The most likely outcome of the IEEPA tariff refund programme is not that importers get their money back. It is that a permanent residue of roughly $11bn to $21bn, somewhere between 7% and 13% of the approximately $166bn collected, is never claimed at all, with cumulative certified refunds plateauing in the $145bn to $155bn band by 31 December 2027. The signals point to a claim curve that has already flattened hard, and to a tail made up of small importers whose individual claims are worth less than the cost of pursuing them. This is a forecast about who never gets paid, not about how fast the payers move.
In short
- The prediction: cumulative certified IEEPA refunds likely plateau at roughly $145bn to $155bn of the approximately $166bn collected, leaving $11bn to $21bn permanently unclaimed, scored at 31 December 2027.
- Signal 1: per a 17 September 2026 trade advisory citing CBP, about $134.7bn has been accepted into CAPE and about $122bn certified and transmitted to Treasury, but the daily intake rate has fallen roughly 87% between the spring and late-summer windows.
- Signal 2: Costco’s fourth-quarter release on 24 September 2026 booked a $0.15 per diluted share non-recurring IEEPA refund benefit, confirming that the large, well-advised importers have already been paid.
- Signal 3: Phase 3, opening 6 October 2026, covers finally liquidated entries and effectively requires litigation; practitioners quote $10,000 to $15,000 for a Court of International Trade matter against an average unclaimed entry the data implies is worth around $1,200.
- The counter-signal: CBP has automated reliquidation before and contingency-fee recovery firms have an obvious incentive to harvest the tail, either of which could close the gap faster than the decay curve implies.
Why this matters now
The Supreme Court held 6-3 in Learning Resources, Inc. v. Trump on 20 February 2026 that the International Emergency Economic Powers Act does not authorise the President to impose tariffs, but it did not decide the refund question. That was left to the Court of International Trade, which ordered US Customs and Border Protection to refund the duties. According to CBP and court filings, roughly 330,000 importers paid or deposited an estimated $166bn across more than 53 million entries. The refund is therefore one of the largest single transfers back to the private sector in modern US trade administration.
What makes it analytically interesting is that the transfer is not automatic. CBP built a new module inside the Automated Commercial Environment called CAPE, the Consolidated Administration and Processing of Entries, and importers must come forward and evidence their own entries. A refund system that requires an affirmative claim always produces a shortfall. The only open questions are how large that shortfall is and where it concentrates.
Most coverage to date has tracked the headline payout figure, which is genuinely impressive and rising. The more useful question for anyone modelling 2027 retail margins or advising cross-border sellers is the shape of the curve, not its current level. A claim-based programme does not converge on its ceiling; it converges on the point where the marginal claim costs more than it returns. That point appears to be arriving.
Scale is worth pausing on. The estimated $166bn pool is larger than the annual revenue of all but a handful of US retailers, and it was collected from a population dominated numerically by small importers rather than by the household names. That is the reverse of how tariff burden is usually discussed, and it is why the refund unwind is a small-business story at least as much as a big-box one.
Signal 1: CBP’s own throughput data shows the curve flattening
A trade advisory published on 17 September 2026, citing CBP figures, reported that more than 286,000 CAPE declarations had been submitted, with more than 201,000 accepted for processing, covering approximately 27.2 million entries. Roughly $134.7bn in potential and certified refunds had been accepted into CAPE, and approximately $122bn including statutory interest had been certified and transmitted to the US Treasury for payment. A further $1.3bn across 20,184 individual refunds was reported on hold purely because importers had not supplied valid ACH banking details.
Set that against an earlier reading. As of 10 July 2026, roughly $121.75bn had been accepted for processing and about $86.3bn repaid including interest. CAPE Phase 1 opened on 20 April 2026. The two windows therefore run 81 days and 69 days respectively.
The arithmetic is the signal. From launch to 10 July, accepted dollars ran at roughly $1.50bn per day. From 10 July to 17 September, the increment was about $12.95bn, or roughly $188m per day. That is a decline of approximately 87% in the average daily intake rate, and it happened while the programme was fully operational and widely publicised.
The composition is more telling than the rate. Accepted dollars now represent about 81% of the estimated $166bn pool, but the 27.2 million entries covered represent only about 51% of the 53 million-plus entries involved. Dividing through, the average accepted entry carries roughly $4,950 in IEEPA duty, while the average entry not yet covered implies roughly $1,210. The claimed population and the unclaimed population are not the same kind of importer, and that asymmetry is the whole forecast.
One further detail deserves attention. The gap between 286,000 declarations submitted and 201,000 accepted implies roughly 85,000 submissions that failed validation and have not been successfully refiled. CBP has already set Phase 3 for October 6 to reach the oldest and hardest entries. That a third of submissions still fail on a mature system suggests the remaining filers are less well resourced than the ones already through.
| As-of date | Declarations submitted | Declarations accepted | Dollars accepted into CAPE | Implied daily intake |
|---|---|---|---|---|
| 20 April 2026 (Phase 1 launch) | 0 | 0 | $0 | n/a |
| 11 May 2026 | 126,237 | 86,874 passed validation | ~$35.5bn anticipated on liquidated entries | not comparable (different measure) |
| 10 July 2026 | not disclosed | not disclosed | ~$121.75bn (~$86.3bn repaid) | ~$1.50bn per day since launch |
| 17 September 2026 | 286,000+ | 201,000+ | ~$134.7bn (~$122bn certified) | ~$188m per day since 10 July |
A caution on that table: the 11 May figure measures anticipated refund and interest on entries already liquidated or reliquidated, which is a narrower concept than dollars accepted into CAPE. It is included for context on declaration volumes, not as a directly comparable dollar reading. The clean comparison is the 10 July to 17 September window, and that window is the one that has decelerated.
Signal 2: Costco’s fourth quarter confirms who is already paid
Costco reported its fourth quarter and fiscal 2026 results on 24 September 2026, for the period ended 30 August 2026. Net sales reached $93.9bn, up 11.2%, with diluted earnings per share of $6.75 against $5.87 a year earlier. Digitally enabled sales rose 19.5% in the quarter and 20.9% across the 52 weeks. Fiscal-year additions to property and equipment came in at $6.435bn.
The line that matters here is smaller and easy to miss. The company disclosed a non-recurring benefit of $0.15 per diluted share from IEEPA tariff refunds received in the quarter, less partial reinvestment of those refunds in increased member values. On roughly 445 million diluted shares that is on the order of $67m net of reinvestment, against a gross refund reported elsewhere at a considerably higher figure. Costco separately routed part of a $184m tariff refund into produce and meat prices.
Treat this as a timing signal rather than an earnings story. A refund large enough to move consolidated EPS by fifteen cents arrived, was processed, and was recognised inside a quarter that closed on 30 August 2026. That is what the front of the distribution looks like: sophisticated in-house trade compliance, clean importer-of-record records, and the scale to make the filing effort trivially worthwhile.
Extend the logic. If the largest and best-advised importers cleared their claims by late August, the $134.7bn already accepted is disproportionately theirs, and what remains is by construction the harder, smaller, worse-documented residue. The signals point to a programme that has already harvested its easy dollars. The deceleration in Signal 1 and the composition in Signal 2 are the same fact observed from two directions.
Signal 3: Phase 3 turns a filing into a lawsuit
Phase 3 deploys on 6 October 2026 and addresses the entries the earlier phases could not reach, principally those that have already reached final liquidation. Once an entry is finally liquidated, CBP cannot simply reopen it through the ordinary administrative path. Phase 3 therefore covers importers who have filed Court of International Trade actions and obtained court orders directing reliquidation.
That is a materially different product from Phase 1. Trade practitioners describe Phase 3 as considerably less friendly to self-service and substantially more litigious, requiring counsel or an experienced broker to navigate entry status and procedure. Attorneys are quoted at roughly $10,000 to $15,000 for certain Court of International Trade matters. Against an implied average unclaimed entry of roughly $1,210, the economics only work for importers who can aggregate many entries into one action.
Two hard deadlines compound the problem. The statutory protest window runs 180 days from the date of liquidation and cannot be extended; miss it and the right to that entry’s refund is permanently lost. Separately, per CBP’s own status filing to the court, Phase 3 covers finally liquidated entries filed by plaintiffs for which the Court of International Trade has ordered reliquidation, and only plaintiffs that supplied a valid importer-of-record number by 30 July 2026 can file from 6 October. The non-litigant importer holding finally liquidated entries is left outside the mechanism altogether, dependent on the government’s appeal.
The importer-of-record condition is where this gets structurally awkward for e-commerce. Many small sellers imported under a customs broker’s or a freight forwarder’s arrangements and may not hold a clean, current IOR record of their own. CBP has also begun voiding dormant importer numbers, which removes precisely the credential Phase 3 requires. An importer who stopped importing in 2025, which describes a great many sellers squeezed out by the tariffs themselves, is the single likeliest person to be owed money and unable to collect it.
Who is likely in the tail
The profile that emerges from the arithmetic is reasonably specific. It is an importer with a modest number of entries, an average duty per entry in the low four figures, no in-house customs function, and a relationship with a broker that ended when the importing did. Nothing about that profile is exotic; it describes a large share of the marketplace sellers who built businesses on cross-border sourcing between 2023 and 2025.
Two secondary groups sit alongside them. The first is importers whose entries were subject to protests filed on other grounds, which puts them on a different procedural track with its own eligibility rules and deadlines. The second is the 20,184 refunds, worth about $1.3bn, reported as held purely for missing ACH banking details, a category that signals disengagement rather than ineligibility.
That last group is the cleanest available proxy for filer attention. These importers did the hard part, filed successfully and were approved, and then did not complete a banking form. If a meaningful share of that $1.3bn is still sitting unpaid in mid-2027, it is strong evidence that the residual population is not simply waiting for a better process.
| Signal | Date | Source type | What it measures | What it implies for the tail |
|---|---|---|---|---|
| CAPE throughput and certification | 17 Sep 2026 | Trade advisory citing CBP operational data | Dollars accepted, dollars certified, declarations, entries | Intake rate down ~87%; 81% of dollars but only 51% of entries covered |
| Costco fiscal Q4 disclosure | 24 Sep 2026 | Company results release and 8-K exhibit | Refund recognised in a quarter ended 30 Aug 2026 | Large, well-advised importers are already through the queue |
| Phase 3 design and deadlines | 17 Sep to 6 Oct 2026 | Trade advisory and practitioner commentary | Eligibility rules, litigation cost, statutory windows | Marginal claim cost exceeds marginal claim value for small filers |
What the pattern suggests
Run the decay forward mechanically and the result is stark. If each successive window adds the same fraction of the previous window’s increment, the ratio observed (roughly $12.95bn following roughly $121.75bn) implies remaining intake of under $2bn and a terminal figure near $136bn. That would leave close to $30bn, or about 18% of the pool, unclaimed. A pure two-point geometric extrapolation is fragile, and it should not be treated as the base case.
Two forces likely push the terminal figure above that floor. Phase 3 is a step function rather than a continuation of the curve, unlocking a tranche of finally liquidated entries that the earlier phases structurally could not touch. And a recovery industry is forming, in which brokers and contingency-fee firms aggregate small claims and take a percentage, which changes the unit economics for exactly the importers the decay curve writes off.
Balancing the mechanical floor against those two step-ups gives the base case: cumulative certified refunds of roughly $145bn to $155bn by 31 December 2027, with $11bn to $21bn permanently unclaimed. Assign that roughly 55% confidence. The distribution is asymmetric, because the ways this goes better are discrete and identifiable while the ways it goes worse are simply the continuation of an observed trend.
The companion claim is more diagnostic and easier to check. Measured in importers rather than dollars, the shortfall is likely far larger: accepted declarations plausibly end 2027 below 270,000 against roughly 330,000 importers who paid. Because a single importer can file multiple declarations, the true share of importers covered is likely lower than a naive declaration count suggests. Expect a programme that returns most of the money to a minority of the claimants.
It is worth being explicit about why the distribution skews this way. Every mechanism identified above removes claimants permanently rather than delaying them: a lapsed importer-of-record number, an expired 180-day protest window, a missed 30 July 2026 cutoff. None of those reverse with time or patience, which is what distinguishes this from an ordinary processing backlog.
The optimistic paths, by contrast, all require someone to act deliberately and at scale. CBP would have to extend automation into a category it has so far handled case by case, a court would have to order class-wide reliquidation, or a recovery industry would have to industrialise inside roughly fifteen months. Each is plausible on its own; none is the default. A forecast should weight the trend it can observe more heavily than the interventions it can only imagine.
| Scenario | Probability | Terminal certified refunds | Permanently unclaimed | What you would see first |
|---|---|---|---|---|
| Base case: tail persists | ~55% | $145bn–155bn | $11bn–21bn (7–13%) | Phase 3 adds a visible step in Q4 2026, then intake resumes decaying |
| Tail closes: automation plus intermediaries | ~20% | above $158bn | under $8bn (under 5%) | CBP announces automated reliquidation for a class of entries |
| Decay dominates | ~20% | $136bn–145bn | $21bn–30bn (13–18%) | Phase 3 step is small; monthly CBP updates go quiet |
| Pool restated or unscoreable | ~5% | not comparable | not measurable | Interest accrual restates the denominator or CBP stops publishing |
Wider context: the Section 122 pool repeats the same test
The IEEPA refund is not the only unwind in flight. Hours after the Supreme Court ruling on 20 February 2026, the administration invoked Section 122 of the Trade Act of 1974 and imposed a 10% global import surcharge effective 24 February 2026. Section 122 permits temporary surcharges of up to 15% for no more than 150 days absent an Act of Congress. That surcharge expired at 12:01am EDT on 24 July 2026, precisely at the statutory maximum.
The Court of International Trade had already invalidated the Section 122 surcharge in May 2026, with the ruling’s ultimate fate uncertain on appeal. A replacement took effect at the same minute the surcharge lapsed: Section 301 duties of 10% to 12.5% on the products of roughly 60 economies, covering on the order of 99.4% of US imports. The tariff burden did not go away, but a second potential refund pool was created.
That second pool is the natural test of this article’s thesis. It covers a five-month window rather than roughly two years, so per-importer amounts are considerably smaller, while the fixed cost of filing, proving entries and, where necessary, litigating is essentially unchanged. If the tail thesis is right, the Section 122 pool should show a proportionally worse claim rate than IEEPA. If claim rates come in similar, the marginal-cost story is weaker than argued here.
Downstream, the argument over where recovered money belongs is already live. A class action against Target has put a $994m tariff refund windfall at stake, and a multidistrict panel has been asked to consolidate related suits involving Amazon and Costco. The Target tariff refund class action frames the question that unclaimed dollars never even reach: whether a refunded surcharge belongs to the importer or to the consumer who paid it at checkout.
Implications for retailers, sellers and platforms
For large retailers, the refund is now largely a completed balance-sheet event and an earnings-comparison problem. Analysts modelling fiscal 2027 need to strip non-recurring refund benefits out of the base, or they will read a clean margin comparison as deterioration. Costco’s explicit flagging of the item as non-recurring is the template; not every filer has been as clear.
The allocation question is separate and unresolved. The prevailing expectation has been that the windfall is likelier to end in buybacks than in price cuts, and the partial reinvestment Costco disclosed is a partial counter-example rather than a refutation. Retailers that visibly route refunds to shelf prices likely buy themselves a better position in the litigation that follows.
For marketplace sellers and small cross-border importers, the operative advice is narrow and time-sensitive. Establish whether the business is a plaintiff in the trade-court litigation and whether a valid importer-of-record number was supplied by 30 July 2026, because those conditions gate Phase 3 and cannot be satisfied retroactively. Check liquidation dates against the 180-day protest window, since that clock is statutory and runs regardless of whether anyone is watching it.
For platforms and logistics providers, there is a service opportunity in the tail that the arithmetic makes obvious. Aggregating thousands of small claims turns an uneconomic individual filing into an economic portfolio action, which is precisely the mechanism that would push the outcome toward the optimistic scenario. Whether that industry scales fast enough is the single biggest swing factor in this forecast.
For investors, the practical consequence is a comparability problem running through fiscal 2027 reporting. Refund benefits have landed in different quarters, been described in different language, and in at least one case been partly reinvested in prices rather than dropped to the bottom line. Screening on year-over-year margin change without normalising for these items likely produces false signals in both directions across the retail and consumer sectors.
Caveats: what could go wrong
The strongest objection is that CBP has repeatedly automated its way around friction. The agency already reliquidates certain accepted entries without a separate claim and extended CAPE to reconciliation-flagged entries from 29 June 2026. A broader automated sweep, or a Court of International Trade order directing class-wide reliquidation without individual filings, would collapse the tail quickly and falsify the central claim.
The second objection is commercial. Contingency-fee recovery firms typically take 10% to 25% and are well suited to harvesting exactly the small, numerous claims this analysis writes off. If that industry industrialises through 2027, the terminal figure rises toward the optimistic scenario even though the underlying per-claim economics are unchanged.
Third, the denominator is not stable. Statutory interest continues to accrue on unpaid amounts, which inflates the certified total without changing the number of importers made whole. The approximately $166bn figure comes from CBP and court filings and could be restated, and legislative or judicial action could extend deadlines that this analysis treats as fixed.
Fourth, there is a timing-versus-terminal confusion to guard against. A government funding lapse or a CBP resourcing constraint would slow certification without changing how much is eventually claimed, and a slowdown of that kind could be misread as confirmation of the thesis. The prediction here is about the terminal level, not the pace.
Fifth, and most practically, the forecast may simply become unscoreable. CBP is under no obligation to keep publishing CAPE throughput statistics, and reporting has so far come largely through trade advisories rather than a standing public dashboard. If that reporting stops, the honest answer in December 2027 will be that the question cannot be settled from public data.
How to score this prediction
The primary claim resolves on one number: cumulative certified IEEPA refunds, including statutory interest, as reported by CBP or a trade advisory citing CBP, at or near 31 December 2027. A reading between $145bn and $155bn confirms the base case. Below $145bn confirms the decay scenario; above $158bn falsifies the thesis in favour of the tail closing.
The companion claim resolves on accepted CAPE declarations, with a threshold of 270,000 at end-2027 against the roughly 330,000 importers who paid. The Section 122 claim resolves on whether that pool’s claim rate, measured as claimed dollars over collected dollars, comes in below the IEEPA rate at a comparable point after its own refund mechanism opens.
Three interim checkpoints are worth watching. The size of the step that Phase 3 produces in the fortnight after 6 October 2026 is the first and most informative. Whether the $1.3bn held for missing ACH details resolves or persists into 2027 is the second, since it is a pure proxy for filer engagement. The emergence of a named, scaled tariff-recovery intermediary is the third.
Readers who want to track the official mechanism directly can follow CBP’s own programme page for IEEPA duty refunds, which carries the phase schedule and filing requirements.
Frequently asked questions
What exactly is being predicted here?
That cumulative certified IEEPA tariff refunds plateau at roughly $145bn to $155bn of the approximately $166bn collected, leaving $11bn to $21bn permanently unclaimed as of 31 December 2027. The companion claim is that accepted CAPE declarations end 2027 below 270,000 against roughly 330,000 importers who paid. Both are checkable against CBP’s published throughput figures.
If $122bn has already been certified, is the story not basically over?
For the largest importers, yes. The forecast is about the remaining tail, which represents roughly 49% of affected entries but only about 19% of the dollars. That composition is what makes the last stretch structurally harder than the first.
Why would anyone leave a tariff refund unclaimed?
Because the marginal claim can cost more than it returns. The data implies an average unclaimed entry carries roughly $1,210 in duty, while Court of International Trade matters are quoted at roughly $10,000 to $15,000. Add a lapsed importer-of-record number or a missed 180-day protest window and the claim becomes impossible rather than merely uneconomic.
What is CAPE Phase 3 and why does 6 October 2026 matter?
Phase 3 is the stage that addresses finally liquidated entries, which earlier phases could not reopen administratively. It opens on 6 October 2026 and, per CBP’s status filing, covers finally liquidated entries filed by plaintiffs for which the Court of International Trade has ordered reliquidation, limited to those that supplied a valid importer-of-record number by 30 July 2026. Importers that never sued are not covered and remain dependent on the government’s appeal.
What is the strongest argument that this prediction is wrong?
That CBP automates the remainder. The agency has already extended automatic handling to certain entry categories, and a class-wide reliquidation order would close the tail without individual filings. A rapidly scaling contingency-fee recovery industry would have a similar effect through a different mechanism.
Does the Section 122 surcharge create a second refund pool?
Potentially. The 10% global surcharge ran from 24 February 2026 until it lapsed at the 150-day statutory maximum on 24 July 2026, and the Court of International Trade invalidated it in May 2026 with the outcome on appeal still open. Because the window was short, per-importer amounts are smaller against an unchanged fixed filing cost, which is why the tail effect there should be proportionally worse if this analysis is right.
Do consumers get any of this money back?
That is being litigated rather than administered. A class action against Target has put a $994m windfall at stake and a multidistrict panel has been asked to consolidate related suits, but no general consumer restitution mechanism exists. Unclaimed refunds never enter that argument at all, because the money stays with the Treasury.
What should a small cross-border seller do this month?
Confirm whether the business is a plaintiff in the trade-court proceedings and supplied a valid importer-of-record number by 30 July 2026, since those conditions gate Phase 3 and cannot be fixed retroactively. Pull liquidation dates for affected entries and measure them against the non-extendable 180-day protest window. If the aggregate claim is small, the realistic route is a broker or recovery firm that can pool it with others.
How confident is this forecast?
Roughly 55% on the base case band, with about 20% on a materially worse outcome and 20% on the tail closing. The residual 5% covers the possibility that the pool is restated or CBP stops publishing the statistics, in which case the prediction becomes unscoreable rather than right or wrong. All figures cited are as reported by CBP, trade advisories and company filings, and are subject to revision.