based on OEP plan selections: the snapshot CMS publishes reflecting who chose or was assigned a plan during Open Enrollment. Effectuated enrollment (the number of people who actually pay their first premium and maintain coverage) is always lower. In a normal year, the gap is 3-5%. This is not a normal year, and the gap will almost certainly be larger.
Here’s why. Of the 23.1 million plan selections, 8.8 million (38%) were passive auto-renewals: consumers who took no action and were automatically re-enrolled in their existing plan. In 2025, 42% of all consumers were paying ≤$10/month after APTC. In 2026, that share collapsed to 27%. The consumers caught in between (people who were paying near-zero in 2025, passively renewed, and are now facing a real monthly premium for the first time) are the highest attrition risk in the entire marketplace.
We can estimate this population directly. Using each state’s 2025 ≤$10 share as a proxy for the composition of that state’s auto-renewals, approximately 2.5 million passive renewals transitioned from ≤$10 to a meaningful premium. Over half of these are concentrated in five states: Texas (392K), Florida (341K), California (291K), Georgia (239K), and South Carolina (92K). Fifty-one percent of HC.gov counties, representing half of marketplace enrollment, experienced a ≤$10 share drop of more than 15 percentage points.
For these consumers, the OEP plan selection is essentially a placeholder. They were enrolled in a near-zero-premium plan, took no action to shop or switch, and are now staring at a bill. Under the ACA’s 3-month grace period rules, a consumer who doesn’t pay their January premium has until March 31 to pay before their coverage is retroactively terminated back to January 31. Most who haven’t paid by now likely won’t pay.
What does this look like for an actual consumer? I shared this example on LinkedIn back in October. Consider a 47-year-old in Gwinnett County, Georgia, earning $25,000 who was enrolled in Centene’s lowest-cost Silver plan at $0/month net premium in 2025. If they did nothing, they were auto-renewed and now owe $227.27/month. Had they actively shopped, they could have switched to Kaiser’s lowest-cost Silver for $89.54/month, or chosen a bronze plan from Oscar at $23.41. That $137/month gap is the cost of inaction, and Gwinnett is not an outlier. In Georgia, Centene fell off the first page of Georgia Access results in counties representing over 600,000 members. Every one of those passively renewed Centene members is now paying significantly more than the cheapest available option, and many may not know it if they’ve been on a zero premium plan for a while.
What’s the attrition rate? The non-effectuation risk varies by segment. For active re-enrollees and new consumers (14.3M), I’d estimate 5-15% non-effectuation — these people actively chose their plan, but even active shoppers face sticker shock and some will bail.
For the 7.3M auto-renewals who weren’t in the ≤$10-to-paying group (call them low-risk passive renewals) the rate is higher, perhaps 10-20%, because they still didn’t actively engage even though their premiums moved substantially.
For the 1.55M at-risk passive renewals (≤$10 to a real premium), the attrition rate is likely much higher still. David Anderson has published work suggesting an overall attrition impact of roughly 15% with the presence of a zero premium plan, but that figure encompasses all consumers (active and passive, new and renewing), making it a floor for the passive-renewal-specific effect. From my own experience working with carriers, I’ve seen 30 to 45% of passively renewed members who went from $0 to a non-zero premium disenroll for non-payment within the first quarter.
Applying these segmented rates:
The competitive disruption layer makes this worse in specific markets. Half of all auto-renewed consumers were on Silver plans and in 38% of HC.gov counties (773 counties), the cheapest Silver issuer changed. States with the highest disruption scores — Indiana (72% Silver auto-renewal share, -16.5% enrollment), Ohio (52%, -19.5%), Arizona (54%, -15.6%) — are the ones where the non-effectuation wave will hit hardest, because their auto-renewal populations were overwhelmingly Silver and overwhelmingly exposed to the benchmark repricing.
The bottom line: 23.1 million plan selections is the ceiling, not the floor. Effectuated enrollment is more likely to land in the 19-20 million range, with the precise number depending on how many of those 2.5 million zero-to-nonzero passive renewals actually pay their first bill. By the time CMS publishes effectuated enrollment data (typically with a 6-month lag), the real story of this OEP will be clearer, and it will almost certainly show a steeper decline than the headline plan selection numbers suggest.
So What Now?
If you’ve made it this far… congrats. I opened this post by saying the story underneath the numbers is more interesting than “enrollment went down.” I hope I’ve shown that. But I want to close with something that doesn’t get said enough in ACA analysis: the marketplace didn’t break this year.
The SLCSP moved 30%. Issuers repositioned. Consumers (at least the ones who actively engaged) followed the math. Bronze grew substantially almost everywhere. Gold increased where Silver inverted. Brokers helped where they could. Auto-renewals absorbed the rest (maybe….). The plumbing of the individual market worked more or less as designed, which is both reassuring and alarming: reassuring because the architecture held, alarming because “as designed” includes 8.8 million people getting passively rolled into plans they didn’t choose at prices they may not have seen.
That’s the tension at the center of this data. The ACA marketplace is built on the assumption that consumers will shop. The enhanced subsidies papered over what happens when they don’t: $0 premiums meant even passive enrollees were fine. Now the subsidies have returned to pre-ARPA levels, and the gap between “enrolled” and “covered” is about to become quite visible.
Three things I’m watching
First, the March 31 grace period cliff. That’s tomorrow, as I write this; every passively renewed consumer who hasn’t paid a premium by end of month loses coverage retroactively to January 31. The scale of that cliff (hundreds of thousands of people, probably over a million) will show up in the effectuated enrollment data later this year, and it will be the first real comprehensive measure of how much damage inertia did.
Second, whether states and issuers learn the right lesson from the disruption data. The “Goldilocks” finding that moderate competitive disruption produced the best enrollment outcomes should inform how regulators think about rate review, market entry, and network adequacy. A marketplace with zero churn isn’t healthy; it’s stagnant. But a marketplace where half the Silver landscape reshuffles and 38% of consumers don’t notice isn’t healthy either. There’s a middle ground, and the states that find it will outperform.
Third, what happens to the 2027 rate cycle. Issuers are filing rates right now, and they’re doing it with incomplete information about who actually effectuated in the market. If a carrier assumed 90% retention and the real number is 75%, their 2027 risk pool just changed materially. And they can’t just use their internal data: carrier can’t assume their own effectuation rates are illustrative of the market’s. The non-effectuation wave doesn’t just affect 2026 coverage: it feeds forward into next year’s pricing, which feeds forward into next year’s SLCSP, which feeds forward into next year’s subsidy math. They also need to be re-thinking their risk adjustment assumptions; risk adjustment is not just a function of member risk scores - it’s a function of metal mix, and that mix fundamentally changed.
The marketplace is still the largest source of individual health coverage in the country. It’s not going anywhere. But anyone looking at the 23.1 million headline and concluding that the post-ARPA transition was manageable is reading the wrong number. The real number comes later. And it will be lower.
Update: Thanks to Darren Michael for pointing out a few small errors I corrected; the only meaningful one is around the disruption model - an earlier version I had started with in this analysis used 4 variables, but one of them (passive rate) was really a mediator variable and not a dependent variable so I excluded from the analysis. However, in my editing, the final version of the post left in a reference to this old variable. The analysis itself is unchanged.