MLR Rebates Fall to $759 Million as Insurers Tighten Toward the Margin Floor

KFF's July 2026 tally puts this year's Affordable Care Act rebate at $759.2 million across the individual, small-group and large-group markets, down from $1.6 billion in 2025 and $958 million in 2024, the smallest payout since 2018's $706.7 million (KFF, July 2026). That decline is not a consumer refund story. It is evidence that carriers priced closer to the 80/85 percent medical loss ratio floor than they have in years.

A shrinking rebate reads, on the surface, as good news for policyholders' premium dollars and worse news for insurer profitability. The mechanics say otherwise. Rebates exist because an insurer collected more premium than its medical loss ratio floor allows it to keep after paying claims and quality improvement costs; a smaller industry-wide rebate simply means fewer carriers cleared that threshold by a wide margin. Under the three-year rolling formula CMS uses to calculate them, this year's number is not a real-time read on 2026 pricing at all. It is a lagging indicator built from 2023, 2024 and 2025 experience, arriving just as carriers file 2027 rates that assume the opposite of what the rebate implies.

The Three-Year Formula Behind a One-Year Headline

CMS requires individual and small-group insurers to spend at least 80% of premium revenue on medical claims and quality improvement activities, and large-group insurers to spend at least 85%; any carrier that falls short in a given state and market segment owes the shortfall back as a rebate. But the calculation is never based on a single year. Insurers report their MLR annually, and CMS averages three consecutive years of that experience before determining whether a rebate is owed, which is why 2026's $759 million reflects financial data from 2023, 2024 and 2025, not 2026 itself (KFF, July 2026). The rebate checks mailed this fall go to people and employers who were covered in 2025, the most recent of the three years in the averaging window.

That three-year lag is a deliberate smoothing mechanism, not an oversight. A single bad claims year, a large-claimant spike, or a pandemic-era utilization swing would otherwise whip an insurer's rebate obligation from zero to material and back again annually, which is exactly the kind of volatility a credibility-weighted actuarial calculation is supposed to dampen. The tradeoff is that the published rebate number always describes an insurer's already-completed pricing cycle, not its current one. The rolling average pulls in 2023's post-pandemic utilization catch-up, 2024's stabilizing trend, and 2025's markedly tighter individual-market experience, and blends all three into a single figure that reads as "this year's rebate" but is functionally a three-year backward-looking average. Reading the $759 million total as a signal about how insurers are pricing today, rather than how they priced from 2023 through 2025, is the single most common misread of this release.

The historical rebate series makes the smoothing effect visible. Rebates peaked at roughly $2.46 billion in 2020 and $2.01 billion in 2021, when insurers overestimated 2020's pandemic-depressed claims and priced 2021 too rich relative to actual cost, then fell to $1.03 billion in 2022, $947 million in 2023 and $958 million in 2024 as the three-year windows absorbed that overpricing and margins normalized (healthinsurance.org, using KFF data). The 2025 spike to $1.6 billion reflected a window still carrying some of that residual excess margin. This year's $759 million is the first cycle where the rolling average has fully rotated past the CSR-era pricing distortions and into a period where insurers, in KFF's words, "reflect this period of margin normalization rather than continued insurer profitability stemming from the earlier CSR-related premium increases" (KFF, July 2026).

Where the Rebate Concentrates: the Individual Market Carries Most of It

Rebates are not distributed evenly across the three markets the MLR rule covers. In the most recent full-detail cycle, the individual market accounted for close to 72% of total rebate dollars despite covering only about 11% of the U.S. population, a concentration KFF's underlying CMS data confirms year after year (healthinsurance.org, using KFF data). Average per-person rebates in that cycle ran $233 in the individual market, $190 in the small-group market and $91 in the large-group market, on a base of 5.1 million individual-market recipients and 3.5 million people with employer coverage (KFF, July 2026). Large blocks of employer coverage, particularly self-funded and ASO arrangements that fall outside MLR entirely, dilute the group-market numbers; fully-insured individual policies carry no such exemption, which is part of why that market absorbs a disproportionate share of the total.

The market-level loss ratios explain the pattern without fully resolving it. KFF's 2025 data put the individual market's aggregate simple loss ratio, medical claims divided by premium, before quality improvement credit, at 93%, against 87% in small group and 91% in large group (KFF, July 2026). All three sit above their respective 80% and 85% floors on a market-wide average basis, which on its face argues against widespread rebate exposure. But MLR is not computed as an industry average; it is computed separately for each legal entity, in each state, in each market segment. A market where the aggregate ratio comfortably clears the floor can still contain individual carrier-state blocks that fall well short of it, and those specific blocks are what generate the rebate obligation. The 72% concentration in the individual market is consistent with that market containing more of these below-floor outliers, likely a mix of insurers that priced defensively against continued post-subsidy-cliff enrollment uncertainty and ended up running unusually rich loss ratios in specific states, even as the broader market average looks healthy.

The Quality Improvement Credit Sitting Between Simple Loss Ratio and the Floor

The gap between a market's simple loss ratio and its actual rebate exposure is not only a market-average-versus-individual-block problem; it is also a numerator problem. The MLR that CMS compares against the 80/85 floor is not the raw ratio of claims to premium. It is incurred claims plus documented quality improvement expenditures, divided by earned premium net of certain federal and state taxes and licensing fees, reported annually on each insurer's Supplemental Health Care Exhibit. Quality improvement activities, care coordination programs, wellness initiatives, health information technology investments tied to clinical outcomes, can add several percentage points to the numerator before an insurer's calculated MLR is measured against the floor at all. A carrier with a simple loss ratio of 76% in a given state block is not automatically 4 points short of an 80% floor; if it can document 3 points of qualifying QI spend, it is only 1 point short, and the rebate owed shrinks proportionally.

That distinction matters for how a pricing or reserving actuary should read the 93%/87%/91% simple loss ratios cited above. Those figures, sourced from aggregate claims data, sit above the respective floors before any QI credit is even applied, which means the markets are running with real cushion on a blended basis. But QI credit is documented and audited at the individual filer level, and it varies enormously by carrier: a large national insurer with an established care-management infrastructure can typically substantiate a meaningfully larger QI credit than a smaller regional plan building that infrastructure from scratch. An insurer's realized rebate exposure is therefore a function of three separate variables stacked on top of each other: the three-year rolling claims experience, the credibility adjustment tied to block size, and the QI expense credit tied to care-management maturity. Two carriers with identical claims experience and identical block size can still owe different rebates if one has a more developed, better-documented quality improvement program.

The rebate liability itself is not a one-time cash event handled entirely after the fact, either. Insurers are required to accrue an estimated MLR rebate liability during the plan year itself, updating the estimate each quarter as claims experience develops, which means the actuarial and finance functions inside a health insurer are effectively reserving against this obligation in real time rather than discovering it when CMS's three-year window closes. A carrier that under-accrues that liability during 2025 and 2026, because its interim claims-development assumptions ran optimistic, faces the same kind of adverse true-up dynamic reserving actuaries recognize from casualty loss development: a liability that looked adequately provided for at an early valuation date turns out to be understated once the full experience period matures.

Credibility Adjustments: Why Smaller Blocks Escape Rebates Larger Ones Cannot

A structural feature of the MLR rule, separate from the three-year average, determines who actually owes money once a shortfall is identified: the credibility adjustment. An insurer's reported claims experience is only as reliable as its underlying enrollment, and CMS's methodology recognizes that small blocks generate statistically noisy loss ratios that do not reflect true expected cost. Insurers covering fewer than 1,000 life-years in a given state and market are exempt from the rebate requirement entirely because their experience is not considered credible enough to calculate a meaningful MLR. Insurers with between 1,000 and 75,000 life-years are classified as "partially credible" and receive an upward adjustment to their reported MLR, on a sliding scale that adds anywhere from 1.2 to 8.3 percentage points depending on block size, with the smallest partially-credible blocks receiving the largest boost (KFF, Explaining Health Care Reform: Medical Loss Ratio).

The mechanical effect is straightforward: a small regional carrier and a national carrier can report the identical simple loss ratio in the same state and owe entirely different rebates. A block of 5,000 life-years running a 76% simple loss ratio against an 80% floor might receive a 6-point credibility adjustment, pushing its calculated MLR to 82% and clearing the requirement with no rebate owed. A fully-credible national carrier with tens of millions of covered lives reporting that same 76% ratio in a given state receives no such adjustment and owes the full 4-point shortfall back to policyholders. This is precisely why the smallest blocks in a market can consistently escape rebate exposure that larger, better-capitalized competitors cannot: the rule is compensating for statistical noise in small-sample experience, not for size or market power, but the practical effect skews rebate incidence toward larger, more credible books.

Block size (life-years)	Credibility status	MLR treatment
Under 1,000	Not credible	Exempt from rebate requirement
1,000 to 75,000	Partially credible	Reported MLR adjusted upward by 1.2 to 8.3 points, sliding by size
Above 75,000	Fully credible	No adjustment; reported MLR compared directly to the 80/85 floor

A Shrinking Rebate Against a Rising 2027 Filing Season

The rebate trend and the rate-filing trend are now pointing in opposite directions, and that divergence is the more useful signal for pricing actuaries than the rebate total on its own. ACA marketplace insurers pushed through an average premium increase of roughly 20% to 25.5% entering 2026, the steepest jump since 2018, and preliminary 2027 filings from 77 insurers across 16 states and the District of Columbia show a median proposed increase of 14%, with 20 insurers requesting increases above 20% (KFF/Peterson-KFF Health System Tracker, July 2026). Insurers attribute roughly four percentage points of the 2026 increase, and an expected four more points in 2027, to a sicker risk pool left behind after the enhanced premium tax credits expired at the end of 2025 and healthier enrollees exited the marketplaces, a shift that has already pushed average post-subsidy premium payments up 58% this year (KFF/Peterson-KFF Health System Tracker, July 2026).

Read together, a declining rebate and an accelerating rate filing are not contradictory; they describe the same underlying dynamic from two different points in the pricing cycle. The $759 million figure reflects 2023-2025 experience, a period when insurers were still working through post-pandemic cost catch-up and had not yet absorbed the subsidy-cliff risk-pool shift that began in earnest with 2026 enrollment. The 2027 filings reflect carriers' current-year view of a materially worse risk pool, priced defensively precisely because the last several years of thin rebates gave them little cushion to work with. A market running close to its MLR floor, evidenced by a rebate total near a multi-year low, has less room to absorb a fresh adverse-selection shock without either raising rates sharply or eroding underwriting margin. The 14% median 2027 ask is largely that math working itself out in real time.

What a Shrinking Rebate Signals to a Pricing Actuary

The instinct to read a falling rebate total as evidence of benign experience gets the causality backward. A rebate is owed when premium is priced too rich relative to realized claims; a shrinking rebate total means fewer insurer-state-market combinations priced that rich, which is exactly what happens when margins compress and pricing actuaries have less cushion built into their assumptions to begin with. A market running at $759 million in rebates, against a 14% median rate ask for the following year, is not a market with room to spare. It is a market where the historical buffer between priced premium and realized cost has narrowed enough that a single adverse development, whether from risk-pool deterioration, a Medicaid-adjacent cost shock, or a slower-than-modeled provider contracting cycle, would show up as a missed rate filing rather than a rebate check.

The credibility-adjustment mechanics compound that read for actuaries pricing smaller blocks specifically. A partially-credible block that has avoided rebates because of a favorable credibility adjustment, rather than because its underlying claims experience genuinely cleared the floor, is carrying a form of statistical forgiveness that a growing block will eventually lose. Pricing teams building 2027 assumptions off a small block's clean multi-year rebate history should isolate how much of that clean record was earned economics versus credibility-table arithmetic before extending the same assumption into a period of larger expected enrollment, when the same experience would no longer receive the adjustment.