Insuring the Clawback: Why RADV Extrapolation Turned Medicare Advantage Risk Adjustment Into an Unreserved Balance Sheet Liability and the Case for Building a Regulatory Recoupment Insurance Market
🎧 Podcast episode for paid subscribers only. Also available on Spotify.
To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast).
Abstract
The setup: risk adjustment as a contingent liability nobody reserves for
RADV in one sitting: sampling, extrapolation, and the dead FFS adjuster
The 2025 escalation: audit everything, all at once
Sizing the exposure: what the clawback math actually looks like
Why this risk is uninsured today
The model: regulatory recoupment insurance, underwritten at the chart level
Precedents that say this works: tax insurance, R&W, and litigation finance
Who buys, who sells, and how the market structures itself
Failure modes and the honest stress test
The endgame: the insurer as private ratings agency for coding risk
MA plans and delegated risk entities carry a large, unquantified regulatory clawback exposure from RADV audits, which since the 2023 final rule can extrapolate sample error rates across entire contracts without a fee for service adjuster
CMS moved in 2025 from auditing roughly 60 contracts per year to all eligible contracts annually, with a plan to clear the PY2018 through PY2024 backlog and a coder workforce expanded from about 40 to approximately 2,000 plus automation
Almost no plan books a reserve for this because the exposure is deemed not reasonably estimable under ASC 450, which is accounting speak for “we have no idea and would rather not guess in public”
The essay proposes a new business: regulatory recoupment insurance, a specialty line that underwrites RADV and related audit clawback exposure at the chart and HCC level, priced off proprietary shadow audits
Direct analogs exist and work: tax liability insurance, reps and warranties insurance, and litigation finance all converted low frequency, high severity legal outcomes into priced, transferable risk
Buyers include MA plans, delegated provider groups, PE acquirers of value based care assets, and lenders; sellers start as an MGA with a fronting carrier, reinsurance, and eventually ILS for tail risk
The durable moat is not capital, it is loss data; the underwriter that sees thousands of shadow audits becomes the de facto private ratings agency for coding integrity, which is worth more than the premium
The setup: risk adjustment as a contingent liability nobody reserves for
Start with a question that sounds naive and is not. If a Medicare Advantage plan submitted diagnosis codes years ago, got paid on them, and the government has now announced its intention to audit every contract every year and extrapolate any error rate it finds across the entire payment base, what is that liability worth, and where does it sit on the balance sheet?
The answer, at basically every publicly traded plan and every delegated risk entity in the country, is that it sits nowhere. Read the 10-Ks. The RADV discussion lives in the risk factors and legal proceedings sections, phrased as some version of “we are unable to estimate the outcome or range of loss.” No accrual, no reserve, no dedicated capital. The exposure is real enough that Humana sued the federal government over the audit methodology, real enough that CMS itself has projected billions in recoveries, and real enough that the DOJ has spent years litigating chart review practices at the largest carrier in the sector. But it is not booked. It floats.
That gap between economic reality and accounting treatment is where new financial products come from. Every specialty insurance line in history started as an exposure that was obviously real, obviously large, and conveniently unquantified, right up until someone built the machinery to quantify it and then sold protection against it at a margin. Directors and officers liability worked this way. Environmental impairment worked this way. Tax insurance, more recently and more instructively, worked exactly this way. RADV recoupment risk is sitting at that same starting line, and the audience for this argument, people who actually read risk adjustment methodology for fun, already knows the exposure is not hypothetical.
