MarketAxess Holdings Inc. (NASDAQ: MKTX) — An Amortizing Annuity, Priced to Amortize Faster Than It Ever Has

Independent research note. Report date: 2026-07-17. Fiscal year ends December 31; all figures USD unless noted. Price $115.50 (close 2026-07-16).

Standing disclaimer: The analysis in sections 1–15 below is deliberately position-free — it contains no buy/sell recommendation and no price target, and discusses valuation only as embedded expectations and scenarios. The sole exception is the clearly-labeled Claude’s Take block immediately below, which is the author’s own subjective opinion.

⚡ Claude’s Take

This is the author’s own independent, subjective opinion and general information only. It is not investment advice. The analysis in sections 1–15 below takes no position and carries no price target.

Verdict: AVOID-here / HOLD — a real franchise decaying at an unknown rate, not yet cheap enough to underwrite the unknown. Not a short. MarketAxess is not a fraud, not a melting ice cube, and not a bad business: it still earns an ~18% ROIC on total capital (~35% on operating capital) against an ~8.5% WACC, converts 119% of net income into owner free cash flow, carries net cash, and dilutes shareholders by only ~0.25% a year. Volumes have never declined. But it has produced zero EPS growth in six years ($7.85 in 2020 → ~$7.75 normalized today) because it is monetizing every unit it trades ~25% worse than it did in 2021, and nothing in the evidence dates the end of that decay. At $115.50 the market embeds credit fee-per-million falling ~8%/year in perpetuity — worse than the −6.9%/yr MarketAxess has actually delivered, and worse than Q1-2026’s −5.0%. That is the bull case, and it is a real one. It is not enough for me here, because the base case only discounts at ~9.5% against an ~8.5% WACC — roughly a point of compensation for a business whose central variable has no visible floor, whose management is paid on a metric that ignores it, and whose insiders will not buy it. My accumulation zone is ~$85–95 (≈11–12x normalized EPS of ~$7.75, an owner-FCF yield approaching 9%) — a margin of safety wide enough to survive a bear case in which operating margin falls to ~29%. I’d want that price or a fee-per-million stabilization print; at $115.50 I have neither.

What the market is pricing correctly, and why I can’t call this a mispricing. The de-rating is earned, and the sharpest fact in this report is not the share loss — it is that the moat is real but attached to the wrong protocol. Open Trading, the genuine all-to-all liquidity network, is shrinking in absolute dollars ($178.5M → $175.6M) while credit volume grew ~25%; the protocols winning — portfolio trading, blocks, dealer matching — are ones where a 1,800-counterparty anonymous network confers no advantage at all, because a portfolio trade is a bilateral balance-sheet transaction needing one dealer who can warehouse the basket, not a crowd. MarketAxess is not being disrupted; it is being routed around, and it is helping — its own three headline initiatives are, by management’s admission, the ones that “come in at that lower” price point. It is buying volume with price. Meanwhile capital allocation actively destroyed value: $646M of buybacks at a blended $206.02 is worth $362M today — $283.9M incinerated, ~$8.07 per current share — funded latterly by levering a never-drawn balance sheet, and executed by a management paid on adjusted operating income with no EPS, no ROIC, no ROE and no TSR metric anywhere in the plan (the CEO’s bonus was $1,525,000 in both 2024 and 2025 — identical to the dollar — while missing target both years). The framing is falling knife, not abandoned value: MKTX carries a positive Value loading into a Value factor that returned +13.8% over the past year and still lost 45.4%; ~86% of its variance is idiosyncratic; there is no hostile regime to wait out. The tape rejects good news — the December targets/$505M authorization and the Q4 print each popped and fully round-tripped within weeks. And the CEO who spent ~$1M of his own money at $238 in 2023 has not bought a share at $170, $146, $116, or the $109 low. Zero insider open-market purchases since January 2024. When the people who can see the fee-per-million report before I can won’t buy it, I want their discount, not their price.

Catchy tag: “The toll booth is real. The traffic is being routed around it.” Conviction: medium. Bull trigger (flips me constructive): two or three consecutive quarters of fee-per-million stabilizing (decay inside ~2–3%, or flat) while volume compounds ~10% — because §6.2 proves that if fee capture merely holds, operating margin returns toward 50% on scale economics that never stopped working; that plus an insider buying at these levels would move me quickly. Bear trigger (flips me negative outright): fee-per-million decay accelerates past −8% with Open Trading dollars still shrinking and high-grade share breaking below ~16% — that is the amortizing-annuity path, and at that outcome $115.50 is not defensible above a 4% discount rate, meaning today’s “cheapness” is an illusion and the price is not a floor.

📈 Stock Price Action — Five-Year Event Map

Over the trailing five years MarketAxess round-tripped from a high of ~$463 (2021-08-04) to a five-year low of $109.09 on 2026-06-25, closing $115.50 on 2026-07-16 — 79.1% below the all-time-high close of $553.43 (2020-12-22), inside a 52-week range of $109.09–$211.60. The stock has fallen in five of the last six calendar years (2021 −27%, 2022 −31%, 2023 +6%, 2024 −22%, 2025 −19%, 2026 YTD −36%), and every rally inside that span has been fully retraced.

Cycle narrative. (1–2) The five-year window opens at the top: MKTX peaked at ~$463 on 2021-08-04 and gave back 48% into mid-2022 as pandemic-era volume and fee tailwinds normalized against a multiple set for permanence (Interp). (3) The 2022–23 rate- and credit-volatility surge drove record volumes; the Q4’22 release delivered the largest single up-day of the five years, +10.2% (Fact). (4) That rally fully retraced, terminating in the largest one-day drop of the period, −17.8% on 2024-01-31 (Q4’23 release) (Fact); the market read the print as confirmation of share erosion and fee compression rather than a cyclical dip (Interp). (5–6) A 2024 recovery to ~$281 was again fully round-tripped over fourteen months, punctuated by two earnings breaks (2025-02-05 −8.9%; 2025-08-06 −10.1%) (Fact). (7) Management’s response is on the record — the 2025-12-09 8-K announced medium-term targets, raised the repurchase authorization to $505mn and committed to a $300mn ASR; the stock rose 4.9% that day and 5.5% after the Q4’25 print, and both pops were fully retraced within weeks (Fact). (8) The 2026 leg to the five-year low was a grind, not a gap: across the −36.9% slide, no single session fell more than 4.7%, and the low itself traded 1.2× average volume — orderly distribution, with no capitulation (Fact). Notably, no 8-K in the corpus explains the 2026-06-25 low — the final leg is not event-driven at the filing level (Fact).

1. Executive Summary

MarketAxess operates electronic trading platforms for fixed-income securities — principally corporate credit — connecting roughly 2,000 institutional investor and broker-dealer firms and charging a commission per million dollars of face value traded. It is not a dealer, an exchange, or an asset manager: it is a venue and a workflow layer. The architecture is excellent — 868 employees generate $846.3M of revenue (~$975k per head), at 59% gross and 40% operating margins, with net cash, ~0.25%/yr dilution, and stock-based compensation of only 3.7% of revenue. Free cash flow is real: owner FCF (after capex, capitalized software and SBC) was $293.2M in FY2025 — 119% of net income. Nothing here is broken in the way a melting ice cube is broken.

And yet the company has produced no earnings growth in six years. Diluted EPS was $7.85 in 2020 and ~$7.75 normalized today; operating income is lower than in 2020 ($374.7M → $341.8M) on 22.8% more revenue; operating margin fell monotonically from 54.4% to 40.4%; ROE fell from ~34.7% to ~19.5%. The stock is 79.1% below its 2020 high, having hit a five-year low three weeks ago. The entire decline is one variable. Credit fee-per-million has fallen for four consecutive years — $184.78 → $138.87, ~−25%, ~−6.9%/yr, accelerating to −7.6% in 2025 — and a decomposition settles the cause beyond argument: holding fee capture at 2021 levels, FY2025 operating margin would have been 50.8% — higher than 2021’s 48.25% — despite a 39.5% opex build. The entire 14-point margin decline is fee erosion; operating leverage more than paid for the investment. In FY2025 MarketAxess moved 10.0% more bonds and collected $8.6M more in credit fees.

The mechanism is the report’s central finding, and it is subtler than “losing share to Tradeweb.” MarketAxess’s genuine moat — Open Trading, the anonymous all-to-all pool where liquidity truly begets liquidity — is shrinking in absolute dollars ($178.5M → $175.6M of commissions) while credit volume grew ~25%. The protocols taking its place — portfolio trading, blocks, dealer matching sessions — are ones where a 1,800-counterparty network confers no advantage whatsoever, because a portfolio trade is a bilateral balance-sheet transaction requiring one dealer who can warehouse a basket and hedge its duration cheaply, not a crowd of anonymous counterparties. The moat is real, and it is attached to the wrong protocol. Management’s own 10-K concedes clients use these workflows “in lieu of more established trading protocols designed to generate price competition on individual bonds” and that they carry a “lower-fee structure.” Protocol substitution is economically indistinguishable from a price cut. Share loss compounds it and is genuinely MarketAxess-specific: US high-grade share fell 21.0% (2021) → 17.1% (Q1-26) and high-yield 15.2% → 12.2%, while its share of the electronic high-grade market collapsed from 60.0% to 30.7% — the electronification tailwind arrived and competitors took all of it.

Capital allocation converted a decelerating franchise into a value-destroying one. $646.2M of buybacks 2021–2025 at a blended $206.02 is worth $362.3M at $115.50 — $283.9M destroyed — latterly funded by drawing $220M on a never-before-used revolver to execute a $300M ASR at $171.84. Management is paid on adjusted operating income with no EPS, ROIC, ROE or TSR metric in the plan; the CEO’s bonus was $1,525,000 in both 2024 and 2025 — identical to the dollar — despite missing target in both. No insider has made an open-market purchase since January 2024, across the entire slide from ~$270 to $109.

What the bulls have, and it is not nothing. At $115.50 — on a rebuilt EV of $3.78bn (not the $6.49bn the vendor feed reports) and normalized EPS of ~$7.75 (not the $8.45 the screen shows) — MarketAxess trades at 14.9x normalized earnings, 8.7x EBITDA and a 7.1% owner-FCF yield, roughly 41% below the cheapest point in its entire twelve-year public multiple history. Solving the perpetuity, the price embeds ~+1.0–1.3% perpetual owner-FCF growth, which at trend volume requires fee-per-million to decay ~8%/year forever — worse than anything MarketAxess has ever delivered. Protocol mix-shift is arithmetically bounded (portfolio trading cannot exceed 100% of the book), so its contribution must terminate; and Q1-2026 delivered the first inflection in five years — revenue +11.9%, operating income +14.2%, the first YoY margin expansion in the series (+0.87pt). Returns still comfortably exceed the cost of capital.

This memo takes no position; the labeled Claude’s Take above does. The honest conclusion of the analysis is that both tails are live: if the base case holds, the market discounts these cash flows at ~9.5% against an ~8.5% WACC — cheap, but only modestly; if the bull case is right the stock is badly mispriced; and if the bear case is right — fee-per-million at −8% against +8% volume, margin to 28.7% — $115.50 is not defensible at any discount rate above 4%, and the price is not a floor. The valuation does not resolve the debate. It collapses it onto a single monthly disclosure: whether fee per million stabilizes.

2. Business Overview

What it is. MarketAxess operates electronic trading platforms for fixed-income securities, principally corporate credit. It connects approximately 2,000 institutional investor and broker-dealer firms — roughly 1,800 potential Open Trading counterparties and ~200 broker-dealers as of the FY2025 10-K — and charges a commission per million dollars of face value traded. It is a venue and a workflow layer, not a principal risk-taker in the economic sense (though it stands as matched principal in Open Trading, which matters for the balance sheet; see §6.6). [FACT — FY2025 10-K]

The first and most important fact about the model is the ratio of output to input: 868 employees (554 US, 314 international) generate $846.3M of revenue — ~$975,000 per employee. MarketAxess is a software-economics business wearing a financial-markets costume: high fixed costs, near-zero marginal cost per incremental trade, and therefore enormous operating leverage in both directions. That symmetry is the whole story of the last five years. [FACT — FY2025 10-K]

Revenue segmentation — and the absence of a cushion.

[FACT — FY2022 and FY2025 10-Ks] Within commissions, FY2025 splits into variable transaction fees $600.4M (70.9% of total revenue — Credit $542.0M, Rates $28.2M, Other $30.3M**)** and fixed distribution fees $134.2M (15.9%). The non-commission “recurring” lines total $111.6M — just 13.2% of revenue. Including fixed distribution fees, anything resembling recurring revenue is 29.0%; ~71% is a volume lottery. [FACT — FY2025 10-K]

This matters more than any other structural feature. Unlike Tradeweb, which carries a meaningful fixed-fee/subscription cushion (~⅓ of revenue), MarketAxess is ~87% levered to volume × price. There is no material subscription annuity to smooth a bad protocol mix — and it is now in one. [INTERPRETATION]

The diversification story is not in the numbers. Information services grew $39.3M (2022) → $53.2M (2025); of FY25’s +$2.7M, $0.9M was FX and $1.8M net new contracts — ~+3.6% organic. Post-trade grew $36.9M → $44.5M; of FY25’s +$2.0M, $1.4M was FX and only $0.6M net new — ~+1.4% organic. Technology services jumped $3.0M → $13.9M on the Pragma acquisition — bought, not built. These lines are real but small and compounding at 2–4% organically ex-FX. They cannot offset credit fee erosion, and they are not a diversification engine. [FACT — FY2025 10-K MD&A] This raises a question the company has never answered: if the Composite+ pricing data is genuinely differentiated, why does it compound at 3%? [OPEN QUESTION]

Products, with the load-bearing detail.

• U.S. high-grade (HG) — the historic core and largest credit product. FY2025 market ADV $39.0bn.
• U.S. high-yield (HY) — FY2025 market ADV $12.2bn.
• Emerging markets (EM) — hard- and local-currency debt in ~30 currencies with ~200 broker-dealers.
• Eurobonds — European credit.
• Municipal bonds — FY2025 market ADV $10.5bn, +45.5% YoY — the fastest-growing addressable pool disclosed.
• U.S. government bonds / rates — FY2025 market ADV $1,044.3bn; MarketAxess share 2.4%. A rounding error, and load-bearing for §4.
• Other — equities/FX algorithms (Pragma, 2023) and ETF/derivatives RFQ (RFQ-hub majority stake, 2025).

A genuine and underappreciated positive: EM + Eurobonds together were 40.5% of total credit ADV in 2025, up from 32.7% in 2020 — real diversification away from the contested US high-grade core, and the segment where revenue actually grew ~20% with record commissions. [FACT — FY2025 10-K]

Protocols — the actual unit of competition. The single most important analytical move in this report is to stop thinking of MarketAxess as competing for products and start seeing it compete for protocols. A corporate bond can be traded five materially different ways, each with different economics, each favoring a different kind of venue:

1. Disclosed RFQ — client pings known dealers for competing quotes. >60% of MKTX credit volume in 2025. Highest fee. Favors the venue with the widest dealer connectivity.
2. Open Trading (all-to-all) — MarketAxess’s crown jewel. Anonymous; any participant can price any other’s inquiry; MarketAxess stands in the middle as matched principal. High fee. The only protocol with a genuine liquidity externality.
3. Portfolio trading (PT) — a basket of up to ~2,100 bonds traded all-or-none at one aggregate price with one dealer. Explicitly lower-fee. Favors the dealer with the biggest balance sheet and cheapest duration hedge.
4. Block trading — large single tickets. ~⅓ of MKTX credit ADV as of early 2026.
5. Dealer-initiated / matching sessions (Mid-X, MIDEX) — dealers cross risk at a mid-point derived from Composite+. Lower-fee.

The 10-K states the consequence in plain language: “portfolio trading is generally provided under a lower-fee structure than other protocols and the growth of portfolio trading on our platform will likely have a negative impact on our average credit variable transaction fee per million.” And: “Our dealer clients have also increased their usage of matching sessions offered by competing platforms.” [FACT — FY2025 10-K, Risk Factors and MD&A; repeated verbatim in the Q1-2026 10-Q]

Note also a mechanical wrinkle: certain US high-grade fee plans are denominated in basis points of yield, making fee-per-million sensitive to bond duration — a genuinely exogenous, cyclical driver management has leaned on heavily as an explanation. §6 shows that alibi has now expired. [FACT — FY2025 10-K]

The economics: a monotonic five-year decline.

[FACT — 10-K MD&A FY2021–FY2025; ROE rebuilt from filings, see §6.4. FPM figures disclosed directly in each 10-K MD&A.]

Every line falls in every year. Credit fee-per-million has fallen four consecutive years — −9.6%, −5.0%, −5.3%, −7.6% — cumulatively ~−25%, and the decline accelerated in 2025 to its worst rate since 2022. It fell a further −5.0% YoY in Q1-2026 to ~$132. [FACT]

The stated cause migrated, and the migration is the tell. In FY2022–FY2023 management attributed the decline to bond duration and dealer migration to fixed distribution fees — genuinely exogenous rate-cycle effects. By FY2024 the language became “product and protocol mix-shift reflecting lower levels of U.S. high-yield activity and increased portfolio trading.” By FY2025 it is unambiguous: “mainly due to protocol mix-shift reflecting increased portfolio trading.” The explanation moved from “the rate cycle did this to us” to “our own mix is degrading.” One reverses; the other does not. [FACT for the quotes; INTERPRETATION for the reading]

FY2025 crystallizes the entire model in one line: credit trading volume +10.0%, credit variable transaction fees +$8.6 million. MarketAxess moved a tenth more bonds and got almost nothing for it. Total revenue grew +3.6%. [FACT — FY2025 10-K MD&A]

Verdict — §2. A genuinely high-quality business model being run at a deteriorating price. The architecture is excellent and nothing is broken in the melting-ice-cube sense — volumes are growing, not shrinking, revenue still compounds, and the balance sheet carries no risk of consequence. But the model has one structural vulnerability the last five years have exposed completely: with ~87% of revenue as volume × price and no subscription cushion, MarketAxess has no defense against an adverse protocol mix shift. It wins more volume every year and monetizes each unit ~25% worse than in 2021 — which is why six years of volume growth produced zero EPS growth. With fee-per-million falling ~6–8%/yr, volume must grow ~10%+ just to stand still: a treadmill that accelerates as the mix worsens. The business-quality question is therefore not “is this a good business?” — it is — but “at what level do the returns stop falling?” §4 argues the answer is not yet visible.

3. Industry Dynamics

Structure. The electronic fixed-income trading industry is a small oligopoly of venues sitting between dealers (the incumbents being disintermediated) and the buy-side: MarketAxess (credit-anchored), Tradeweb (rates-anchored, pushing hard into credit), Bloomberg (the electronic RFQ rail bundled into the Terminal), ICE (BondPoint/TMC), Trumid (venture-backed credit challenger), plus dealer-backed consortia and interdealer brokers (TP ICAP). The profit pool sits with the venues on a per-million toll; the dealers retain the balance-sheet-intensive risk business. [FACT — FY2025 10-K, Competition]

Market size and the electronification runway — real in volume, largely illusory in profit. The industry’s growth engine is the conversion of voice trading to electronic. MarketAxess’s own disclosure puts electronic penetration of US high-grade at 60% in FY2025, up from 35% four years earlier — a genuine, powerful secular wave. High-yield is disclosed at ~30%; emerging markets and municipals are far less penetrated (munis’ addressable ADV grew +45.5% YoY). [FACT — FY2021–FY2025 10-Ks]

But the arithmetic that governs the industry is brutal, and it is the most important thing in this section. Full electronification from here implies roughly ~2.0x volume. Trend fee-per-million decay implies roughly ~0.56x price. The product is ~+1.5%/yr revenue. [INTERPRETATION — derived] This is not a forecast; it describes the realized record: MarketAxess compounded revenue at ~4.2% post-2020 with falling earnings. And the pool is small: the entire US high-grade + high-yield fee pool at 100% electronification is only ~$1.6–1.8bn. [INTERPRETATION — derived] There is no volume outcome that rescues a falling rate. The “long runway” management sells is real in volume and largely illusory in profit.

Competitive intensity is rising, and fee compression is industry-wide — which exonerates the industry, not the company. The Tradeweb benchmark settles it: TW’s Q4-2025 fee capture fell in every single asset class (blended −10.2%), with cash-credit fee-per-million down −14.3% YoY — a faster decline than MarketAxess’s −7.6%. (Caveat with teeth: TW and MKTX category definitions differ, so the trend is robust but the levels are indicative only and should not be compared directly.) [FACT for both series; INTERPRETATION for the comparison] Nobody is winning on price; the price is simply falling. But share loss is company-specific: Tradeweb is gaining (record 22% share) while MarketAxess is losing on both high-grade and high-yield. MarketAxess is losing on two of the three terms in Revenue = volume × share × fee-per-million, rescued only by cyclical market volume (TRACE +8.0% in FY25). Anyone saying “it’s just the industry” is exactly half right. [INTERPRETATION]

The moat mechanism that the bull case rests on is now obsolete — this is the section’s key structural insight. The classic argument for why bond venues are winner-take-most: fixed income has one less layer of abstraction than equities. There is no Reg NMS / NBBO analog mandating a best-execution sweep across venues, so the platform is the exchange, liquidity concentrates, and share should compound. That claim has been overtaken by events — the abstraction layer arrived anyway, from vendors rather than regulators. MarketAxess’s own 10-K names “EMS and OMS Providers… offer aggregation of trading venue liquidity” as a competitive category. [FACT — FY2025 10-K, Competition] EMS/OMS aggregation does privately what Reg NMS did publicly: it collapses single-venue captivity to near zero. Per Greenwald, scale is only a barrier when combined with captivity — strip the captivity and entrants reach incumbent scale. Trumid did exactly that. (Corollary worth noting: a bond best-execution mandate, often cited as a tail risk, would be an anticlimax — the moat eroded without one.) [INTERPRETATION]

The Bloomberg ceiling. Bloomberg is the electronic RFQ rail nobody models and everybody uses, bundled into a ~$30k/yr Terminal seat. Its venue exists to defend that subscription, not to earn a per-million toll — which means it can price at or near zero indefinitely. This sets a fee ceiling neither pure-play can contest, and Bloomberg will never exit. [INTERPRETATION]

The Marathon capital-cycle read — textbook, mid-bust, no recovery signal. MarketAxess earned a 54.4% operating margin and ~30% ROIC in 2020. Per Capital Returns, such returns attract capital — and they did: Tradeweb’s credit push, Trumid, ICE, TP ICAP, dealer-backed venues, and the portfolio-trading protocol itself. Returns are now mean-reverting exactly on schedule: operating margin 54.4% → 40.4%, and net income lower in 2025 ($246.6M) than in 2020 ($299.4M) despite 22.8% more revenue. [FACT] Critically, every recovery signal is absent: the field is widening, capacity is growing, price discipline is deteriorating, and nobody is quitting. The capital cycle has not turned. [INTERPRETATION] This framing deserves emphasis: much of what looks like company-specific failure is the capital cycle working as designed on a business that earned too much for too long.

Regulation. TRACE post-trade reporting (FINRA) underpins the share estimates used throughout this memo. MiFID II governs the European book. An April-2026 FINRA proposal to suppress affiliate back-to-back TRACE reports could restate every market-share figure in this report — MarketAxess’s Open Trading is matched-principal, so it may inflate both numerator and denominator. Direction unknown. [FACT for the proposal; OPEN QUESTION for the impact — this is a genuine unquantified risk to the entire share dataset]

Verdict — §3. A structurally deteriorating industry: genuinely excellent five years ago, being competed back toward ordinary at speed. It now favors neither scale players nor specialists, because the scale advantage that defined it has been dissolved by vendor-side aggregation. The secular tailwind is real but is being fully consumed by price: volume roughly doubles, price roughly halves, revenue creeps. The profit pool is small, a structurally indifferent competitor (Bloomberg) caps pricing, and the capital cycle is mid-bust with no recovery signal. A rational allocator should expect industry returns to keep mean-reverting toward the cost of capital. The honest counterweight: this remains a ~40%-operating-margin, capital-light, net-cash industry, and the less-contested corners (EM, Eurobonds, munis) are genuinely more attractive — but they are too small to carry the whole.

4. Competitive Position

The claim to be tested. The bull case — which rests on an unqualified network-effect moat, a proposition asserted widely in promotional bull-case commentary and treated here as a hypothesis to falsify rather than as evidence — rests on one proposition: MarketAxess’s Open Trading all-to-all pool is a network effect; liquidity begets liquidity; the deepest pool of anonymous credit liquidity is structurally unassailable.

Greenwald gives a decisive test, and it is not a matter of opinion. A network effect is economies of scale combined with customer captivity, and its signature is market-share stability. If share moves more than ~5 percentage points over 5–8 years, no barrier exists; under ~2 points, barriers are formidable. A genuine liquidity network in an electronifying market should compound share: as the market goes electronic, the deepest pool should win disproportionately. We need not argue about whether the network feels liquid. We need only look at the share.

The share record — MarketAxess’s own disclosure (estimated share of total, all-means market volume, %):

[FACT — FY2021–FY2025 10-Ks and Q1-2026 10-Q] 2022 was the peak; every product has fallen since. High-yield’s −5.4pp in three years fails Greenwald’s “>5pp = no barriers” test outright.

The single most important fact in this report. Dividing MarketAxess’s disclosed share of the total market by its disclosed electronic penetration of that market yields its share of the electronic market — the arena where it actually competes. Both series are defined against the same denominator in every 10-K, verbatim: electronic penetration is “the level of electronic trading as a percentage of all means of trading,” and market share is “our estimated market share of total U.S. high-grade corporate bond volume.” The division is valid. [FACT — basis verified across FY2021–FY2025 10-Ks]

MarketAxess’s share of the electronic U.S. high-grade market halved — 60.0% → 30.7% — in four years, during the largest electronification wave in the product’s history: the very tailwind it sells to investors. [FACT — derived arithmetic on MKTX’s own continuously-disclosed series; independently corroborated: Morningstar estimates ~33% on a high-grade+high-yield combined basis, and rebuilding our derivation on that same basis gives 32.2%. Two independent methods agree within one point.]

The electronification tailwind arrived, and competitors captured all of it. The FY2025 10-K prints MarketAxess’s falling share and its claim of “a long runway for market share growth” in the same paragraph. [FACT]

A necessary caveat, stated plainly: the equivalent high-yield derivation (~76% → ~42%) is not reliable and we do not lead with it. MarketAxess revised high-yield penetration 20%→30% in a single year and then froze it at “approximately 30.0%” for four consecutive years — that is un-updated boilerplate, not a measured series, and any derivation resting on it inherits the defect. Rebased to 2022, electronic high-yield share fell roughly 60% → 42%. The high-grade series is the one to rely on. [ASSUMPTION/OPEN QUESTION]

Naming the moat in Greenwald’s taxonomy. MarketAxess has economies of scale in a single protocol — Open Trading, anonymous all-to-all odd-lot credit — essentially without customer captivity. Supply/cost advantage: absent. Demand/captivity: absent, because of multi-homing. The buy-side is already on MarketAxess and Tradeweb and Bloomberg and Trumid, reachable through the same EMS. Nobody switches; they simply route the next ticket elsewhere. There is no switch to make, so there is no switching cost — and best-execution obligations legally require clients to shop each inquiry, destroying habit. Scale without captivity is not a barrier to entry. A large multi-homed network is a directory, not a moat: counterparties grew 1,700 → 1,800 in four years (+6%) while share collapsed. [INTERPRETATION, on Greenwald’s framework]

The cannibalization tell — the fact that proves the moat is mis-attached. Open Trading variable transaction fees are declining in absolute dollars: $178.5M (2023) → $178.0M (2024) → $175.6M (2025) — while credit volume rose ~24.8% over the same span. [FACT — FY2023–FY2025 10-Ks, corroborated in PwC’s Critical Audit Matter] Open Trading has been stalled at 35–37% of eligible volume for four years, and price improvement per dollar traded roughly halved (~10.1 → ~4.9bps) — that being its entire customer value proposition (partly cyclical; caveated). The differentiated protocol is shrinking in dollars while cheap substitute protocols grow.

Why: a portfolio trade is not a network transaction. This is the mechanism, and it is the report’s central insight. A portfolio trade is a bilateral balance-sheet transaction: you do not need 1,800 anonymous counterparties, you need one dealer who can warehouse a 2,000-line basket and hedge its duration cheaply. Blocks and dealer matching sessions are the same in kind. The protocols that are growing are precisely those where MarketAxess’s moat mechanism confers no advantage at all. And management’s three headline “key initiatives” — portfolio trading, block, dealer-initiated — are, by its own admission on the call, the ones that “come in at that lower” price point. MarketAxess is not being disrupted. It is being routed around — and it is helping. Growth is being bought with price. [INTERPRETATION, resting on the FACTs above]

The rates asymmetry — real, but the naive version is false. The common claim is that Tradeweb wins credit because it can auto-spot the Treasury hedge on a credit trade, and MarketAxess cannot. In fact MarketAxess does offer U.S. Treasury hedging. The real mechanism is liquidity depth, not feature availability — and the evidence for unfixability is MarketAxess’s own failed attempt: government-bond share went 2.6% (2021) → 3.5% (2022 peak) → 2.4% → 2.4%. It tried, invested (LiquidityEdge, ~$150M, 2019), and gave the gains back. [FACT] This is one of three reinforcing mechanisms, not a monocausal explanation. [INTERPRETATION]

Head-to-head, with numbers.

• vs. Tradeweb — the direct competitor and the share gainer: record 22% share, institutional RFQ ADV +30%, revenue ~+18%, and a ~⅓ fixed-fee cushion MarketAxess lacks. TW is compressing its own fee capture faster (−14.3% cash credit) — it is buying the share, but it is winning it.
• vs. Bloomberg — the unmodeled incumbent rail, bundled into the Terminal, structurally indifferent to per-million economics. A permanent price ceiling.
• vs. Trumid — the existence proof that captivity is gone: a venture-backed entrant reached relevant scale in a market supposedly protected by a network effect.
• vs. ICE (BondPoint/TMC) — adjacent, sub-scale in institutional credit, but a persistent bid for the same flow.

Management’s own scorecard confirms it. The FY2025 10-K discloses that PSU payouts on the US-credit-market-share metric were written down 27.6%, and the share metric paid 0% in 2024. [FACT — DEF 14A / FY2025 10-K] The company’s own compensation machinery has registered the share loss even as the MD&A calls the runway long.

A correction to a common market narrative, on the evidence. A widely-repeated narrative — typically sourced to Tradeweb’s own competitive commentary — describes MarketAxess as “resurgent” and frames MarketAxess share gains as a risk to Tradeweb. MarketAxess’s filings falsify this on every measure: high-grade share 20.4% → 17.1%, high-yield 17.1% → 12.5%, credit fees +1.6%, adjusted net income +0.2%, fee-per-million −7.6%. The claim traces to Tradeweb management commentary — a hypothesis, not evidence — and is internally incoherent (Tradeweb cannot post a record 22% share while MarketAxess is resurgent, with no donor). The “resurgent MarketAxess” narrative does not survive its own filings. [FACT / INTERPRETATION]

Verdict — §4. A weak and narrowing advantage — a real moat, mis-attached. This is neither clean pole of the question. MarketAxess genuinely built the deepest anonymous all-to-all credit pool in the world, and that pool has a genuine liquidity externality. But it commands scale without captivity, in an industry where vendor-side aggregation has stripped captivity to zero — and the volume is migrating to protocols where its network is irrelevant, while the network’s own revenue shrinks in absolute dollars. The network-effect hypothesis, tested on the metric Greenwald says is decisive, is falsified: share fell on every product from a 2022 peak, and share of the electronic high-grade market halved. What remains is a large, useful, multi-homed directory with a strong brand and real technology — worth something, but not a barrier to entry, and not the compounding franchise the 2020 multiple was paid for. Crucially, this is not a melting ice cube: revenue has never declined, margins are 40%, and returns still exceed the cost of capital by a wide margin (§6.4). The right question is not “broken or over-punished?” but “at what ROIC does the mean-reversion stop?” — and that is a valuation question (§10), not a business-quality one.

5. Growth History and Forward Opportunities

The historical record. Revenue grew $689.1M (2020) → $846.3M (2025): a +4.2% CAGR post-2020, and +3.6% in FY2025. Against this, operating income fell 8.8% ($374.7M → $341.8M) and diluted EPS went nowhere ($7.85 → ~$7.75 normalized). [FACT] The 2018–2020 period is a different company: revenue compounded ~26%/yr into a COVID-volatility windfall peak (FY2020 revenue +34.8%, operating margin 54.4%) that has never been repeated. FY2020 is not a valid baseline for anything; FY2021 is the cleaner base. [INTERPRETATION]

The decomposition — where the growth went. Credit volume grew +48.7% (2021→2025), from ~$2.625tn to ~$3.903tn implied. Credit variable fees grew +11.7% ($485.0M → $542.0M). The exact bridge:

[FACT — derived from 10-K MD&A; validated: implied FY25 volume growth of +10.0% matches the 10-K’s stated “10.0%” exactly]

MarketAxess surrendered 76% of its volume-driven revenue growth to fee erosion. That single number is the growth story of the last five years. [FACT]

Is the growth high or low quality? Low — and deteriorating in composition. It is (a) volume-driven, not price-driven — the company is winning tickets by pricing them cheaper; (b) increasingly acquired rather than organic — technology services went $3.0M → $13.9M on Pragma, and RFQ-hub (2025) contributes to “Other” variable fees, flattering Q1-2026; © concentrated in the lowest-fee protocols — portfolio trading ADV +71%, block ADV +33%, while Open Trading, the high-fee differentiated protocol, shrinks in dollars; (d) unaccompanied by operating leverage in the reported numbers — though §6.2 proves the leverage exists and is being consumed by price; and (e) not cushioned — the recurring lines compound at 2–4% organically on 13.2% of revenue. [FACT/INTERPRETATION]

Forward opportunities — honestly assessed. There are four, and they are not nothing:

1. Electronification runway. Genuine: high-grade at 60% penetration, high-yield ~30%, EM and munis far lower. But §3’s arithmetic caps what it is worth: ~2.0x volume against ~0.56x price ≈ +1.5%/yr revenue. Real in volume, largely illusory in profit. [INTERPRETATION]
2. The less-contested corners — the best genuine growth in the business. EM + Eurobonds are now 40.5% of credit ADV (from 32.7% in 2020), with record FY25 commissions and Q1-26 volumes +29.8%/+20.4%. Revenue outside US credit grew ~20%. Municipals’ addressable ADV grew +45.5%. This is real, and it is where MarketAxess is actually winning — but it is too small to carry an $846M revenue base against the US credit decay. [FACT/INTERPRETATION]
3. Adjacencies via M&A — Pragma (equities/FX algos), RFQ-hub (ETF/derivatives RFQ). Bought, not built; see §7 for whether they earn their cost of capital.
4. The Q1-2026 inflection. Revenue +11.9%, operating income +14.2%, op margin +0.87pt — the first YoY margin expansion in five years. [FACT — Q1-2026 10-Q] The first evidence that volume growth can outrun fee erosion. It is also flattered by the RFQ-hub acquisition and an easy comp, and fee-per-million still fell 5.0%. One quarter is not a trend. [INTERPRETATION]

Verdict — §5. Low-quality growth: volume bought with price, on a treadmill that accelerates. MarketAxess grows units at a healthy ~10%/yr and converts almost none of it into profit, because it surrenders ~7%/yr of unit price to do so — the arithmetic of a business trading margin for relevance. The genuine bright spot (EM/Eurobonds/munis, ~20% growth on 40% of credit ADV) is real and under-appreciated but cannot outrun the core’s decay at current weights. The forward opportunity set is not empty; it is simply worth far less than the volume headline implies, because the industry converts volume into revenue at a shrinking rate. Growth without economics is not investable, and for five years this has been growth without economics.

6. Financial Quality

6.1 The quality-of-earnings gate: both headline EPS numbers are wrong, in opposite directions

No valuation conclusion is possible until this is resolved, and it resolves cleanly. [FACT — 10-Q 2025-03-31 Note 8; FY2025 10-K Notes 9 & 20; EDGAR XBRL]

The screen shows TTM GAAP EPS of $8.45 against FY2025 GAAP EPS of $6.64 — an apparent +27% surge. It is not growth. It rebuilds exactly from EDGAR XBRL (Q2-25 $1.91 + Q3-25 $1.84 + Q4-25 $2.50 + Q1-26 $2.20 = $8.45): the “jump” is Q1-2025’s anomalous $0.40 rolling out of the window.

Q1-2025 had zero operating content. Pre-tax income was $96.15M vs. $96.72M a year earlier — flat. The EPS collapse to $0.40 was 100% tax: an 84.3% effective tax rate, driven by a $54.9M provision for unrecognized tax benefits booked after a New York State tax court ruled in a matter MarketAxess was not a party to, reversing a lower court ruling that had supported its historical filing position. [FACT]

The twist that changes the answer. FY2025’s full-year notable for prior-period uncertain tax positions is only $23.6M, not $54.9M — and at 9M-2025 it was still the full $54.9M. A ~$31.3M reserve release therefore landed in Q4-2025, flattering Q4’s $2.50 GAAP EPS by ~$0.82. [FACT — verified across four documents; the UTB balance moved $56.4M → $22.4M] Ex-notable, Q4-25 earned $1.68 — the weakest quarter of the year. Ex-notable net income declined through 2025: Q1 $70.0M, Q2 $74.5M, Q3 $68.3M, Q4 $61.9M.

FY2025 GAAP is depressed by the charge; TTM GAAP is flattered by its release. Neither is the run-rate. Two independent routes converge on normalized run-rate EPS of ~$7.75 (trim toward ~$7.60 if the now-recurring “repositioning” severance of ~$0.14/sh is also deducted); normalized effective tax rate ~26%. The $7.39 figure is MarketAxess’s own disclosed non-GAAP reconciliation, not our arithmetic. [FACT]

The charge was real cash, not a paper reserve: FY25 current state & local tax provision jumped to $41.2M from $9.7M. The New York State matter is now closed (closing agreement 2026-02-18 covering 2015–2023); the New York City exam remains open and widened from 2016–2018 to 2016–2023 between the 10-K and the 10-Q, with $11.2M of unrecognized tax benefits live. [FACT — residual risk, quantified]

Implication. Against FY2024 ex-notable EPS of $7.28, normalized growth is ~+3%/yr — and a ~5% shrinking share count is doing much of that work. At $115.50 the true multiple is 14.9x normalized, not the 13.7x the screen shows.

6.2 The 14-point margin decline, decomposed — it is price, not opex

Operating margin fell 54.38% (2020) → 48.25% → 45.51% → 41.86% → 41.72% → 40.39% (2025). Operating income has not grown in five years: $374.7M → $341.8M (−8.8%) on revenue +22.8%. From 2021 to 2025: revenue +21.1%, opex +39.5%, operating income +1.4%. [FACT — ties exactly to filings]

The natural hypothesis is that management over-invested. It is wrong. Holding credit fee-per-million at 2021’s $184.78, FY2025 revenue would have been $1,025.4M and operating income $521.0M — an operating margin of 50.81%:

[FACT — arithmetic ties exactly: 48.25 + 2.56 − 10.42 = 40.39]

Had fee capture merely held, margins would have expanded ~2.6 points despite a 39.5% opex build. The entire margin decline is fee erosion. Operating leverage on volume more than covered the investment. This is the most important analytical result in the memo, and it cuts both ways: it convicts the bear case on mechanism, and it hands the bull case a proven ~50% incremental margin waiting to be released if the fee rate ever stabilizes.

The counterfactual is an upper bound and is labeled as such [ASSUMPTION] — it assumes all FY25 volume would clear at 2021 pricing, whereas some portfolio-trading volume is genuinely incremental and exists only at a low price. But a bounding test constrains how much of the decline is benign mix: if the 2021 base book had held $184.78, the incremental $1.278tn of volume must have cleared at just ~$44.58/million to produce the actual $542.0M — far below any plausible protocol rate. ⇒ The base book must itself have repriced. This is not purely benign incremental mix. [INTERPRETATION — and the single most load-bearing unresolved question in the report, because a mix effect terminates when the mix completes, while a price effect need not]

Mix is the mechanism; price is the substance. Three proofs: (a) the bounding test above; (b) Open Trading commissions declining in dollars ($178.5M → $175.6M) while volume rose ~25%; © the 10-K’s own language — portfolio trading carries a “lower-fee structure” and is used “in lieu of more established trading protocols designed to generate price competition on individual bonds.” “In lieu of” is substitution. Protocol substitution is economically indistinguishable from a price cut. [FACT for quotes; INTERPRETATION for the reading]

Opex detail (2021→2025), for completeness: technology & communications $42.5M → $78.3M (+84.3%, the fastest line — cloud/hosting); G&A $14.5M → $26.6M (+83.4%); employee compensation $170.9M → $248.5M (+45.4%); D&A $53.4M → $76.7M (+43.5%, acquired-intangible amortization from Pragma/RFQ-hub); professional/consulting $41.9M → $31.5M (−24.9%); clearing $16.1M → $16.6M (+3.2%). [FACT] The build is real, but it is not the cause.

6.3 Revenue composition — ~71% is a volume lottery

Covered in §2 and not repeated: variable transaction fees are 70.9% of revenue; truly subscription-like revenue is 13.2%, growing 2–4% organically ex-FX. There is no annuity here to cushion a bad mix. [FACT]

6.4 Returns — still comfortably above the cost of capital, and the vendor data is wrong

A data correction that materially affects the verdict. ROIC.ai’s ROE and ROIC for MarketAxess do not reconcile to any reported equity figure and must not be used. Its claimed FY2025 ROE of 16.749% on net income of $246.6M implies equity of $1,472.5M — but reported equity is $1,145.7M (ending) / $1,388.7M (beginning). Its FY2020 claim of 43.054% implies $695.4M against reported $955.1M / $770.1M. Neither matches. (Its margin data, by contrast, ties exactly.) This is the same failure mode documented for other tickers in LEARNINGS.md. [FACT — verified]

Rebuilt from the filings (net income per EDGAR ÷ average equity):

[FACT] The direction the vendor data implied (roughly halved) is right; the levels are not — and the levels matter, because the bear case leans on them. FY2025 NOPAT is $251.9M (operating income $341.8M × (1 − 26.3% normalized tax)). On total capital (average equity + debt, cash included) ROIC ≈ 18.3%; on operating capital excluding the cash/regulatory pile, ≈ 35%.

WACC — and why CAPM is unusable here. AZI and FactorsToday both report a beta of 0.11355, which would imply a cost of equity of ~4.7% for an equity with ~33% annualized volatility and a −76% five-year maximum drawdown. That is not a credible discount rate. [FACT for the beta; INTERPRETATION for the rejection] We use a judgment build-up instead: risk-free ~4.2%, equity risk premium ~4.5%, beta ~0.9–1.0 → cost of equity ~8.3–8.7%; debt is trivial (4.8% of capital, $220M at 5.1%). WACC ≈ 8.5% (band 8–9%). [ASSUMPTION — stated explicitly because §10 rests on it]

⇒ ROIC of ~18% (total capital) to ~35% (operating capital) against a WACC of ~8.5% is a spread of +10 to +26 points. MarketAxess still earns well above its cost of capital. In Greenwald’s diagnostic framing this sits comfortably inside — not at the bottom edge of — the band where competitive advantages are present. The business is not broken on returns; it is decaying on price. The concern is the trend, and that incremental capital is being deployed into a shrinking fee pool. [INTERPRETATION]

6.5 Cash flow and SBC — genuinely high quality, and the strongest part of the story

[FACT — rebuilt from cash flow statements]

This is real, clean cash. Three points deserve emphasis because they are unusually favorable and cut against the bear case:

• SBC is genuinely small — $30.9M in FY2025 = 3.7% of revenue, 10.5% of owner FCF (3.6–4.2% of revenue across 2021–2025). This is a sharp and creditable contrast with platform peers.
• Capitalized software ≈ amortization: $53.0M capitalized in FY25 against ~$50.5M of D&A on the pool; net book value moved $107.3M → $112.4M. Steady state — operating income is not flattered by a capitalization ramp. (The cash cost has risen from 4.7% to 5.9% of revenue, and is correctly deducted in owner FCF.)
• Management’s own FCF definition is flattered by SBC, as always: company-defined FY25 FCF is $346.9M vs. our owner FCF of $293.2M — the ~$54M wedge is the SBC add-back. Use owner FCF.

This is not a company whose free cash flow is an accounting artifact. [INTERPRETATION]

6.6 Balance sheet — still net cash, but the cushion is halving and the posture changed

[FACT — 10-K FY2025 Notes 12/20; 10-Q 2026-03-31] Net cash has roughly halved in five quarters. MarketAxess drew its revolver for the first time in a meaningful way — $220.0M outstanding at 12/31/2025 under a $750M commitment at 5.1%, maturing 2026-08-09 (amended and restated 2026-02-04, same $750M) — to fund buybacks (§7). Total stockholders’ equity fell $1,388.7M → $1,145.7M in FY2025, driven by repurchases.

A caveat with teeth on the cash: much of it sits inside regulated broker-dealer subsidiaries — net capital/financial resources of $552.1M in excess of the $36.6M required — and supports Open Trading matched-principal settlement. The 10-K notes Open Trading growth “is dependent on the willingness of our customers and counterparties.” The pile is partly a moat input, not free cash. [FACT/INTERPRETATION]

Still net cash and unlevered by any normal standard; the August-2026 revolver maturity is housekeeping, not a risk. But levering a never-drawn balance sheet to buy back stock while cash halves is a genuine change in financial posture, and it deserves flagging. [INTERPRETATION]

6.7 One-time items across the five-year set

• FY2020 is a COVID-volatility windfall and is not a valid baseline (revenue +34.8%, operating margin 54.38% — never repeated). Using it as the margin baseline overstates the decline; FY2021 is the cleaner base.
• FY2025: New York UTB reserve +$54.9M (Q1) less ~$31.3M release (Q4) = $23.6M net; repositioning severance $5.1M; acquisition-related $0.6M.
• Q1-2026: repositioning $1.5M + other notables $0.7M − tax $0.5M = $1.6M.
• “Repositioning” severance now appears in FY2025 and Q1-2026 — it is becoming recurring and should not be freely added back (~$0.14/sh in FY25). [INTERPRETATION]

Verdict — §6. High-quality financials attached to a deteriorating price, with the deterioration precisely located. The cash is real (owner FCF $293M, 119% of net income), the SBC is small (3.7% of revenue), the accounting is clean (capitalization ≈ amortization; the vendors are wrong, not the filings), the balance sheet is net cash, and returns remain 10–26 points above the cost of capital. Do economics improve with scale? Yes — demonstrably. That is the tragedy of the numbers: §6.2 proves the scale economics work and are being entirely consumed by price. Had fee capture held, FY2025 would have printed a 50.8% operating margin despite a 39.5% opex build. Every dollar of operating leverage this business generated over five years was handed to clients in the form of a lower fee per million. The financial quality is not in question. The price per unit is — and it is the only thing that is.

7. Capital Allocation

Verdict up front: no. Capital allocation is the weakest link in the story, and it converts a decelerating franchise into an actively value-destroying one. [INTERPRETATION, on the FACTs below]

7.1 The buyback — $283.9M destroyed

[FACT — 2024/2025 exact from the 10-Ks (the exact years carry $495M of the $646M); 2021–22 estimated]

$646.2M deployed at a blended $206.02 is worth $362.3M today. $283.9M destroyed — ~$8.07 per current share. Every vintage is underwater.

A reconciliation worth recording, because the filings appear to disagree and do not: XBRL reports FY25 PaymentsForRepurchaseOfCommonStock of $420.0M (cash) while the equity note reports $360.0M / 1,980,715 shares. They tie to the dollar: $420.0M cash − $60.0M of the ASR parked in additional paid-in capital pending its 2026-02-04 settlement = $360.0M of treasury stock. The correct full-year figures are $420.0M / 2,340,497 shares / $179.45 blended. [FACT — verified]

They borrowed to do it. A never-drawn balance sheet was levered $220M to fund a $300M ASR at $171.84 in December 2025. The CFO on the Q4-25 call: “we did take out about $220 million on our revolver… to put a little bit of leverage on it to do that ASR. So our first order of business is going to be to pay that down over time.” [FACT — Q4-2025 earnings call] The debt outlived the value: the ASR is ~33% underwater; the full FY25 program ~36%.

The timing inverts cheapness — and answers the conviction question: they are pulling back. $0 repurchased in 2023 with ~$100M of authorization idle while the stock traded ~$271; $420M in 2025 at $179; and today, with the stock at $115.50 — the cheapest it has ever been on every metric (§10.2) — ~$205M sits idle while management repays the revolver. Buying high forced abstention when cheap. [FACT for the amounts; INTERPRETATION for the reading] This is the mirror image of the insider record (§7.4): the company spent $420M at $179 and will not spend at $115.

7.2 M&A — bought what it could not build, with one clear failure

• LiquidityEdge (2019, ~$150M) — the US Treasuries platform meant to close the rates gap that §4 identifies as structurally decisive. It failed. Government-bond share went 2.6% → 3.5% (2022 peak) → 2.4% → 2.4%. Mentions in the 10-K went 3 → 1 → 0, and it was dropped from the FY2025 acquisition list entirely. [FACT]
• Regulatory Reporting Hub (Deutsche Börse, 2020) — post-trade; contributes to a line compounding at ~1.4% organically.
• Pragma (2023) — quantitative/algorithmic trading tech; the source of technology services’ $3.0M → $13.9M jump. Bought, not built.
• RFQ-hub (2025-05-09, 90.3% for $82.3M) — ETF/derivatives RFQ; flatters Q1-2026’s “Other” variable fees and therefore the inflection in §5.

A structural point that matters more than any single deal: MarketAxess “operates as a single reporting unit.” Goodwill is therefore tested against the fair value of the entire company — which means ~$394M of goodwill and intangibles (34% of book) is effectively never impairable at a $4.1bn market cap. The clean impairment record is an artifact of segment reporting, not evidence of allocation discipline. LiquidityEdge failed and no writedown will ever say so. [FACT / INTERPRETATION]

7.3 Incentives — management is not paid on anything that has gone wrong

There is no EPS metric. No ROIC. No ROE. No TSR — absolute or relative — anywhere in the plan. [FACT — DEF 14A 2025 & 2026]

• Annual bonus: adjusted operating income (75% for the CEO, raised from 60%) + individual/strategic (25%).
• Long-term: 50% PSUs on US credit market share / revenue growth ex-US credit / operating margin; 50% time-vested with no performance condition at all.

The bonus grid has no threshold. It is linear from $0 adjusted operating income = 0%, paying 50% at half of target profit. FY2024 missed by 3.9% → 96% funding; FY2025 missed by 5.8% → 94% funding. CEO Chris Concannon’s bonus was $1,525,000 in both years — identical to the dollar. [FACT] Adjusted operating income is defined as “operating income before… the impact of cash incentives” — measured before the bonus it funds.

Goalposts ratchet off the company’s own prior-year result — “set in line with the Company’s 2025 results.” [FACT] This is precisely how the operating-margin metric funded at 78% while margin compressed from 54% to 40%.

Pay-versus-performance: FY2025 TSR of $33.56 against a peer group at $194.20; CEO pay rose 22% to $7,101,027. Say-on-pay support rose from 94% to 98%. [FACT — DEF 14A 2026] (A trap for future readers: the 2026 proxy silently rebases the TSR series; do not mix it with the 2025 proxy’s.)

The honest counterweight, and it is real: the PSUs did fail — paying 43%/45%, with the share metric at 0% in 2024 and share-metric payouts written down 27.6%. [FACT] The plan is not a rubber stamp. But a 43% payout is a high floor against a −66% five-year stock, and half the long-term award vests on time alone.

7.4 Insider behavior — the tell

Across the entire 60-month Form 4 corpus (430 deduped transactions, re-parsed from 225 raw ownership XMLs) there are exactly two open-market purchase events, ever — 5,270 shares, ~$1.29M — and both predate the decline:

Zero purchases dated 2024-01-01 or later — across the entire slide from ~$270 to the $109.09 five-year low to $115.50. Sell:buy ratio ~30:1 (~$39.2M sold). No insider has ever bought a share below $200. [FACT — re-derived from raw XML]

The datum bites because of the 2023 baseline. Concannon proved he will spend ~$1M of his own money when he believes the stock is mispriced — at $238. He has not repeated it at $170, $146, $116, or $109. That is a revealed preference, not a constraint. [INTERPRETATION]

Founder Rick McVey’s tell: sold 50,000 shares at ~$270.06 = $13,503,229 on 2024-11-12/14, six days after the Q3-24 print — discretionary, explicitly not 10b5-1-planned (aff10b5One=false), and notable because his 2021 sale was plan-designated, so he knows the difference. He retains 562,029 shares, so he is not exiting. But $13.5M sold discretionarily at $270 against $0 bought at $109–170 is the asymmetry. [FACT / INTERPRETATION] (Note: only ~24.3% of sales carry a 10b5-1 flag, but the tag did not exist before 2023, so the plan/discretionary split is not reliably computable across the full window.)

7.5 The SBC credit that indicts the record

Dilution is only ~0.25%/yr and the share count genuinely fell 6.4% — this is a real, creditable contrast with platform peers. And EPS still fell from $7.85 to $6.64. $646M of buybacks bought approximately 16 cents of EPS. [FACT/INTERPRETATION] The buyback was not mopping up dilution; there was almost none to mop. It was a deliberate deployment of $646M into a decaying franchise at an average price 78% above today’s.

Verdict — §7. Management has not allocated capital intelligently, and this is the clearest negative verdict in the memo. $646M was deployed at $206.02 into a business whose fee rate was falling every year, funded latterly by levering a pristine balance sheet — destroying $283.9M, ~$8.07 per current share — while the one acquisition that addressed the company’s decisive structural gap (LiquidityEdge/rates) failed and can never be written down because of single-segment reporting. The pattern is procyclical in the worst way: $0 at $271, $420M at $179, idle at $115. And the incentive plan explains it: management is paid on adjusted operating income measured before the bonus, against goalposts reset to last year’s result, with no EPS, ROIC, ROE or TSR anywhere — so a decade of fee-per-million erosion, a halving of ROE and a −79% stock have cost the CEO nothing, whose pay rose 22% in the year TSR reached $33.56 against a peer group’s $194.20. To management’s credit, dilution is minimal, SBC is small, the PSU share metric did pay zero in 2024, and the dividend is well covered. But capital allocation is the bridge between business value and shareholder value, and here it has been a wrecking ball.

8. Changes and Headwinds — Last Two Years

The CEO transition (2023) — orderly, and the subsequent board build is the interesting part. The board elected Chris Concannon CEO on 2023-01-03 (announced 01-09, effective 2023-04-03); founder Rick McVey became Executive Chairman. Concannon had been COO/director since January 2019; the 8-K states expressly there were “no arrangements or understandings.” This was a four-year-telegraphed internal succession — no rupture. [FACT — 8-K 2023-01-09] Concannon’s background is electronic market structure (ex-Cboe President/COO, ex-Bats CEO, ex-Virtu/Nasdaq/Instinet). The Virtu thread is worth flagging: Concannon is ex-Virtu, the interim principal accounting officer came from Virtu, and in January 2026 the board seated Virtu’s sitting CEO Douglas Cifu. A board stacking itself with electronic-market-structure operators reads as one that knows it faces a structural problem, not a cyclical one. [FACT for the appointments; INTERPRETATION for the reading]

The disclosure withdrawal (FY2023 10-K). MarketAxess’s market-share table went from seven rows to five: “Composite Corporate Bond,” emerging-markets debt, and Eurobonds were deleted simultaneously and never restored. [FACT — verified by diffing table row labels across FY2021–FY2025 10-Ks; a grep count does not reveal this] “Composite Corporate Bond” — the company’s own headline metric, which included EM and Eurobonds — appeared 4× in the FY2022 10-K, 1× in FY2023, and 0× in FY2024 and FY2025. MarketAxess still reports EM and Eurobond volumes (+29.8%/+20.4% in Q1-26) but never the share.

We state the defensible version and explicitly decline the tempting one. The motive is not established, and the obvious accusation is wrong: the very filing that deleted the rows attributed rising Eurobond volumes to “increases in estimated market volumes and our estimated market share” — MarketAxess said share was rising in the deletion year. [FACT] The defensible finding is narrower and still meaningful: the withdrawal removed verifiability, and the attribution language shifted — share was cited as a volume driver in 2021–2023 and never once in 2024–2025. [FACT / INTERPRETATION] (A circulating third-party “18.2% → 15.7%” Eurobond-decay figure does not reconcile to the 10-K basis (12.1% → 15.4%, rising) and rests on a different denominator; this note cites the disclosure deletion, not the rate.)

Fee-per-million erosion became structural (2024–2025). The defining change of the period, covered in §2 and §6: the stated cause migrated from duration to protocol mix, and in Q1-2026 duration flipped to a tailwind and fee-per-million still fell 5.0%. The cyclical alibi has expired. [FACT]

Portfolio trading became a prisoner’s dilemma with no exit. The 10-K concedes offering PT costs price; refusing it costs share. Blocks and dealer matching sessions compound the same dynamic. [FACT — FY2025 10-K risk factors]

M&A: RFQ-hub (2025-05-09, 90.3% for $82.3M) — ETF/derivatives RFQ, and a real contributor to the Q1-2026 inflection that the bull case leans on. Pragma (2023) continues to carry technology services. [FACT]

The New York tax matter (2025–2026). A $54.9M Q1-2025 charge, a ~$31.3M Q4-2025 release, a New York State closing agreement on 2026-02-18 covering 2015–2023 — and a New York City exam that widened from 2016–2018 to 2016–2023 between the 10-K and the 10-Q, with $11.2M live. [FACT] Genuine cash ($41.2M current state/local provision vs. $9.7M), now largely resolved with a quantified residual.

Capital-structure change (Dec 2025). Medium-term targets announced; buyback authorization raised to $505M; a $300M ASR executed at $171.84, funded partly by the first meaningful $220M revolver draw in the company’s history. [FACT — 8-K 2025-12-09] The stock rose 4.9% that day and fully retraced within weeks.

Sell-side capitulation (mid-2026), in real time. Rothschild Buy→Neutral, PT $189→$134 (2026-06-11); Goldman $130 (06-30); Morgan Stanley $129 (07-10); Piper $128 (07-15). Price targets have converged at or below spot, and the cuts are still arriving. [FACT — news feed]

Regulatory (April 2026). The FINRA proposal on affiliate back-to-back TRACE reporting could restate every share figure in this report. Direction unknown. [OPEN QUESTION]

Verdict — §8. On balance these changes weaken the thesis, and the most important one weakens it structurally. The single decisive development of the last two years is not an event but a reclassification: fee-per-million erosion moved, on management’s own telling, from a cyclical duration effect to a structural protocol-mix effect — and Q1-2026 confirmed it by delivering a duration tailwind and a falling fee rate anyway. Around that, the company withdrew the disclosures that would let outsiders verify its competitive position, levered a pristine balance sheet to buy stock at $179 that now trades at $115, and watched the sell-side converge on targets at or below spot. The genuine positives are real but smaller: an orderly CEO succession into a market-structure operator, a board deliberately reinforced with electronic-trading expertise (itself a tell that management sees the problem clearly), a resolved New York tax matter, and a Q1-2026 print that is the first inflection in five years. The environment has changed, and not in MarketAxess’s favor.

9. Risk Analysis

The shape of the risk. This is an unusually concentrated risk profile: rows 1–4 are all the same risk wearing different clothes, and together they are essentially the entire investment case. There is no leverage risk, no liquidity risk, no concentration risk, and no plausible path to a permanent loss of capital from the balance sheet — MarketAxess will still be here in ten years earning something. The question is only what “something” is. The risks that would ordinarily dominate a financial-sector memo (credit, funding, solvency) are absent; the risk that dominates here is the slow, compounding, unhedgeable erosion of price per unit — a risk that does not announce itself in any single quarter and cannot be observed except in a monthly disclosure.

10. Valuation Discussion (Embedded Expectations)

No price target. No recommendation. This section analyses what the price implies; it does not say what the price should be.

MarketAxess at $115.50 presents an unusually clean embedded-expectations problem, because the business has essentially one value driver — the product of credit volume growth and credit fee-per-million decay — and the market is pricing that product explicitly. The task is to solve for what decay rate the price embeds, and compare it to the one MarketAxess actually delivers.

10.1 The live multiples — rebuilt, because the screens are wrong in both directions

Two data traps must be cleared. First, the earnings (§6.1): normalized run-rate EPS is ~$7.75, not the $8.45 the screen shows nor the $6.64 FY25 GAAP prints. Second, the enterprise value. ROIC.ai reports an EV of $6.494bn — built from a stale 12/31/2025 snapshot (price $181.25, 57% above spot) and carrying $284.9M of debt against an actual $220.0M revolver. It is unusable. Third-party market caps of ~$4.35bn likewise use the FY2024 10-K cover-page share count of 37.65M; the Q1-2026 10-Q cover (2026-05-04) reports 35,539,453 shares — the ASR retired ~1.6M. [FACT]

The correct EV is ~$3.78bn, not $6.49bn — a 42% vendor error. (Cross-check: yfinance market cap $4.1bn — ties.) A caveat with teeth: much of that cash sits in regulated broker-dealer subsidiaries supporting Open Trading settlement, so it is not fully free; the honest EV band is $3.78bn (all cash netted) to $4.11bn (none netted). [FACT]

Three notes. (1) EV/EBITDA of 8.7x flatters MarketAxess: EBITDA excludes ~$53M/yr of capitalized software, a real and rising cash cost. On cash EBITDA the multiple is 10.2x — the honest figure. (2) The dividend is comfortably covered at a 40% payout. (3) Consensus forward EPS implied by the 13.2x forward P/E is ~$8.75 — ~13% above our normalized $7.75. [OPEN QUESTION] Consensus is likely on an adjusted basis and pro-forma the reduced share count; we do not adopt it. It matters only in that the sell-side is not modelling decline.

10.2 Own-history context — cheaper than at any point in twelve years

⚠️ The AZI valuation_index percentiles are degenerate for MKTX — P/E, P/B, P/S and composite all return an identical 0.418 with null history. They are not a percentile read and are not used. [FACT] We hand-built the range from ROIC’s twelve-year year-end multiple series instead (which reconciles: its FY25 P/E of 27.2x = $181.25 / $6.64 ✓).

Across twelve years, the cheapest MarketAxess ever traded — the intraday-low P/E of its cheapest calendar year (2014) — was 23.3x. Today it trades at 13.7x GAAP / 14.9x normalized: roughly 41% below the cheapest point in its entire public multiple history. [FACT] The same holds on sales (4.71x vs. a twelve-year minimum-of-lows of 6.64x), EV/EBITDA (8.7x vs. ~11.6x) and tangible book. On every metric, on its own history, MKTX has never been here.

This is context, not a conclusion. A stock below its own twelve-year floor is either (a) mispriced, or (b) correctly recognised as a different business than the one that earned those multiples. Those 2014–2021 multiples were paid for a franchise with ~54% operating margins, ~35% ROE and a rising fee rate. Today’s is a 40%-margin, 19.5%-ROE business whose fee rate has fallen 24.8% in four years. The de-rating is not evidence of error; a lower multiple is the correct response to lower growth. The only question is how much lower. [INTERPRETATION]

10.3 The comp set — and why it may be the wrong frame

(Peer multiples cross-checked against each name’s own recent disclosures and market data.)

MKTX trades at a ~29% forward-P/E discount to the exchange cohort and ~40% to Tradeweb; on EV/EBITDA the cohort discount is ~50%. [FACT]

Is the discount to Tradeweb adequate? On growth, plainly not. Tradeweb at ~22x grows ~18% and is taking share; MarketAxess at ~14.9x grows normalized EPS ~3% — with buybacks doing much of the work — and is losing it. Crudely growth-adjusted, Tradeweb’s PEG is ~1.2; MarketAxess’s is ~5.0. On the metric that decides a duopoly — who is winning — the cheaper stock is the more expensive one. A 40% price discount does not compensate for a 15-point growth gap and the wrong side of a share war. [INTERPRETATION]

But the exchange cohort may simply be the wrong frame, and this is the sharpest observation in the section. FactorsToday’s factor-similarity screen returns MarketAxess’s nearest neighbours as Coloplast (0.703), Telus (0.667), Vonovia (0.630), Enbridge (0.572), Fortis (0.535), York Water (0.538) — low-beta defensive ADRs and regulated utilities — with the actual exchanges (CME 0.604, Deutsche Börse 0.576, ICE 0.546, Tradeweb 0.537) ranking below several of them. Realized beta is 0.114. [FACT] The equity no longer trades like a growth exchange. It trades like an ex-growth bond proxy. [INTERPRETATION]

Take that seriously, because it yields a sharper answer than the exchange comp. What multiple does a ~3%-grower with a 7.1% owner-FCF yield deserve? Fortis and Enbridge — utilities with growth comparable to or better than MarketAxess’s — trade at ~19x. On that comp, 14.9x looks cheap. The difference is decisive: a utility’s unit price rises with inflation under a regulator; MarketAxess’s unit price falls 5–7% a year with no floor and no regulator. A bond proxy whose coupon compounds is a bond proxy. A bond proxy whose coupon shrinks is an amortizing annuity — and an amortizing annuity must trade at a discount to a growing one. The repricing from compounder multiple to sub-utility multiple is therefore directionally correct; whether 14.9x is the right level depends entirely on the amortization rate. That is the whole valuation, and it is solvable.

10.4 The embedded-expectations solve — what $115.50 actually underwrites

Method. Owner FCF $293.2M (§6.5) is the cash the business actually generates. WACC ≈ 8.5% (band 8–9%) [ASSUMPTION — see §6.4 for why CAPM is unusable here]. Solving a Gordon perpetuity, Value = FCF × (1+g) / (WACC − g), for the growth rate the price implies:

At $115.50 the market embeds perpetual owner-FCF growth of roughly +0.7% to +1.7%, centred near +1.0–1.3%. [FACT, conditional on the WACC assumption] Against ~2.5% expected inflation, that is perpetual real decline of ~1.5%/year, forever. The market is not pricing terminal decline — it is pricing permanent nominal stagnation.

Now convert that into the only variable that matters. Hold margins flat; credit variable fees are 64.0% of revenue and the other 36.0% grows ~3%. Then g = 0.64 × g_credit + 0.36 × 0.03, where g_credit = (1 + volume growth) × (1 + FPM growth) − 1:

This is the finding of the section. At today’s price, if credit volume keeps growing at anything like its four-year trend, the market is embedding fee-per-million decaying ~8%/year in perpetuity — faster than the −6.9%/yr MarketAxess has actually delivered over four years, and materially faster than Q1-2026’s −5.0%. [FACT] Only if you also assume the volume engine halves to ~+6%/yr does the embedded decay fall back to the realized run-rate. The price simultaneously underwrites the worst fee outcome ever realized and a deceleration in the one thing that is working. [INTERPRETATION]

Cross-check: §3’s independent structural arithmetic (~2.0x volume × ~0.56x price → ~+1.5%/yr revenue) sits above the +1.0–1.3% the price embeds. Two independent methods converge: the market is pricing the structural arithmetic, and then some. [INTERPRETATION]

What the market is underwriting correctly. (1) Fee decay is structural, not cyclical — Q1-26 had a duration tailwind and fees still fell 5.0%; protocol mix does not mean-revert. (2) The fee pool is small (~$1.6–1.8bn at full electronification); no volume outcome rescues a falling rate. (3) Share loss is real and MarketAxess-specific. (4) Open Trading — the moat itself — is shrinking in dollars. (5) Diversification is not a rescue (13.2% of revenue at 2–4% organic growth). The bears have the better of the argument on all five, and the de-rating is earned. Anyone arguing this is a mispriced compounder must explain why operating income is lower than in 2020.

What the market may be underwriting incorrectly. (1) A mix shift is a finite transition, not a perpetual rate — portfolio trading cannot exceed 100% of the book, so the mix component of decay must terminate by construction. The counter is §6.2’s bounding test, which implies the base book itself repriced — meaning some decay is genuine price, which need not terminate. [OPEN QUESTION — the single most load-bearing unresolved question in this report.] (2) Zero credit for proven operating leverage: §6.2 shows that had fee capture held, FY25 margin would have been 50.8% despite the 39.5% opex build. Any stabilization drops through at ~50% incremental margin. (3) Zero credit for Q1-2026 — the first margin expansion in five years. (4) The buyback is now large relative to the cap ($505M authorized against $4.1bn); at 14.9x, repurchase is genuinely accretive in a way it never was at 50x.

And where the market may not be bearish enough. At +1.0% embedded growth, the price still assumes operating income never declines in nominal terms. The bear path below has it declining. The current price is not a floor.

10.5 Scenario analysis — driven by the fee/volume identity

Base year FY2025: revenue $846.3M (credit variable fees $542.0M = 64.0%), opex $504.4M, operating income $341.8M, owner FCF $293.2M, 35.5M shares. Horizon FY2031. All figures are ASSUMPTIONS except the FY2025 base.

Bear — “the amortizing annuity, confirmed.” Fee decay accelerates to −8%/yr as portfolio trading, blocks and dealer-run all-to-all commoditize the residual RFQ book; Open Trading keeps shrinking in dollars; share erodes (HG 17.1% → ~14%). Opex compounds at 4% — technology and communications has run at +84% since 2021 and cloud costs do not decline on command. Revenue is flat and operating income falls ~1.6%/yr; margin lands at 28.7%. Owner FCF compounds at −6.9%/yr. This path cannot be reconciled with $115.50 at any sensible discount rate — the implied rate is below 4%, i.e. below the risk-free. The bear is not priced in.

Base — “the trend, forever.” Volume at trend, decay moderating to −5.5% as the easiest mix shift completes, margins flat at 40.4% (operating leverage exactly offsetting fee erosion, as for five years). Revenue +3.0%/yr, owner FCF +2.8%/yr, EPS to ~$9.61 on a 14%-shrunk share count. This is the last five years, projected. At $115.50 the market discounts these at ~9.5% — ~100bp above our 8.5% WACC. The price is modestly conservative if the base case holds, not dramatically so.

Bull — “Q1-2026 was the turn.” The mix shift matures, decay slows to −3%, volume re-accelerates to +12% on block (ADV +33%) and portfolio-trading (ADV +71%) momentum. The proven operating leverage is then not consumed: margin recovers to 48.8% — exactly what §6.2’s decomposition says is available. Owner FCF +10.8%/yr, EPS ~$14.70. Implied discount rate ~14%.

Sensitivity — one variable dominates. Holding all else at base, each 1 point of annual fee decay changes FY2031 operating income by ~$50M (~13%) and swings the supportable perpetual growth rate by ~65bp. Volume matters roughly half as much; opex roughly a quarter as much. Fee per million is not one of several drivers; it is the driver. Every other line in this report is second-order to the monthly volume report’s fee disclosure. [FACT — from the model above]

10.6 Sum-of-the-parts — considered and rejected

Three reasons. First, it is not constructible: MarketAxess “operates as a single reporting unit” (10-Q Note 2) — no segment profitability is reported, so any SOTP requires inventing margins, which is decoration, not analysis. Second, the stub is immaterial: information services ($53.2M) + post-trade ($44.5M) + technology ($13.9M) = $111.6M, 13.2% of revenue, compounding at 2–4%; at a generous 6x sales the whole stub is ~$670M, ~16% of the cap. Third, it changes nothing: >84% of value remains in the credit franchise, whose value is entirely the fee question. Not warranted. [INTERPRETATION]

Verdict — §10. The market is pricing MarketAxess as an amortizing annuity, and — measured against the company’s own five-year record — it is pricing the amortization at a rate MarketAxess has never actually delivered. At $115.50, on a rebuilt EV of ~$3.78bn and normalized EPS of ~$7.75, it trades at 14.9x, 8.7x EBITDA and a 7.1% owner-FCF yield — ~41% below the cheapest point of its own twelve-year history, ~29% under the exchange cohort, ~40% under Tradeweb. The price embeds ~+1.0–1.3% perpetual growth, requiring fee-per-million to decay ~8%/year forever against a realized −6.9%/yr and Q1-26’s −5.0%. The de-rating is earned and the market is right about the mechanism — fee erosion is structural, the pool is small, share is genuinely being lost, the differentiated protocol is shrinking, and the diversification compounds at 3%. But the market is now extrapolating a bounded transition as a perpetuity and giving zero weight to a proven asymmetry — that if fee capture merely stabilizes, margins return toward 50% because the scale economics never stopped working. The honest conclusion is that both tails are live and the base case is only modestly favourable (~9.5% implied vs. ~8.5% WACC). If the bear case is right, $115.50 is not defensible above a 4% discount rate — the price is not a floor and the “cheapness” is an illusion. The valuation does not resolve the debate. It collapses it onto a single monthly datapoint: whether fee per million stabilizes. Q1-2026’s −5.0% is the first mild evidence in five years that it might. One quarter is not a trend.

11. Variant Perception

The consensus belief. Sell-side consensus is not modelling decline: the implied forward EPS of ~$8.75 sits ~13% above our normalized $7.75. But the direction of travel is capitulation, in real time — Rothschild Buy→Neutral with a target cut from $189 to $134 (2026-06-11), Goldman $130, Morgan Stanley $129, Piper $128 — targets have converged at or below spot and the cuts are still arriving. [FACT] The prevailing view has shifted from “temporarily out-of-favour secular grower” to “structurally challenged incumbent losing a protocol war,” and the price has moved with it: −79.1% from the high, −36% year-to-date, a five-year low three weeks ago.

The strongest bull case. The market has confused a bounded transition with a perpetuity, and is selling a 7.1% owner-FCF yield on a business earning 18–35% returns on capital. Portfolio trading cannot exceed 100% of the book — the mix component of fee decay must terminate by construction — and when it does, §6.2 proves what is waiting: had fee capture merely held, FY2025 would have printed a 50.8% operating margin despite a 39.5% opex build. The scale economics never broke; they were handed to clients. Any stabilization drops through at ~50% incremental margin. Meanwhile the price embeds ~8%/yr decay forever — worse than MarketAxess has ever delivered — while Q1-2026 posted revenue +11.9%, operating income +14.2% and the first YoY margin expansion in five years; EM/Eurobonds (40.5% of credit ADV) grew ~20%; high-yield share rose 190bp YoY in the June monthly; net cash is $328M; and $505M of authorization sits against a $4.1bn cap, genuinely accretive at 14.9x in a way it never was at 50x. You are paid to wait at a 41% discount to the cheapest this stock has ever been.

The strongest bear case. This is an amortizing annuity, the amortization has no floor, and the people who can see it first will not buy it. Fee-per-million has fallen four straight years and accelerated to −7.6%; the stated cause migrated from cyclical duration to structural mix; and in Q1-2026 duration flipped to a tailwind and fees fell 5.0% anyway — the alibi is gone. The moat is real but mis-attached: Open Trading’s commissions are shrinking in absolute dollars while volume grows 25%, because the winning protocols are bilateral balance-sheet trades where an 1,800-counterparty network is worthless. Share of the electronic high-grade market halved, 60.0% → 30.7%, during the biggest electronification wave in the product’s history — the tailwind arrived and competitors took all of it. Vendor-side EMS/OMS aggregation has stripped the captivity that made scale a barrier, so Greenwald says there is no barrier; Marathon says returns mean-revert and every recovery signal is absent. Capital allocation destroyed $283.9M, management is paid on a metric that ignores all of it, and no insider has bought a share since January 2024 — including a CEO who spent $1M of his own money at $238. At −8% decay, margin goes to 28.7% and $115.50 requires a sub-4% discount rate to justify.

The 3–5 assumptions that actually matter.

1. Is fee-per-million decay a bounded mix transition or a permanent repricing of the base book? The single most load-bearing question in the report. §6.2’s bounding test — incremental volume would have had to clear at ~$44.58/mm for the base book to have held — says the base repriced. If that is right, the decay does not terminate and the bull case fails at its foundation.
2. Does the ~50% incremental margin still exist, and will it ever be allowed to drop through? §6.2 says the economics are intact and being consumed. The bull case is entirely a bet on this being released.
3. Was Q1-2026 an inflection or a comp? Revenue +11.9% and the first margin expansion in five years — against an easy comp, flattered by RFQ-hub, with fees still −5.0%.
4. Does share loss stabilize? HG 17.1% and falling on the quarterly series; the June monthly showed +10bp. One month is not the trend (see below).
5. Is the ~8%/yr embedded decay a genuine over-discount, or the market correctly pricing an acceleration we can’t yet see?

What would falsify each side. Kills the bull: two or three more quarters of fee decay at or beyond −7%, with Open Trading dollars still shrinking and high-grade share breaking below ~16% — that is the amortizing annuity confirmed, and the price is then not a floor. Kills the bear: fee-per-million flat-to-−2% for two or three consecutive quarters while volume compounds ~10% — margins would begin re-expanding toward the 50% §6.2 identifies, and the ~8% embedded decay would be revealed as a gross over-extrapolation. Either falsification arrives in a monthly disclosure, not a strategy pivot.

The factor-positioning read — where consensus may be offsides, and where it is not. [FACT unless noted] The tape is not ambiguous: price below all three EMAs with the EMAs descending (117.12 / 127.01 / 157.38); Sharpe deteriorating monotonically as the window shortens (y5 −0.76 → y1 −1.65 → m6 −1.91 → m3 −2.59) — the decline is accelerating; Momentum loading negative in all four nested models and Quality zeroed in every one — cheap, falling, with no quality signature: the value-trap signature in factor space (Interp).

The decisive datum, and it forecloses the “abandoned value name” story: MKTX carries a +0.27 Value loading into a Value factor that returned +13.8% over the past year (z = +1.64) — and lost 45.4% anyway. Its factor tilts were tailwinds and it fell regardless. With R² of 14.4% and specific vol of 25.9% against 28.6% realized, ~86% of variance is idiosyncratic. There is no hostile regime to wait out; the market is repricing this company’s earnings stream (Interp). Realized beta of 0.114 confirms it is not falling with the market. And there was no capitulation: across the −36.9% slide to the June low no single session fell more than 4.7%, and the low itself traded 1.2× average volume — orderly institutional distribution, not a flush (Fact → Interp). The tape also rejects good news: the December targets/$505M authorization (+4.9%) and the Q4’25 print (+5.5%) each fully round-tripped within weeks.

Two honest counterweights to that read. First, the whole neighbourhood is being repriced: the Financial Data Titans custom basket is −21.9% over 252 days (z = −2.01), a near-extreme complex-wide de-rating — consistent with §3’s capital-cycle diagnosis and not entirely MarketAxess-specific. Second — and this is a genuine tension the memo must not paper over — the blade may be price-per-trade rather than franchise share. The June-2026 monthly volume report showed high-grade share up ~10bp to 17.9%, high-yield share up 190bp YoY to 14.9%, block ADV +33% and portfolio-trading ADV +71%, while fee-per-million fell on protocol mix. A share-losing franchise and a mix-compressing one look identical on a five-year chart and are entirely different investments. (Two necessary caveats: this is a single month against a multi-year quarterly series showing 17.1% and falling — MarketAxess’s own risk factors warn that month-end portfolio trading “can drive significant swings in … estimated market share” — and the June figures rest on a secondary source, an IR press release carried by a news feed, with no primary filing in the corpus. Do not place 17.9% on the annual trend line.) [FACT with caveats]

Where we come out. Consensus is offsides in both directions, which is why this is a genuinely hard name rather than an easy short or an easy value buy. Consensus is too optimistic on earnings — the sell-side’s ~$8.75 is ~13% above a properly normalized $7.75, and it is not modelling decline at all. But the price is arguably too pessimistic on the fee rate — embedding ~8%/yr decay in perpetuity against a −6.9% realized and a −5.0% most recent quarter. Both can be true simultaneously: the analysts have the wrong number and the market has the right idea, expressed too harshly. The variant perception worth holding is narrower than either camp’s: this is neither a broken business nor a mispriced compounder — it is a real franchise with a real moat attached to a losing protocol, decaying at a rate nobody, including management, has yet been able to date. The evidence that would settle it is not a strategic announcement or an analyst day. It is one line in a monthly volume report.

12. Fact vs. Interpretation Table

13. Open Questions

1. Is fee-per-million decay a bounded mix transition or a permanent repricing of the base book? The single most load-bearing unresolved question in this report, and the one that decides the investment. §6.2’s bounding test — the incremental $1.278tn of volume would have had to clear at ~$44.58/mm for the 2021 base book to have held its price — implies the base book itself repriced. If so, the decay does not terminate when the mix completes. If the decay is genuinely mix, it stops by construction. We cannot resolve this from outside the company’s fee schedules.
2. Why does Composite+ compound at only ~3%? Information services grew ~3.6% organically ex-FX and post-trade ~1.4%. If the pricing data is genuinely differentiated — as the moat story requires — this is inexplicably slow. Either the data is less differentiated than claimed, or it is being under-monetized. Both readings matter.
3. Was Q1-2026 an inflection or an artifact? Revenue +11.9%, operating income +14.2%, first margin expansion in five years — against an easy comp, flattered by RFQ-hub (acquired 2025-05-09), with fee-per-million still −5.0%. Two or three more quarters resolve it; nothing else will.
4. What is the true, current market share, and will FINRA’s April-2026 TRACE proposal restate it? Open Trading is matched-principal and may inflate both numerator and denominator of every share figure in this report. Direction unknown. This is an unquantified risk to the entire share dataset the bear case rests on.
5. Why did management withdraw the EM, Eurobond and Composite Corporate Bond share disclosures in FY2023 — and why has the attribution language never again cited share as a volume driver? We explicitly decline the tempting inference (the same filing said Eurobond share was rising). But the withdrawal removed verifiability precisely where the competitive question is hardest, and it has never been restored or explained.
6. Why has no insider bought a share since January 2024? The CEO spent ~$1M of his own money at $238 in 2023 and has not repeated it at $170, $146, $116 or $109. Does he see something in the fee schedule that outsiders cannot, or is this simply inertia? The single cheapest test of the bull case would be an insider buying.
7. Can the ~50% incremental margin §6.2 identifies ever actually be released, or is it structurally committed to clients? The entire bull case is a bet on the former.
8. Is Bloomberg’s Terminal-bundled venue a permanent fee ceiling, and at what level? Nobody models it; everybody trades on it. It cannot be observed and cannot be competed away.
9. What did LiquidityEdge (~$150M, 2019) actually earn, and what has Pragma earned? Single-segment reporting means no acquisition will ever be marked to its result, and none can ever be impaired at a $4.1bn cap.

14. What Must Be True

For the bull case — “the transition is bounded and the leverage gets released”

1. Fee-per-million decay is predominantly mix, not price — and therefore terminates when portfolio trading, blocks and matching sessions reach their natural share of the book. (§6.2’s bounding test is the direct challenge to this; it implies the base book repriced.)
2. The ~50% incremental margin is real and releasable — §6.2 establishes it exists; the bull case requires it to drop through rather than be handed to clients.
3. Volume keeps compounding at ~10% on electronification, blocks (+33% ADV), portfolio trading (+71% ADV), and the genuinely-growing less-contested corners (EM/Eurobonds at 40.5% of credit ADV growing ~20%; munis’ addressable ADV +45.5%).
4. Share stabilizes near current levels rather than continuing to bleed.
5. The ~8%/yr embedded decay is a gross over-extrapolation of a −6.9% realized, −5.0% most-recent rate.

Falsification test (specific, dated, observable): two or three consecutive quarters — through the Q3-2026 and Q4-2026 prints — of credit fee-per-million decaying at or beyond −7% YoY, with Open Trading commissions still declining in absolute dollars and US high-grade share breaking below ~16%. Any one of those alone is noise; together they confirm the amortizing annuity, put FY2031 operating margin on the path to ~28.7%, and mean $115.50 requires a sub-4% discount rate to justify — i.e. the price is not a floor. The bull case dies in the monthly volume report, not in a strategy announcement.

For the bear case — “the amortizing annuity with no floor”

1. Fee decay is genuine price, not mix — the base book has repriced and will keep repricing, because EMS/OMS aggregation stripped the captivity that let MarketAxess hold price.
2. Open Trading keeps shrinking in dollars — the differentiated protocol continues to be cannibalized by protocols where the network is worthless.
3. Share loss continues — high-grade below 16%, electronic high-grade share continuing down from 30.7%.
4. Bloomberg and Tradeweb hold the fee ceiling down — nobody exits, nobody disciplines price, the capital cycle stays mid-bust.
5. Capital allocation stays procyclical — management, paid on adjusted operating income against goalposts reset to last year, keeps deploying into a shrinking pool.

Falsification test (specific, dated, observable): credit fee-per-million flat to −2% YoY for two or three consecutive quarters while credit volume compounds ~10% — with Open Trading commissions returning to growth in absolute dollars. That combination would mean the mix transition has matured; §6.2’s arithmetic then forces operating margin back toward 50% at ~50% incremental margins, revenue growth re-rates toward the bull path, and the ~8%/yr decay the price embeds is revealed as a gross over-extrapolation of a transitional effect. An insider open-market purchase at these levels — the first since January 2024 — would be strong corroborating evidence, and its continued absence is the bear’s best ongoing confirmation.

The symmetry worth noting. Both falsification tests are the same disclosure, read in opposite directions, and both arrive within two quarters. That is unusually clean: this thesis is not a multi-year act of faith in either direction. It is a wager on one number, and the number reports monthly.

15. Source Appendix

See Appendix B — Source Appendix below for the full, itemized source list (primary SEC filings with accessions and URLs; earnings-call transcripts; quantitative data providers with their documented error modes; and secondary/framework sources).

APPENDIX A — Standard Diligence Questionnaire

MarketAxess Holdings Inc. (NASDAQ: MKTX) — supplemental to the research memo of 2026-07-17. Price $115.50 (close 2026-07-16). All figures USD; fiscal year ends December 31.

This appendix answers a standard diligence question set from the evidence assembled in this analysis. It is supplemental to the memo and carries no recommendation and no price target — the labeled Claude’s Take is the only place a position appears. Where a question does not map to an electronic-venue business model, that is stated and the correct sector analog is given. Management commentary is treated throughout as a hypothesis requiring external validation, not as evidence.

General

What thoughtful questions have other investors asked about this company?

Four debates dominate the external literature, and all four are live in this report.

1. “Is fee-per-million decay duration mechanics or price competition?” — the sell-side’s central argument, and management’s. The mechanics are real: US high-grade fee plans are denominated in basis points of yield, so fee-per-million is a function of DV01. Management’s published sensitivities are that every +100bps of yield cuts fee-per-million by ~$5–6, and every year of shortened maturity by ~$15. Third-party analysts have explicitly flagged the risk that “price competition is being masked by the DV01 story.” The evidence settles this against management: the stated cause migrated in the company’s own MD&A from duration (FY2022–23) to “protocol mix-shift reflecting increased portfolio trading” (FY2025), and in Q1-2026 duration was a tailwind and fee-per-million fell 5.0% anyway [FACT — FY2022–FY2025 10-K MD&A; Q1-2026 10-Q]. The cyclical alibi has expired. [INTERPRETATION]

2. “Is MarketAxess losing to Tradeweb, and is the loss self-perpetuating?” — the most-argued question in the name. The bull framing rests on the proposition that fixed income has one less layer of abstraction than equities: with no Reg NMS / NBBO analog forcing a best-execution sweep across venues, the platform is the exchange, liquidity concentrates, and share is the network-effect tell. This framing is now obsolete, and its obsolescence is the structural finding here: the abstraction layer arrived anyway — from vendors rather than regulators. MarketAxess’s own 10-K names “EMS and OMS Providers… offer aggregation of trading venue liquidity” as a competitive category [FACT — FY2025 10-K, Competition]. EMS/OMS aggregation does privately what Reg NMS did publicly. If the framing’s own logic is accepted, the answer is damning on its own terms: share of the electronic US high-grade market fell 60.0% → 30.7% in four years [FACT — derived from MKTX’s continuously-disclosed series].

3. “Did management strategically misread portfolio trading?” — a specific, dated criticism. MarketAxess publicly characterized PT as an “undifferentiated sub-category of RFQ” that would fade when volatility returned; Tradeweb was roughly a year earlier to the protocol, and PT’s share of TRACE has since risen toward ~10%. The 10-K now concedes PT carries a “lower-fee structure” and is used “in lieu of more established trading protocols designed to generate price competition on individual bonds” [FACT — FY2025 10-K]. Portfolio-trading ADV grew +71%. The misread is not the interesting part; the trap is: offering PT costs price, refusing it costs share, and both roads lead down. [INTERPRETATION]

4. “Is Open Trading a hub or a stress-window backstop?” — the sharpest question anyone has asked, and the report answers it. Independent analysis has observed that Open Trading penetration spiked in the March-2020 and 2023 stress windows but has been ~flat ex-stress for four years, concluding participants treat it as a backstop, not the default hub. The filings now corroborate and worsen this: Open Trading variable transaction fees are declining in absolute dollars — $178.5M (2023) → $178.0M (2024) → $175.6M (2025) — while credit volume rose ~24.8% [FACT — FY2023–FY2025 10-Ks; PwC Critical Audit Matter], with Open Trading stalled at 35–37% of eligible volume for four years and price improvement per dollar traded roughly halved (~10.1 → ~4.9bps).

A fifth question nobody outside appears to have asked, and it is the one that matters: is the decay a bounded mix transition or a permanent repricing of the base book? The bounding test in the Financial Quality section implies the base book itself repriced — for the 2021 base to have held its price, the incremental $1.278tn of volume would have had to clear at ~$44.58/million, far below any plausible protocol rate. A mix effect terminates by construction; a price effect need not. This is the single most load-bearing unresolved question in the analysis. [INTERPRETATION]

A methodological caution on the promotional literature: a widely-circulated bull one-pager asserts “one of the most powerful competitive moats in financial services: the network effect”. It is marketing copy, was treated as a hypothesis to falsify, and is falsified by the share record. Separately, the common narrative describing MarketAxess as “resurgent” is an assertion traceable to Tradeweb management commentary, not evidence, and contradicted by MarketAxess’s filings on every measure. It does not survive its own filings. [FACT / INTERPRETATION]

Cyclicality & Earnings Nature

Are earnings at a cyclical high or low?

Neither — and that is the finding. This is the wrong frame for MarketAxess, and applying it is the most common analytical error in the name.

The cyclical inputs are, if anything, favorable: TRACE market volumes rose +8.0% in FY2025; MarketAxess’s own credit volume grew +10.0%; and in Q1-2026 bond duration — the exogenous, genuinely cyclical driver — was a tailwind. [FACT] Yet operating income is lower today than in 2020 ($374.7M → $341.8M) on 22.8% more revenue, and net income is lower ($299.4M → $246.6M). [FACT — 10-Ks]

Earnings are at a structural low relative to their own history, reached with cyclical conditions at or above normal. Normalized EPS of ~$7.75 is essentially flat against $7.85 in 2020 — zero EPS growth in six years. The decline is not a cycle to wait out; it is the cumulative effect of credit fee-per-million falling four consecutive years, $184.78 → $138.87 (−24.8%), and −5.0% again in Q1-2026 to ~$132. [FACT — 10-K MD&A each year]

One genuine caveat against the “no cycle” reading: FY2020 was a COVID-volatility windfall (revenue +34.8%, operating margin 54.4%) that has never been repeated and is not a valid baseline for anything; FY2021 is the cleaner base. [INTERPRETATION] And a legitimate cyclical upside exists that this report does not dismiss: corporate-bond volatility drives Open Trading penetration, and the company frames 2014/2017/2021/2024 as abnormally low-volatility years. That framing fails its own arithmetic — four of the last ten years is not an anomaly, it is the base rate. [INTERPRETATION]

Note on the reported numbers before any cyclical read is attempted (§6.1): both headline EPS figures are wrong, in opposite directions. FY2025 GAAP EPS of $6.64 is depressed by a $54.9M New York State tax reserve booked in Q1-2025 (an 84.3% effective tax rate on flat pre-tax income); TTM GAAP EPS of $8.45 is flattered by the ~$31.3M release of that same reserve in Q4-2025. Neither is the run-rate. Normalized is ~$7.75. [FACT — verified across four documents; UTB balance moved $56.4M → $22.4M]

Driven by the external environment or internal actions?

Overwhelmingly internal — and the decomposition proves it beyond argument.

Holding credit fee-per-million at 2021’s $184.78, FY2025 operating margin would have been 50.81% — higher than 2021’s actual 48.25% — despite a 39.5% opex build:

[FACT — arithmetic ties exactly: 48.25 + 2.56 − 10.42 = 40.39] (The counterfactual is an upper bound and is labeled as such [ASSUMPTION] — it assumes all FY25 volume would clear at 2021 pricing, whereas some PT volume is genuinely incremental and exists only at a low price.)

The entire 14-point margin decline is fee erosion. Operating leverage on volume more than paid for the investment. The natural hypothesis — that management over-invested — is simply wrong. [FACT for the arithmetic; INTERPRETATION for the reading]

The uncomfortable half of “internal” is that MarketAxess is an active participant in its own compression. Its three headline key initiatives — portfolio trading, block, dealer-initiated — are, by management’s own admission on the call, the ones that “come in at that lower” price point. It is buying volume with price. [FACT for the quote; INTERPRETATION] The external environment (volumes, electronification, duration) has been supportive throughout.

How stable are revenues?

Structurally unstable in composition, empirically stable in aggregate — and the aggregate stability is misleading.

[FACT — FY2025 10-K]

~71% of revenue is a volume lottery. Truly subscription-like revenue (information + post-trade + technology) is only 13.2% — and it compounds at 2–4% organically ex-FX (information services +3.6%, post-trade +1.4% in FY25). [FACT — FY2025 10-K MD&A] Including fixed distribution fees, anything resembling recurring revenue is 29.0%.

This matters more than any other structural feature. Unlike Tradeweb, which carries a meaningful fixed-fee cushion (~⅓ of revenue), MarketAxess is ~87% levered to volume × price. There is no material subscription annuity to smooth a bad protocol mix — and it is now in one. [INTERPRETATION]

Empirically, revenue has never declined: $689.1M (2020) → $846.3M (2025), a +4.2% CAGR. That stability is what makes the name genuinely hard — this is not a melting ice cube. But the stability is at the revenue line only; operating income fell 8.8% across the same span. [FACT]

Outlook for products/services?

Product by product, on the disclosed numbers:

• US high-grade — the historic core; market ADV $39.0bn. Share 21.0% (2021) → 17.1% (Q1-26). Contested and losing.
• US high-yield — market ADV $12.2bn. Share 15.2% → 12.2%. The worst product; −5.4pp in three years fails Greenwald’s “>5pp = no barriers” test outright.
• Emerging markets + Eurobonds — the genuine bright spot: 40.5% of total credit ADV in 2025, up from 32.7% in 2020, with record FY25 commissions, Q1-26 volumes +29.8% / +20.4%, and revenue outside US credit growing ~20%. Real, under-appreciated — and too small to carry an $846M base against US credit decay. [FACT/INTERPRETATION]
• Municipals — addressable ADV +45.5% YoY, the fastest-growing pool disclosed; MarketAxess’s share here is small but rising.
• US government bonds / rates — market ADV $1,044.3bn; MarketAxess share 2.4%. A rounding error, and load-bearing: the company tried (LiquidityEdge, ~$150M, 2019), got share to 3.5% in 2022, and gave it all back to 2.4%. [FACT]
• Information / post-trade / technology services — real but compounding at 2–4% organically; technology services’ jump ($3.0M → $13.9M) is Pragma, bought not built.

The forward-looking honest answer is that the product-level outlook is favorable in volume and adverse in price, and price wins the arithmetic (see below). [INTERPRETATION]

How big will this market be — growing, shrinking, domestic or international?

Growing in volume, small in profit, and increasingly international at the margin.

The engine is electronification: MarketAxess discloses electronic penetration of US high-grade at 60% in FY2025, up from 35% four years earlier — a genuine, powerful secular wave; high-yield ~30%; EM and munis far lower. [FACT — FY2021–FY2025 10-Ks]

But the governing arithmetic is brutal: full electronification from here implies roughly ~2.0x volume; trend fee-per-million decay implies roughly ~0.56x price; the product is ~+1.5%/yr revenue. [INTERPRETATION — derived] This is not a forecast — it describes the realized record: revenue compounded ~4.2% post-2020 with falling earnings. And the pool is small: the entire US high-grade + high-yield fee pool at 100% electronification is only ~$1.6–1.8bn. [INTERPRETATION — derived] There is no volume outcome that rescues a falling rate.

A further caution that deserves airing: the assumption that electronification accelerates total volumes is taken as given and is unevidenced — over twenty years e-trading went from ~nil to ~40% of high-grade, yet aggregate corporate-bond turnover as a percentage of bonds outstanding “looks basically trendless” in SIFMA data. Electronification may redistribute the pie without growing it. [INTERPRETATION]

Geography: the domestic US credit core is where the share war is being lost; the international book (EM ~30 currencies with ~200 broker-dealers, Eurobonds under MiFID II) is where MarketAxess is actually winning. The mix is shifting international not by design so much as by differential decay. [INTERPRETATION]

Business Quality & Competitive Moat

Is the industry getting more or less competitive?

More — decisively, and on every observable.

The field is widening, not consolidating: MarketAxess, Tradeweb (rates-anchored, pushing hard into credit), Bloomberg (the Terminal-bundled RFQ rail), ICE (BondPoint/TMC), Trumid (venture-backed challenger that reached relevant scale), plus dealer-backed consortia, interdealer brokers (TP ICAP), and — critically — the dealers themselves, via portfolio trading and matching sessions. The 10-K concedes: “Our dealer clients have also increased their usage of matching sessions offered by competing platforms.” [FACT — FY2025 10-K]

Fee compression is industry-wide, which exonerates the industry, not the company. Tradeweb’s Q4-2025 fee capture fell in every single asset class (blended −10.2%), with cash-credit fee-per-million −14.3% YoY — a faster decline than MarketAxess’s −7.6%. (Caveat with teeth: TW and MKTX category definitions differ; the trend is robust, the levels are indicative only.) [FACT for both series; INTERPRETATION for the comparison] Nobody is winning on price; the price is simply falling.

But share loss is company-specific. Tradeweb is at a record 22% and gaining while MarketAxess loses on both high-grade and high-yield. MarketAxess is losing on two of the three terms in Revenue = volume × share × fee-per-million, rescued only by cyclical market volume. Anyone saying “it’s just the industry” is exactly half right. [INTERPRETATION]

The Marathon capital-cycle read — textbook, mid-bust, no recovery signal. MarketAxess earned a 54.4% operating margin and ~30% ROIC in 2020. Per Capital Returns, such returns attract capital — and they did (Tradeweb’s credit push, Trumid, ICE, TP ICAP, dealer venues, and the PT protocol itself). Returns are mean-reverting exactly on schedule: margin 54.4% → 40.4%; net income lower in 2025 than 2020 on 22.8% more revenue. [FACT] Every recovery signal Marathon looks for is absent: the field is widening, capacity growing, price discipline deteriorating, nobody quitting, no capex-to-depreciation collapse, no consolidation. The capital cycle has not turned. [INTERPRETATION] Much of what looks like company-specific failure is the capital cycle working as designed on a business that earned too much for too long.

How profitable is the business (ROIC, ROE)?

Still comfortably above the cost of capital, and falling — and the vendor data is wrong.

A data correction that materially affects the verdict: ROIC.ai’s ROE and ROIC for MarketAxess do not reconcile to any reported equity figure and must not be used. Its claimed FY2025 ROE of 16.749% on net income of $246.6M implies equity of $1,472.5M against reported equity of $1,145.7M (ending) / $1,388.7M (beginning). Its FY2020 claim of 43.054% implies $695.4M against reported $955.1M / $770.1M. Neither matches. (Its margin data, by contrast, ties exactly.) [FACT — verified]

Rebuilt from the filings (net income per EDGAR ÷ average equity):

[FACT]

FY2025 NOPAT is $251.9M (operating income $341.8M × (1 − 26.3% normalized tax)). On total capital (average equity + debt, cash included) ROIC ≈ 18.3%; on operating capital excluding the cash/regulatory pile, ≈ 35%. Against a WACC of ~8.5% (band 8–9%) [ASSUMPTION — CAPM is unusable here: the reported beta of 0.11355 would imply a ~4.7% cost of equity on an equity with ~33% annualized volatility and a −76% five-year drawdown], that is a spread of +10 to +26 points.

In Greenwald’s diagnostic framing (sustained after-tax ROIC of 15–25%+ = advantages present; 6–8% = absent), MarketAxess sits comfortably inside — not at the bottom edge of — the band where competitive advantages exist. The business is not broken on returns; it is decaying on price. [INTERPRETATION] The concerns are the trend (ROE roughly halved in five years) and that incremental capital is being deployed into a shrinking fee pool.

How profitable is the industry — how many competitors, what barriers to entry?

A highly profitable industry being competed back toward ordinary at speed — with, on Greenwald’s test, no remaining barrier to entry.

Competitor count: you can count the serious institutional-credit venues on one hand (MarketAxess, Tradeweb, Bloomberg, Trumid, ICE) — which on Greenwald’s heuristic suggests barriers. The share test overrides the count.

Greenwald’s decisive test is market-share stability: >5pp of movement over 5–8 years = no barriers; <2pp = formidable barriers.

[FACT — FY2021–FY2025 10-Ks; Q1-2026 10-Q] 2022 was the peak; every product has fallen since. High-yield fails the test outright.

And the single most important derived fact in the engagement: dividing MarketAxess’s disclosed share of the total market by its disclosed electronic penetration of that market yields its share of the electronic market — the arena where it actually competes. Both series are defined against the same denominator in every 10-K, verbatim (penetration is “the level of electronic trading as a percentage of all means of trading”; share is “our estimated market share of total U.S. high-grade corporate bond volume”). The division is valid. [FACT — basis verified across FY2021–FY2025 10-Ks]

MarketAxess’s share of the electronic US high-grade market halved during the largest electronification wave in the product’s history — the very tailwind it sells to investors. The tailwind arrived and competitors captured all of it. [FACT — derived from MKTX’s own continuously-disclosed series; independently corroborated: Morningstar estimates ~33% on a HG+HY combined basis, and rebuilding our derivation on that same basis gives 32.2% — two methods agree within one point]

A necessary caveat: the equivalent high-yield derivation (~76% → ~42%) is not reliable and is not relied upon — MarketAxess revised HY penetration 20%→30% in one year then froze it at “approximately 30.0%” for four years. That is un-updated boilerplate, not a measured series. [ASSUMPTION/OPEN QUESTION]

Barriers to entry — named in Greenwald’s taxonomy: MarketAxess has economies of scale in a single protocol (Open Trading — anonymous, all-to-all, odd-lot credit) essentially without customer captivity.

• Supply/cost advantage: absent. No proprietary technology of consequence; TRACE is public; EMS integrations and index partnerships (MSCI, FTSE Russell) are non-exclusive and undifferentiated.
• Demand/captivity: absent, because of multi-homing. The buy-side is already on MarketAxess and Tradeweb and Bloomberg and Trumid, reachable through the same EMS.
• Economies of scale + captivity: the scale is real; the captivity is gone, and per Greenwald scale without captivity is not a barrier — entrants reach incumbent scale because customers are equally available to all. Trumid is the existence proof.

The tell that settles it: counterparties grew 1,700 → 1,800 (+6%) in four years while share collapsed. A large multi-homed network is a directory, not a moat. [INTERPRETATION, on Greenwald’s framework]

Government protection: none. Regulation (best-execution obligation) works against the moat by legally requiring clients to shop each inquiry.

Can the business be easily understood?

Yes — unusually so, and this is a genuine strength of the analysis rather than of the business. MarketAxess earns a commission per million dollars of face value traded. Value reduces cleanly to a three-term identity: Revenue = volume × share × fee-per-million. There are no reinsurance triangles, no loan books, no percentage-of-completion, no deferred revenue of consequence. 868 employees generate $846.3M of revenue (~$975,000 per head) at 59% gross and 40% operating margins. [FACT — FY2025 10-K]

The complexity is not in the business; it is in one term. Volume is disclosed monthly and is growing. Share is disclosed (partially — see the withdrawal below) and is falling. Fee-per-million is disclosed, is falling, and nobody outside the company — including, on the evidence, the sell-side — can date the end of the decline. The entire investment reduces to that one number, and it reports monthly. [INTERPRETATION]

The one genuine opacity is self-inflicted: MarketAxess operates as a single reporting unit (10-Q Note 2), so no segment profitability is reported at all. Nobody outside can say what US credit earns versus EM, what LiquidityEdge earned, or what Pragma earns. [FACT] This is why the memo considered and rejected a sum-of-the-parts: it is not constructible without inventing margins, and >84% of value sits in the credit franchise regardless.

Can it be undermined by foreign low-cost labor?

No — and the question does not map to the business model. Stating it plainly rather than forcing an answer is the honest response.

MarketAxess is a US-regulated electronic venue whose competitive position rests on liquidity concentration and regulatory registration, not on labor arbitrage. Labor is 29.4% of revenue ($248.5M of employee compensation on $846.3M), it is high-skill (engineering, sales, market structure), and 314 of 868 employees are already international. A lower-cost competitor cannot enter by paying people less; it enters by attracting liquidity. [FACT / INTERPRETATION]

The correct sector analog — and it is not a hypothetical, it is exactly what is happening — is vendor/venue aggregation and dealer-balance-sheet substitution:

1. Vendor aggregation. EMS/OMS providers (Aladdin, Charles River, and the named “EMS and OMS Providers… offer aggregation of trading venue liquidity” category in the 10-K) supply the abstraction layer that Reg NMS supplied in equities. [FACT — FY2025 10-K] They collapse single-venue captivity to near zero. Nobody switches away from MarketAxess; they simply route the next ticket elsewhere. There is no switch to make, so there is no switching cost. [INTERPRETATION]
2. Dealer-balance-sheet substitution. The protocols taking share — portfolio trading, blocks, dealer matching sessions — are bilateral balance-sheet transactions. A portfolio trade needs one dealer who can warehouse a ~2,000-line basket and hedge its duration cheaply, not 1,800 anonymous counterparties. [INTERPRETATION]

This is the disintermediation risk that actually exists, and it is more dangerous than the low-cost-labor version would be: the “cheap entrant” is the dealer’s own balance sheet, it cannot be out-invested, and MarketAxess is obliged to offer the very protocols that carry it. [INTERPRETATION]

Do brands matter?

Somewhat, and less than they used to — and the brand is not where the value is.

MarketAxess is a genuinely well-known institutional brand: it invented all-to-all credit trading, it is the reference for Composite+ pricing data, and it is on essentially every institutional credit desk. But a venue brand is not a demand advantage in Greenwald’s sense — it does not create captivity. Traders do not choose a venue out of habit or brand affinity; best-execution obligations legally require them to shop each inquiry, and the EMS presents all venues side by side. Brand gets you onto the desk; liquidity gets you the ticket. MarketAxess is still on every desk and is getting fewer tickets. [INTERPRETATION]

The one place brand demonstrably still pays is Open Trading’s reputation as a stress-window liquidity backstop — penetration spiked in March 2020 and in 2023. That is a real asset with a real option value, and it is not enough: Open Trading’s commissions are shrinking in dollars in the calm periods that constitute most of the calendar. [FACT/INTERPRETATION]

What is the nature of competition?

A price war that nobody calls a price war, fought protocol by protocol rather than product by product.

The single most important analytical move in this engagement is to stop thinking of MarketAxess as competing for products and start seeing it compete for protocols. A corporate bond can be traded five materially different ways, each with different economics, each favoring a different kind of venue:

[FACT for the fee/volume characterizations — FY2025 10-K; INTERPRETATION for the framing]

Competition takes the form of protocol substitution, and protocol substitution is economically indistinguishable from a price cut. The 10-K says so in its own words: portfolio trading is “generally provided under a lower-fee structure” and is used “in lieu of more established trading protocols designed to generate price competition on individual bonds.” “In lieu of” is substitution. [FACT]

This is a prisoner’s dilemma with no exit, in Greenwald’s game-theoretic sense. Offering portfolio trading costs price; refusing it costs share. Neither MarketAxess nor Tradeweb can unilaterally decline, and the structural feature that makes cooperation impossible is Bloomberg: its venue exists to defend a ~$30k/yr Terminal subscription, not to earn a per-million toll, so it can price at or near zero indefinitely and will never exit. [INTERPRETATION] There is no most-favored-nation clause, no advance price announcement, no meet-the-competition mechanism — none of the structural adjustments that tame a prisoner’s dilemma is available on a per-ticket basis to a best-execution-obliged buy side.

The head-to-head, with numbers:

• vs. Tradeweb — the direct competitor and the share gainer: record 22% share, institutional RFQ ADV +30%, revenue ~+18%, and a ~⅓ fixed-fee cushion MarketAxess lacks. TW is compressing its own fee capture faster (−14.3% cash credit) — it is buying the share, but it is winning it.
• vs. Bloomberg — the unmodeled incumbent rail; a permanent price ceiling.
• vs. Trumid — the existence proof that captivity is gone.
• vs. ICE — adjacent, sub-scale, but a persistent bid for the same flow.

The rates asymmetry is real, but the naive version is false. The common claim is that Tradeweb wins credit because it auto-spots the Treasury hedge and MarketAxess cannot. In fact MarketAxess does offer US Treasury hedging. The real mechanism is liquidity depth, not feature availability — and the evidence for unfixability is MarketAxess’s own failed attempt: government-bond share 2.6% → 3.5% (2022 peak) → 2.4% → 2.4%. It tried, invested ~$150M, and gave the gains back. [FACT / INTERPRETATION]

Customers’ switching costs?

Effectively zero — and this is the load-bearing finding of the competitive analysis.

There is no switch to make, so there is no switching cost. The buy-side is simultaneously connected to MarketAxess, Tradeweb, Bloomberg and Trumid — all reachable through the same EMS/OMS layer. Nobody switches; they route the next ticket elsewhere. [INTERPRETATION]

Three forces destroy captivity here, and all three run the wrong way for MarketAxess:

1. Multi-homing is the norm, not the exception — connectivity is already installed on every venue.
2. Vendor aggregation supplied the abstraction layer — the 10-K names EMS/OMS aggregation as a competitive category. [FACT] (A corollary worth noting: a bond best-execution mandate — often cited as a tail risk — would be an anticlimax. The moat eroded without one.)
3. Best-execution obligations legally require clients to shop each inquiry, destroying habit — the one mechanism (per Greenwald) that could sustain captivity in a frequent, repeated purchase.

Test it against the numbers rather than the story: counterparties grew 1,700 → 1,800 (+6%) over four years while share collapsed and Open Trading’s revenue shrank in absolute dollars. If switching costs existed, growing the network by 6% would not be compatible with halving electronic share. The network is a directory. [FACT / INTERPRETATION]

The honest counterweight, and it is real: there is a genuine liquidity externality in Open Trading — a deeper anonymous pool does deliver better prices, and price improvement is a measurable customer benefit. The problem is that the benefit is shrinking (price improvement per dollar traded roughly halved, ~10.1 → ~4.9bps, partly cyclically) and the protocol carrying it is losing ground to protocols where the externality does not apply. The moat is real. It is attached to the wrong protocol. [INTERPRETATION]

Financial Condition & Balance Sheet

Assets not fully recognized on the balance sheet?

Yes, and they are the whole business — as is normal for a software-economics venue, and it cuts both ways.

Unrecognized assets (the franchise):

• The Open Trading network itself — ~1,800 counterparties, ~200 dealers, ~2,000 firms. Built over a decade from 2013 and carried at essentially nothing. This is the genuine unrecognized asset, and its economic value is currently declining: Open Trading commissions fell $178.5M → $175.6M while volume rose ~25%. [FACT]
• The Composite+ pricing dataset — a real proprietary data asset (MarketAxess claims visibility on ~80% more bonds than public sources). But it monetizes at $53.2M growing ~3.6% organically, which is the strongest available evidence that it is either less differentiated than claimed or badly under-monetized. [OPEN QUESTION — flagged in the memo]
• Internally-developed technology — partly capitalized ($112.4M net book value), mostly expensed.

In Greenwald’s asset-value/EPV framing, the comparison is instructive. Reproduction cost of the tangible asset base is modest — tangible book is $801.4M, and much of that is cash. Earnings power value on normalized owner FCF of $293.2M at an 8.5% WACC is ~$3.45bn — roughly 4.3x tangible book. That gap is the franchise, and Greenwald’s rule is that it must be justified by an identifiable barrier to entry. The competitive analysis concludes the barrier is largely gone — which is precisely why the multiple has compressed toward, though not to, an EPV-only valuation. [INTERPRETATION]

A liability-side item that does not appear as an asset and should be understood as one: the $552.1M of net capital held in excess of the $36.6M required inside regulated broker-dealer subsidiaries is recognized as cash, but it is not free cash — it supports Open Trading matched-principal settlement, and the 10-K notes Open Trading growth “is dependent on the willingness of our customers and counterparties.” The pile is partly a moat input, not a distributable asset. [FACT / INTERPRETATION]

Off-balance-sheet liabilities?

Materially none — and the search was conducted properly rather than assumed.

• No pension, no OPEB, no securitizations, no VIEs of consequence, no earnout structures of size.
• Operating leases are recognized on balance sheet under ASC 842 (post-2019); office space only.
• The only genuine contingent exposure is tax, and it is quantified: the New York State matter is closed (closing agreement 2026-02-18 covering 2015–2023); the New York City exam remains open and widened from 2016–2018 to 2016–2023 between the 10-K and the 10-Q, with $11.2M of unrecognized tax benefits live. [FACT] That is a rounding error against a $4.1bn cap — but note the FY25 experience proves this exposure converts to real cash: the current state and local tax provision jumped to $41.2M from $9.7M. [FACT]
• Matched-principal settlement risk is real but collateralized, regulated, and capitalized against ($552.1M excess net capital). It is the reason the regulatory cash exists. [FACT]
• The revolver ($220M drawn of a $750M commitment at 5.1%, maturing 2026-08-09, amended and restated 2026-02-04) is on balance sheet. [FACT]

The genuine “off-balance-sheet” item is not a liability at all — it is a disclosure gap. Single-reporting-unit status means ~$394M of goodwill and intangibles (34% of book) is tested against the fair value of the entire company, and is therefore effectively never impairable at a $4.1bn market cap. LiquidityEdge (~$150M, 2019) failed — government-bond share went 2.6% → 3.5% → 2.4% → 2.4%, 10-K mentions went 3 → 1 → 0, and it was dropped from the FY2025 acquisition list entirely — and no writedown will ever say so. The clean impairment record is an artifact of segment reporting, not evidence of allocation discipline. [FACT (structure) / INTERPRETATION (consequence)]

How conservative is the accounting?

Genuinely clean — this is the strongest part of the story, and the report says so without hedging.

Four tests, all passed:

1. Stock-based compensation is small. $30.9M in FY2025 = 3.7% of revenue, 10.5% of owner FCF (3.6–4.2% across 2021–2025). A sharp and creditable contrast with platform peers, where 15–30% is routine. Dilution is only ~0.25%/yr.
2. Capitalized software ≈ amortization — no ramp flattering earnings. $53.0M capitalized in FY25 against ~$50.5M of D&A on the pool; net book value moved $107.3M → $112.4M. Steady state. [FACT] (The cash cost has risen from 4.7% to 5.9% of revenue, and is correctly deducted in owner FCF.)
3. Cash conversion is excellent and improving — owner FCF was 119% of net income in FY2025. Net income is not running ahead of cash (see below).
4. The company’s own non-GAAP is honest where it counts. The $7.39 normalized FY25 EPS figure is MarketAxess’s own disclosed reconciliation, not our arithmetic. [FACT]

Three genuine caveats, in descending order of seriousness:

• Single reporting unit ⇒ goodwill effectively never impairable (above). This is the one place the accounting structure conceals rather than reveals.
• “Repositioning” severance now appears in FY2025 and Q1-2026 — it is becoming recurring and should not be freely added back (~$0.14/sh in FY25). Trim normalized EPS toward ~$7.60 if you deduct it. [INTERPRETATION]
• Management’s own FCF definition is flattered by the SBC add-back, as always: company-defined FY25 FCF is $346.9M vs. our owner FCF of $293.2M — the ~$54M wedge is SBC. Use owner FCF.

Where the vendors are wrong, the filings are right. ROIC.ai’s ROE/ROIC reconcile to no reported equity figure; its EV of $6.494bn is built from a stale 12/31/2025 snapshot (price $181.25, 57% above spot) — a 42% error. AZI’s valuation_index percentiles are degenerate for MKTX (P/E, P/B, P/S and composite all return an identical 0.418 with null history) and were not used. The accounting is not the problem here; the data feeds are. [FACT — verified]

How CapEx-hungry is the business?

Barely at all in the traditional sense — and the question needs its sector analog to be answered honestly.

Traditional PP&E capex is immaterial: this is a business with 868 employees, leased offices, and cloud infrastructure. The correct analogs are two, and both are properly captured in owner FCF:

1. Capitalized software — the real “capex.” $53.0M in FY2025 = 5.9% of revenue, up from 4.7% five years ago. It is a genuine, recurring, rising cash cost, and it is why EV/EBITDA flatters this company: EBITDA excludes it. On cash EBITDA the multiple is 10.2x, not 8.7x — the honest figure. [FACT / INTERPRETATION] The favorable read is that capitalization ≈ amortization — this is steady-state maintenance, not a growth ramp being used to flatter earnings.
2. Regulatory net capital — the balance-sheet “capex” nobody models. $552.1M sits in regulated broker-dealer subsidiaries in excess of the $36.6M required, supporting Open Trading matched-principal settlement. It is a permanent, non-distributable capital commitment that scales with the business — economically a capital requirement, not idle cash. It is the single reason the honest EV band is $3.78bn (all cash netted) to $4.11bn (none netted) rather than a point estimate. [FACT / INTERPRETATION]

A Marathon note that cuts for the company: the capital-cycle warning signs on the asset side are absent here. Capex-to-depreciation is roughly 1x; there is no asset-growth binge; the balance sheet is not being expanded into a boom. The asset-growth anomaly does not indict MarketAxess. [INTERPRETATION] The capital cycle is punishing this industry through price, not through capacity — which is exactly why the usual supply-side tells (capex ratios, IPO waves, rising leverage) fail to fire, and why the compression has been invisible to screens for five years. That is an important, non-obvious point: in an asset-light industry, the capital cycle expresses itself in fee compression rather than in overbuilt plant.

Capital Allocation & Management

How much FCF does the business generate, how does management use it, what is the philosophy?

Generation is excellent. Deployment is the weakest link in the entire story, and it converted a decelerating franchise into an actively value-destroying one.

[FACT — rebuilt from cash flow statements] At $115.50 that is a 7.1% owner-FCF yield on a $4,105M market cap. This is real, clean cash — not an accounting artifact.

Uses in FY2025: buybacks $420.0M, dividends ~$111M ($3.12/yr on ~35.5M shares), M&A $82.3M (RFQ-hub). Total deployment exceeded owner FCF, and the gap was funded by drawing the revolver and running down cash. [FACT]

The stated philosophy (per the December-2025 targets and the Q4-25 call) is return-of-capital-led: dividend, then opportunistic buyback, then bolt-on M&A. The revealed philosophy is procyclical: buy the most stock when the stock is most expensive. [INTERPRETATION]

Significant acquisitions recently?

Four in the window, one clear failure, none ever markable:

• LiquidityEdge (2019, ~$150M) — the US Treasuries platform meant to close the rates gap the competitive analysis identifies as structurally decisive. It failed. Government-bond share 2.6% → 3.5% (2022 peak) → 2.4% → 2.4%; 10-K mentions 3 → 1 → 0; dropped from the FY2025 acquisition list entirely. [FACT]
• Regulatory Reporting Hub (Deutsche Börse, 2020) — post-trade; contributes to a line compounding at ~1.4% organically.
• Pragma (2023) — quantitative/algorithmic trading tech; the source of technology services’ $3.0M → $13.9M jump. Bought, not built.
• RFQ-hub (2025-05-09, 90.3% for $82.3M) — ETF/derivatives RFQ. It flatters Q1-2026’s “Other” variable fees and therefore the very inflection the bull case leans on — a fact the bull case must net out. [FACT]

On Greenwald’s M&A test (target must sit in an industry with genuine barriers; synergies must be real, measurable and cost-based; premium must not exceed synergy value): LiquidityEdge fails on the first requirement — MarketAxess bought into US Treasuries, a market where it had no advantage and where Tradeweb’s incumbency was the whole point. The revenue-synergy logic (rates liquidity would pull credit share) is exactly the kind of revenue synergy Greenwald says is almost always illusory. It was illusory. [INTERPRETATION]

And no acquisition will ever be marked to its result: single-reporting-unit status means ~$394M of goodwill/intangibles is effectively never impairable at a $4.1bn cap. [FACT/INTERPRETATION]

Buying back shares?

Yes — $646.2M of it, at a blended price 78% above today’s, and it destroyed $283.9M.

[FACT — 2024/2025 exact from the 10-Ks (the exact years carry $495M of the $646M); 2021–22 estimated]

$283.9M destroyed — ~$8.07 per current share. Every vintage is underwater.

They borrowed to do it. A never-drawn balance sheet was levered $220M to fund a $300M ASR at $171.84 in December 2025. The CFO on the Q4-25 call: “we did take out about $220 million on our revolver… to put a little bit of leverage on it to do that ASR. So our first order of business is going to be to pay that down over time.” [FACT — Q4-2025 earnings call] The debt outlived the value: the ASR is ~33% underwater.

The timing inverts cheapness, and it answers the conviction question — they are pulling back exactly when they should lean in. $0 repurchased in 2023 with ~$100M of authorization idle while the stock traded ~$271; $420M in 2025 at $179; and today, with the stock at $115.50 — the cheapest it has ever been on every own-history metric — ~$205M sits idle while management repays the revolver. Buying high forced abstention when cheap. [FACT for the amounts; INTERPRETATION for the reading]

This is Marathon’s procyclical-management pattern in its purest textbook form, and note the double indictment: the buyback was not even mopping up dilution — dilution is only ~0.25%/yr, and the share count genuinely fell 6.4%. And EPS still fell from $7.85 to $6.64. $646M of buybacks bought approximately sixteen cents of EPS. [FACT/INTERPRETATION]

(One reconciliation worth recording, because the filings appear to disagree and do not: XBRL reports FY25 PaymentsForRepurchaseOfCommonStock of $420.0M (cash) while the equity note reports $360.0M / 1,980,715 shares. They tie to the dollar: $420.0M − $60.0M of the ASR parked in additional paid-in capital pending its 2026-02-04 settlement = $360.0M. The correct full-year figures are $420.0M / 2,340,497 shares / $179.45. [FACT — verified])

Issuing large amounts of new shares to insiders?

No — and this is a genuine, creditable positive that the report states without qualification.

SBC is $30.9M = 3.7% of revenue and 10.5% of owner FCF (3.6–4.2% across 2021–2025). Net dilution is ~0.25%/yr. The share count fell 6.4%, and stands at 35,539,453 per the Q1-2026 10-Q cover (2026-05-04). [FACT] Against platform and market-structure peers, where 15–30% of revenue in SBC is routine, this is exceptional discipline.

The credit is real, and it indicts the record rather than redeeming it. There was almost no dilution to mop up — so the $646M was not defensive share-count maintenance. It was a deliberate, discretionary deployment of $646M into a decaying franchise at an average price 78% above today’s. [INTERPRETATION]

Compensation policy of directors/management?

The plan does not measure anything that has gone wrong, and that is not a rhetorical flourish — it is a literal reading of the proxy.

There is no EPS metric. No ROIC. No ROE. No TSR — absolute or relative — anywhere in the plan. [FACT — DEF 14A 2025 & 2026]

• Annual bonus: adjusted operating income (75% for the CEO, raised from 60%) + individual/strategic (25%).
• Long-term: 50% PSUs on US credit market share / revenue growth ex-US credit / operating margin; 50% time-vested with no performance condition at all.

Three structural defects, each independently disqualifying:

1. The bonus grid has no threshold. It is linear from $0 adjusted operating income = 0%, paying 50% at half of target profit. FY2024 missed by 3.9% → 96% funding; FY2025 missed by 5.8% → 94% funding. CEO Chris Concannon’s bonus was $1,525,000 in both years — identical to the dollar — while missing target in both. [FACT]
2. The metric is defined circularly. Adjusted operating income is “operating income before… the impact of cash incentives” — measured before the bonus it funds.
3. The goalposts ratchet off the company’s own prior-year result — “set in line with the Company’s 2025 results.” [FACT] This is precisely how the operating-margin metric funded at 78% while margin compressed from 54% to 40%: reset the bar to last year’s outcome and any outcome clears it.

Pay-versus-performance says the rest: FY2025 TSR of $33.56 against a peer group at $194.20; CEO pay rose 22% to $7,101,027. Say-on-pay support rose from 94% to 98%. [FACT — DEF 14A 2026] (A trap for future readers: the 2026 proxy silently rebases the TSR series; do not mix it with the 2025 proxy’s.)

The honest counterweight, and it is real: the PSUs did fail — paying 43%/45%, with the share metric at 0% in 2024 and share-metric payouts written down 27.6%. [FACT] The plan is not a rubber stamp, and the company’s own compensation machinery has registered the share loss even as its MD&A calls the runway long. But a 43% payout is a high floor against a −66% five-year stock, and half the long-term award vests on time alone.

On Marathon’s test — is pay linked to size/growth or to returns/cash flow? — the answer is neither: it is linked to a profit level measured before the bonus, against a bar reset to last year. A decade of fee-per-million erosion, a halving of ROE and a −79% stock have cost the CEO nothing. [INTERPRETATION]

Motivations of management?

The insider record answers this more honestly than the proxy does, and the answer is unflattering.

Across the entire 60-month Form 4 corpus (430 deduped transactions, re-parsed from 225 raw ownership XMLs) there are exactly two open-market purchase events, ever — 5,270 shares, ~$1.29M — and both predate the decline:

Zero purchases dated 2024-01-01 or later — across the entire slide from ~$270 to the $109.09 five-year low to $115.50. Sell:buy ratio ~30:1 (~$39.2M sold). No insider has ever bought a share below $200. [FACT — re-derived from raw XML]

The datum bites because of the 2023 baseline. Concannon proved he will spend ~$1M of his own money when he believes the stock is mispriced — at $238. He has not repeated it at $170, $146, $116, or $109. That is a revealed preference, not a constraint. [INTERPRETATION]

Founder Rick McVey’s tell: sold 50,000 shares at ~$270.06 = $13,503,229 on 2024-11-12/14, six days after the Q3-24 print — discretionary, explicitly not 10b5-1-planned (aff10b5One=false), and notable because his 2021 sale was plan-designated, so he knows the difference. He retains 562,029 shares, so he is not exiting — the fair reading is diversification, not abandonment. But $13.5M sold discretionarily at $270 against $0 bought at $109–170 is the asymmetry. [FACT / INTERPRETATION] (Caveat: only ~24.3% of sales carry a 10b5-1 flag, but the tag did not exist before 2023, so the plan/discretionary split is not reliably computable across the full window.)

What management does appear genuinely motivated by — and this is the report’s most sympathetic reading of them — is fixing the structural problem rather than denying it. The 2023 CEO succession was orderly and four-years-telegraphed (Concannon had been COO/director since January 2019; the 8-K states expressly there were “no arrangements or understandings”), and Concannon’s background is pure electronic market structure (ex-Cboe President/COO, ex-Bats CEO, ex-Virtu/Nasdaq/Instinet). The Virtu thread is the tell: Concannon is ex-Virtu, the interim principal accounting officer came from Virtu, and in January 2026 the board seated Virtu’s sitting CEO Douglas Cifu. A board stacking itself with electronic-market-structure operators reads as one that knows it faces a structural problem, not a cyclical one. [FACT for the appointments; INTERPRETATION for the reading] They are not asleep. They are paid on a metric that does not measure the thing they are awake to.

Valuation & Market Data

Is the stock an ADR, MLP, or K-1 issuer?

No — none of the three. MarketAxess Holdings Inc. is a plain US Delaware C-corporation listed on NASDAQ, taxed at the corporate level, distributing qualified dividends reported on Form 1099-DIV. There is no K-1, no UBTI, no ADR fee, no foreign withholding, no Up-C structure, no dual-class share class, and no tracking stock. [FACT — FY2025 10-K]

There is no structural friction of any kind here — no complication of the sort that would justify or explain a valuation discount, and none that constrains who can own it. (Notable only by contrast: the factor-similarity screen returns low-beta defensive ADRs and regulated utilities as MKTX’s nearest factor neighbours — which is a statement about how the equity now trades, not about its structure.)

Dividend policy?

Well covered, and one of the few unambiguously sound elements of the capital-return program.

• $0.78/quarter = $3.12/year — a 2.7% yield at $115.50.
• Payout is 40% on normalized EPS of ~$7.75 and ~38% of owner FCF ($111M of $293.2M). [FACT]
• The dividend has been raised steadily and has never been cut.

The coverage is genuine and is not at risk in any scenario in this report — even the bear case (FY2031 owner FCF ~$205M, EPS ~$5.54) covers $3.12 comfortably. [INTERPRETATION] The dividend is the part of the capital-return program that worked; the buyback is the part that did not.

How profitable is the business?

(Answered in full under Business Quality & Competitive Moat above; the valuation-relevant summary:) Operating margin 40.4%; ROE 19.5% (rebuilt — the vendor figures are wrong); ROIC ~18% on total capital / ~35% on operating capital against a WACC of ~8.5% — a spread of +10 to +26 points. Owner FCF $293.2M = 119% of net income. [FACT, save the WACC — ASSUMPTION]

The trend is the story, not the level: ROE 34.7% → 19.5% and operating margin 54.4% → 40.4% in five years, with operating income lower than in 2020. The business remains well above its cost of capital and is decaying toward it. [FACT/INTERPRETATION]

Own-history valuation context, hand-built because the vendor percentiles are degenerate: across twelve years, the cheapest MarketAxess ever traded — the intraday-low P/E of its cheapest calendar year (2014) — was 23.3x. Today it trades at 13.7x GAAP / 14.9x normalized: roughly 41% below the cheapest point in its entire public multiple history. The same holds on sales (4.71x vs. a twelve-year minimum-of-lows of 6.64x) and EV/EBITDA (8.7x vs. ~11.6x). [FACT] This is context, not a conclusion: those 2014–2021 multiples were paid for a franchise with ~54% margins, ~35% ROE and a rising fee rate. The de-rating is not evidence of error; a lower multiple is the correct response to lower growth. The only question is how much lower. [INTERPRETATION]

Is net income diverging from cash from operations?

Yes — and it diverges in the favorable direction, which is the opposite of the usual red flag and is worth stating plainly.

Owner FCF was 119% of net income in FY2025, and the ratio has improved monotonically: 79% (2021) → 83% → 98% → 109% → 119%. [FACT — rebuilt from cash flow statements] In Marathon’s cash-conversion diagnostic, a widening gap between reported earnings and free cash flow is the warning sign of a late-cycle capital binge. MarketAxess shows the reverse: cash conversion is improving as earnings stagnate. [INTERPRETATION]

The wedge is fully explained and benign: D&A ($76.7M, inflated by acquired-intangible amortization from Pragma/RFQ-hub) plus SBC ($30.9M) exceed the cash cost of capitalized software ($53.0M) and capex. There is no receivables build, no revenue-recognition stretch, no capitalization ramp — capitalization ≈ amortization. [FACT]

The one divergence that matters is not between income and cash. It is between volume and revenue. FY2025 crystallizes the entire model in one line: credit trading volume +10.0%, credit variable transaction fees +$8.6 million. MarketAxess moved a tenth more bonds and got almost nothing for it. That is the divergence to underwrite, and it is not an accounting question. [FACT — FY2025 10-K MD&A; INTERPRETATION for the framing]

(A necessary caution on the reported figures: normalized run-rate EPS is ~$7.75 — not the $8.45 the screen shows, nor the $6.64 FY25 GAAP prints. And the correct live enterprise value is ~$3.78bn on 35,539,453 shares — not the $6.494bn the vendor feed reports from a stale 12/31/2025 snapshot, a 42% error. Both screens are wrong before any multiple is computed. [FACT])

Risks & Downside

What factors would cause the stock to decline?

One factor, wearing several costumes — and the concentration is the defining feature of the risk profile.

In descending order of importance:

1. Fee-per-million decay continues or accelerates. (High likelihood / High impact.) The master risk; every other risk is downstream. Four consecutive annual declines, accelerating to −7.6% in FY25, −5.0% in Q1-26; cause migrated from cyclical duration to structural mix. Sensitivity: each 1 point of annual fee decay changes FY2031 operating income by ~$50M (~13%) — roughly 2x volume’s leverage and 4x opex’s. Fee per million is not one of several drivers; it is the driver. [FACT — from the scenario model]
2. Continued share loss in US credit. (High / High.) HG 21.0% → 17.1%; HY 15.2% → 12.2%; electronic HG share 60.0% → 30.7%; Tradeweb at a record 22% and gaining.
3. Open Trading — the moat protocol — keeps shrinking in dollars. (Med-High / High.) $178.5M → $175.6M while credit volume +25%; stalled at 35–37% of eligible volume for four years.
4. Protocol substitution proves permanent rather than a bounded transition. (Med / High.) The §6.2 bounding test implies the base book itself repriced — the single most load-bearing open question in the engagement.
5. Volume cyclicality (Med / Med-High) — ~71% of revenue is a volume lottery with no subscription cushion; FY25 was rescued by TRACE volumes +8.0%. A volume downturn on top of fee decay is the compound scenario nobody is modelling.
6. Capital misallocation continues (Med-High / Med) — the comp plan measures none of the above.
7. FINRA’s April-2026 affiliate back-to-back TRACE proposal (Med / Med) — could restate every share figure in this report; Open Trading is matched-principal and may inflate both numerator and denominator. Direction unknown. [OPEN QUESTION — a genuine unquantified risk to the entire share dataset the bear case rests on]
8. Technology cost growth (Med / Med) — technology & communications +84.3% since 2021, the fastest opex line; cloud costs do not decline on command.
9. NYC tax exam (Med / Low-Med) — $11.2M live. Immaterial to the thesis.

And a warning the price action itself supplies: $115.50 is not a floor. Under the bear scenario (−8% fee decay, +8% volume, margin to 28.7%, owner FCF compounding at −6.9%/yr), the current price cannot be justified at any discount rate above ~4% — i.e. below the risk-free rate. The bear case is not priced in. [INTERPRETATION — from the scenario model] Note also that the tape rejects good news: the December targets/$505M authorization (+4.9%) and the Q4-25 print (+5.5%) each fully round-tripped within weeks, and across the −36.9% slide to the June low no single session fell more than 4.7% — orderly institutional distribution, not a capitulation flush. [FACT]

The factor read forecloses the comforting “abandoned value name” story. MKTX carries a +0.27 Value loading into a Value factor that returned +13.8% over the past year (z = +1.64) — and lost 45.4% anyway. With R² of 14.4% and ~86% of variance idiosyncratic, there is no hostile factor regime to wait out — the market is repricing this company’s earnings stream. Momentum is negative in all four nested models and Quality is zeroed in every one: cheap, falling, with no quality signature — the value-trap signature in factor space. [FACT for the loadings; INTERPRETATION, regime-caveated]

Risk of a catastrophic loss?

Very low likelihood, and structurally so — this is the strongest defensive feature of the business.

The risks that would ordinarily dominate a financial-sector memo are absent:

• No leverage risk. Net cash +$328.1M even after the ASR; $529.9M of revolver capacity available; the August-2026 revolver maturity is housekeeping.
• No liquidity risk. Owner FCF $293.2M/yr; the dividend consumes 38% of it.
• No credit risk. MarketAxess does not lend, does not warehouse inventory, and does not take directional risk.
• No customer concentration. ~2,000 firms, ~1,800 Open Trading counterparties, ~200 dealers — genuinely diversified.

The one genuine catastrophic-tail channel is matched-principal settlement in Open Trading — MarketAxess stands between both sides of an all-to-all trade. It is real, but it is collateralized, regulated, and capitalized against: $552.1M of net capital in excess of the $36.6M required. A settlement failure in a systemic credit dislocation is the theoretical mechanism; it has not been tested to breaking in March 2020 or in 2023, both genuine stress windows in which Open Trading penetration actually rose. [FACT / INTERPRETATION]

A cyber/operational-outage event is the other tail — a venue that cannot trade for a week loses tickets it may not get back on a multi-homed desk where the competing venue is one click away. This is the tail where zero switching costs cut hardest against the company. [INTERPRETATION] No such event is in the record.

The honest framing of the downside is not catastrophe. It is slow disappointment. “MarketAxess will still be here in ten years earning something. The question is only what ‘something’ is.”

Chance of a total loss?

Effectively nil, and it is worth being precise about why rather than merely asserting it.

A permanent total loss of capital requires one of: leverage that cannot be serviced (MarketAxess is net cash); a business that stops generating cash (owner FCF is $293.2M and 119% of net income); fraud (accounting is clean, SBC is 3.7% of revenue, capitalization ≈ amortization, the company’s own non-GAAP reconciliation is honest, and the vendor feeds — not the filings — are what is wrong); or an existential regulatory event (none identified; the FINRA proposal restates measurement, not legality).

None applies. Revenue has never declined. Volumes have never declined. Returns exceed the cost of capital by 10–26 points. This is not a melting ice cube, and the memo is explicit that it is not a short.

The correct downside frame is a terminal multiple risk, not a terminal value risk. The bear case is not zero — it is $5.54 of FY2031 EPS on a 28.7% margin, at which $115.50 is 20.8x forward earnings for a shrinking business. The loss is a de-rating of a permanently profitable company, not an impairment of it. [INTERPRETATION — from the scenario model] The distinction matters: the risk here is opportunity cost and multiple compression compounding over years, not a hole in the balance sheet. Nothing about this business forces a resolution date, which is precisely what makes the decay so hard to underwrite — an amortizing annuity does not default; it just pays you less each year.

Recent News & Events

Has the business environment changed recently?

Yes — and the most important change is not an event but a reclassification, which is why it has been easy to miss.

Fee-per-million erosion moved, on management’s own telling, from a cyclical duration effect to a structural protocol-mix effect. The MD&A language migrated from bond duration and dealer migration to fixed distribution fees (FY2022–23) → “product and protocol mix-shift” (FY2024) → “mainly due to protocol mix-shift reflecting increased portfolio trading” (FY2025). [FACT — 10-K MD&A each year] And Q1-2026 confirmed it: duration flipped to a tailwind and fee-per-million fell 5.0% anyway. The cyclical alibi has expired. [FACT] The explanation moved from “the rate cycle did this to us” to “our own mix is degrading.” One reverses; the other does not. [INTERPRETATION]

Other changes in the window, in rough order of importance:

• Portfolio trading became a prisoner’s dilemma with no exit — the 10-K concedes offering PT costs price while refusing it costs share; blocks and dealer matching sessions compound it. [FACT — FY2025 10-K risk factors]
• Capital-structure change (Dec 2025) — medium-term targets announced; buyback authorization raised to $505M; a $300M ASR at $171.84, funded partly by the first meaningful $220M revolver draw in the company’s history. [FACT — 8-K 2025-12-09] The stock rose 4.9% that day and fully retraced within weeks. Net cash has roughly halved in five quarters: +$709.7M → +$328.1M.
• Sell-side capitulation (mid-2026), in real time — Rothschild Buy→Neutral, PT $189→$134 (2026-06-11); Goldman $130 (06-30); Morgan Stanley $129 (07-10); Piper $128 (07-15). Price targets have converged at or below spot and the cuts are still arriving. [FACT — news feed] (Cited as evidence of the consensus shift; not adopted as a view.)
• The Q1-2026 print — the first inflection in five years, and the bull case’s best evidence. Revenue +11.9%, operating income +14.2%, op margin +0.87pt — the first YoY margin expansion in the series. [FACT — Q1-2026 10-Q] It is also flattered by the RFQ-hub acquisition and an easy comp, and fee-per-million still fell 5.0%. One quarter is not a trend. [INTERPRETATION]
• Regulatory (April 2026) — the FINRA proposal on affiliate back-to-back TRACE reporting could restate every share figure. Direction unknown. [OPEN QUESTION]
• The stock itself — −79.1% from the 2020 high, a five-year low of $109.09 on 2026-06-25, down in five of the last six calendar years, −36% YTD. [FACT]

On balance the environment has changed, and not in MarketAxess’s favor. [INTERPRETATION]

Significant acquisitions?

RFQ-hub (2025-05-09, 90.3% for $82.3M) — ETF/derivatives RFQ, and a real contributor to the Q1-2026 inflection the bull case leans on; it must be netted out before Q1-26 is read as organic. Pragma (2023) continues to carry technology services ($3.0M → $13.9M). [FACT] No transformative deal, no auction premium, nothing that changes the credit franchise’s trajectory. The M&A is bolt-on adjacency — buying what could not be built — and none of it addresses the fee rate. [INTERPRETATION]

Change in accounting policies?

No policy change of substance — the accounting is clean and the recent noise is tax, not accounting.

The one item that looks like an accounting event and is not: the $54.9M provision for unrecognized tax benefits booked in Q1-2025 after a New York State tax court ruled in a matter MarketAxess was not a party to, reversing a lower court ruling that had supported its historical filing position. That drove an 84.3% effective tax rate on flat pre-tax income ($96.15M vs. $96.72M). A ~$31.3M reserve release then landed in Q4-2025, flattering that quarter’s $2.50 GAAP EPS by ~$0.82. [FACT — verified across four documents; UTB balance $56.4M → $22.4M]

This is a valuation-relevant trap, not an accounting-quality problem: FY2025 GAAP is depressed by the charge and TTM GAAP is flattered by its release. Neither is the run-rate. The New York State matter is now closed (closing agreement 2026-02-18 covering 2015–2023); the New York City exam remains open and widened from 2016–2018 to 2016–2023, with $11.2M live. [FACT] The charge was real cash — the FY25 current state and local provision jumped to $41.2M from $9.7M.

One structural disclosure change deserves the section’s real attention, and we state the defensible version rather than the tempting one. The FY2023 10-K’s market-share table went from seven rows to five: “Composite Corporate Bond,” emerging-markets debt and Eurobonds were deleted simultaneously and never restored. [FACT — verified by diffing table row labels across FY2021–FY2025 10-Ks; a grep count does not reveal this] “Composite Corporate Bond” — the company’s own headline metric — appeared 4× in the FY2022 10-K, 1× in FY2023, and 0× in FY2024 and FY2025. MarketAxess still reports EM and Eurobond volumes (+29.8%/+20.4% in Q1-26) but never the share.

The motive is NOT established, and the obvious accusation is wrong: the very filing that deleted the rows attributed rising Eurobond volumes to “increases in estimated market volumes and our estimated market share” — MarketAxess said share was rising in the deletion year. [FACT] The defensible finding is narrower and still meaningful: the withdrawal removed verifiability precisely where the competitive question is hardest, and the attribution language shifted — share was cited as a volume driver in 2021–2023 and never once in 2024–2025. [FACT / INTERPRETATION] (A circulating third-party “18.2% → 15.7%” Eurobond-decay figure does not reconcile to the 10-K basis (12.1% → 15.4%, rising) and uses a different denominator; this note cites the disclosure deletion, not the rate.)

Recent changes — new markets, facilities, management?

New markets: no new asset class of consequence. The genuine expansion is within existing products — EM + Eurobonds now 40.5% of credit ADV (from 32.7% in 2020), growing ~20% — plus adjacencies bought rather than built (Pragma: equities/FX algos; RFQ-hub: ETF/derivatives RFQ). Municipals’ addressable ADV grew +45.5% YoY, the fastest-growing pool disclosed. [FACT]

Facilities: nothing material. 868 employees (554 US, 314 international) in leased offices. No grandiose headquarters — Marathon’s classic late-cycle warning sign is absent here. [FACT / INTERPRETATION]

Management and board — the most interesting change, and the one that reads as a tell:

• CEO transition (2023): the board elected Chris Concannon CEO on 2023-01-03 (announced 01-09, effective 2023-04-03); founder Rick McVey became Executive Chairman. Concannon had been COO/director since January 2019; the 8-K states expressly there were “no arrangements or understandings.” A four-year-telegraphed internal succession — no rupture. [FACT — 8-K 2023-01-09]
• CFO exit 2023-11-09; CIO Naineshkumar Panchal exit effective 2026-04-01, with a CTO search concluded. [FACT — 8-K 2026-02-16]
• Board additions (8-K 2026-01-22): Douglas Cifu and Kenneth Schiciano, effective 2026-03-01. [FACT]

The Virtu thread is worth flagging as the single most informative personnel fact in the file: Concannon is ex-Virtu, the interim principal accounting officer came from Virtu, and in January 2026 the board seated Virtu’s sitting CEO, Douglas Cifu. A board stacking itself with electronic-market-structure operators reads as one that knows it faces a structural problem, not a cyclical one. [FACT for the appointments; INTERPRETATION for the reading] That is simultaneously the most reassuring and the most damning thing in this section: they see it clearly, and they are paid on a metric that cannot see it at all.

Supplemental to the MKTX research memo of 2026-07-17. No recommendation and no price target appear in this appendix. All non-obvious facts trace to the primary sources listed in Appendix B.

APPENDIX B — Source Appendix

MarketAxess Holdings Inc. (NASDAQ: MKTX) · CIK 0001278021 · Report date 2026-07-17 · Price $115.50 (close 2026-07-16)

Sources are listed primary first. Every non-obvious fact in the memo traces to an entry below. Where a fact rests only on a secondary or third-party source, it is marked [SECONDARY-ONLY]. Third-party aggregated data is reconciled to the filings; where a vendor and a filing disagree, the filing wins and the discrepancy is documented in the quantitative-data-providers section below.

1. Primary — SEC filings (CIK 0001278021)

The full trailing 60-month SEC corpus was reviewed. All URLs accessed 2026-07-17.

1.1 Annual reports (Form 10-K) — the spine of the report

1.2 Quarterly reports (Form 10-Q) — relied on

Earlier 10-Qs (2021-2024) were reviewed for the trend build; they are not individually load-bearing.

1.3 Current reports (Form 8-K) — material events

The full 8-K corpus was reviewed; the material-event timeline was built from it.

1.4 Proxy statements (DEF 14A) — compensation and incentives

1.5 Insider filings (Forms 3/4/5) — the insider read

Corpus: the full Form 3/4/5 filing set on EDGAR, comprising 206 unique filings / 430 transactions, period 2021-08-01 → 2026-07-10. All conclusions were re-derived directly from the raw ownership XMLs — see the data-integrity note in the quantitative-data-providers section below.

Individually load-bearing filings:

All Form 4 URLs follow https://www.sec.gov/Archives/edgar/data/1278021/{accession-no-dashes}/ — e.g. the McVey filing at sec.gov.

1.6 Competitor primary filings

• Tradeweb Markets Inc. (NASDAQ: TW) — Q4-2025 / FY2025 earnings materials and monthly volume disclosure. Source of TW’s own disclosed figures: record 22% market share; credit revenue $488.0M (+6.3%); institutional RFQ ADV +30%; cash-credit FPM $126.83 (−14.3% YoY); blended fee capture −10.2%. TW filings: sec.gov EDGAR CIK 0001758730. Caveat: TW’s “Cash Credit” and MKTX’s “credit” are not the same category — MKTX’s includes EM, eurobonds and munis; TW’s strips out derivatives, China and EP flow. The FPM levels are not comparable across the two issuers; only the trend is.

2. Primary — earnings-call transcripts

Company earnings-call transcripts, read in full. Management commentary is treated as a hypothesis requiring external validation, never as evidence — every claim sourced here is cross-checked against a filing.

Coverage note: the transcript set is earnings-call-centric. Non-earnings events (conference presentations, strategy days) were not separately sourced; this is a known gap.

3. Quantitative data providers — third-party, reconciled to filings

All are third-party aggregated data, not primary. For a US filer, EDGAR and the 10-K/10-Q remain primary. Three vendor errors were material enough to change conclusions and are documented below.

3.1 SEC EDGAR XBRL (authoritative; accessed 2026-07-17)

Company Facts / Concept API, CIK 0001278021. Used for numeric spot-checks independent of the HTML filings:

• us-gaap:EarningsPerShareDiluted — FY24 $7.28; FY25 $6.64; Q1-25 $0.40; 9M-25 $4.14; Q1-26 $2.20
• us-gaap:PaymentsForRepurchaseOfCommonStock — FY2025 $420,015,000; FY2024 $75,474,000 — the figure that resolved the $420M-vs-$360M question
• Filing enumeration for the 60-month corpus and the Form 4 index

3.2 Price / OHLCV data — third-party daily provider

A third-party daily price/OHLCV data provider (split- and dividend-adjusted), accessed 2026-07-16/17. Supplied the five-year event map, the 52-week range ($109.09–$211.60), EMAs (21/50/200 = 117.12 / 127.01 / 157.38), and the volume checks on the 2026 decline.

3.3 News feed — third-party news aggregator

A third-party financial-news aggregator, accessed 2026-07-17, used for the recent-events timeline and the “Recent News” diligence answer. Any AI-generated importance/sentiment/impact scores are a third-party signal, not evidence; underlying articles were opened and validated against primary sources where material.

[SECONDARY-ONLY] — the June-2026 monthly volume figures. HG estimated share ~17.9% (+10bp YoY), HY 14.9% (+190bp YoY), block ADV +33%, PT ADV +71% rest on a third-party summary of MKTX’s monthly volume report (2026-07-07). No SEC filing carries these figures — MKTX publishes monthly volume statistics as an IR press release, not an 8-K. The underlying primary release is MKTX’s monthly volume report, 2026-07-07, at investor.marketaxess.com. These are single-month figures and are not comparable to the annual (10-K) or quarterly (10-Q) share series.

3.4 Form 4 parse — data-integrity note

Methodological caveat: all insider conclusions were re-derived directly from the raw ownership XMLs. The aff10b5One (10b5-1 plan) element did not exist before 2023 — 2021/2022 filings are absent on the flag, not false. Flag-only yields 10b5-1 = 1.1% of sale dollars; flag OR footnote text yields 24.3%, which is the defensible figure and the one used.

3.5 Third-party fundamentals aggregator — two documented errors

A third-party fundamentals aggregator (ROIC.ai), accessed 2026-07-17 for ratios, enterprise value, valuation multiples and the three statements. Treated as aggregated data reconciled to the filings, not as primary:

• ✅ Margins tie to the filings exactly and are used.
• ❌ ROE and ROIC are WRONG and are not used. Vendor reports ROE 43.05% (2020) → 16.75% (2025) and ROIC 30.33% → 15.56%. Neither endpoint reconciles to any reported equity figure: 16.749% on FY25 net income of $246.6M implies equity of $1,472.5M against reported $1,145.7M (ending) / $1,388.7M (beginning); 43.054% on FY20’s $299.4M implies $695.4M against reported $955.1M / $770.1M. Returns were rebuilt from the filings: ROE 34.7% (2020) → 19.5% (2025) on average equity; ROIC ~18.3% (total capital) to ~35% (operating capital). The direction the vendor reports is right; the levels are not, and the levels are what the moat diagnostic turns on.
• ❌ Enterprise value is STALE by ~42%. Vendor reports EV $6,494M — a 12/31/25 snapshot at $181.25/sh (+57% vs spot) — and $284.9M of debt against an actual $220M revolver. Live rebuild: $4,105M market cap + $220.0M revolver − $377.3M cash − $170.8M investments = $3,777M. Honest band $3.78bn (all cash netted) to $4.11bn (none netted); much of the cash is regulated broker-dealer net capital ($552.1M excess) supporting Open Trading matched-principal settlement and is not free.

3.6 Own-history valuation percentiles — third-party source

A third-party own-history valuation-percentile source, accessed 2026-07-17. Inputs: price $115.50, TTM EPS $8.4513, BVPS $33.64, TTM sales/sh $23.92 → P/E 13.67x, P/B 3.43x, P/S 4.83x.

• ❌ The percentile ranks are DEGENERATE for MKTX and are NOT reported. P/E, P/B, P/S and composite all return the identical 0.418 with null history — not a credible own-history rank. The own-history multiple range was hand-built instead from a 12-year year-end multiple series (which reconciles: FY25 P/E 27.2x = $181.25 / $6.64 ✓).

3.7 FactorsToday factor model

https://www.factorstoday.com/api — public no-auth factor/risk model, accessed 2026-07-16/17. Supplied the factor-positioning read (Value +0.273 / Momentum −0.054, negative in all four nested models / Quality zeroed in all four; R² = 0.144; specific vol 25.9%; max DD −76.4% at 5y; rs_peak −79.1%).

Caveat: the model reports a realized beta of 0.11355, which would imply a cost of equity of ~4.7% — not a credible discount rate for an equity with ~33% annualized volatility and a −76% five-year maximum drawdown. CAPM was rejected; the analysis uses a labeled judgment build-up (risk-free ~4.2%, ERP ~4.5%, beta ~0.9–1.0 → CoE ~8.3–8.7%; WACC ≈ 8.5%, range 8–9%) explicitly tagged ASSUMPTION. Leaderboard returns are annualized at every horizon, including short windows; short-horizon figures were de-annualized and reconciled against the daily price series before use.

4. Secondary — third-party research, industry data and framework

5. Date-integrity check

6. Claims resting on a single source, or on interpretation — declared

For transparency, the memo’s load-bearing claims that are not direct primary disclosure:

1. MKTX’s share of the electronic HG market (~60% → ~31%) — DERIVED ARITHMETIC, not a disclosed figure. It divides two of MKTX’s own disclosed series that share a denominator (both are measured against the total market: “electronic market share” is defined identically in every 10-K FY2021–FY2025 as “the level of electronic trading as a percentage of all means of trading”; the share series is “our estimated market share of total U.S. high-grade corporate bond volume”). The division is therefore valid, and it is independently corroborated by Morningstar (see §5). Caveats that must travel with it: the electronic-penetration inputs are MKTX’s own rounded estimates presented in a promotional frame (“creating a long runway for future market share growth”), not audited metrics; ±2.5pt rounding moves the FY25 HG figure ~±1.3pts. The HY leg is materially weaker than the HG leg — MKTX revised HY penetration 20% → 30% in a single year (FY21 → FY22 10-K) and then froze it at exactly “approximately 30.0%” for four consecutive years, which reads as un-updated boilerplate rather than a measured series; the ~76% (2021) start point is therefore unreliable and the HY decline should be rebased to 2022 (~60% → ~42%) or explicitly caveated.
2. The margin decomposition (price/mix −10.42pts, opex +2.56pts) — a counterfactual holding FY21 FPM constant; arithmetic ties exactly (48.25 + 2.56 − 10.42 = 40.39). Labeled an upper bound on the FPM effect, because it assumes all FY25 volume would clear at 2021 pricing.
3. Normalized EPS — the FY25 $6.64 + $0.75 = $7.39 bridge is MKTX’s own disclosed non-GAAP reconciliation, not our derivation. The ~$7.75 TTM figure is derived (independently reproduced at $7.75 vs the draft’s $7.77). The two routes differ by 4.9%; the memo states which it underwrites.
4. The Q4-2025 tax release (~$31.3M) — derived, but from four of MKTX’s own documents: the “Reserve for uncertain tax positions” notable is $54,939K in the Q1, Q2 and Q3-2025 10-Qs and $23,631K for FY2025, locating the release in Q4. Corroborated independently by the UTB liability move ($56.4M at 3/31/25 → $22.363M at 12/31/25). Note: the FY25 10-K’s UTB reconciliation shows no discrete decrease line — its “+$19,909 prior periods” is a net figure that already absorbs the Q4 remeasurement. The reconciliation must not be cited as showing “no release.”
5. The FY2025 buyback — $420,015,000 (cash-flow statement) and $360.0M / 1,980,715 sh (equity note) are both correct on different bases and reconcile exactly: $420.0M cash − $60.0M of the ASR sitting in APIC pending its 2026-02-04 settlement = $360.0M recorded as treasury stock. Blended: 2,340,497 shares (incl. the 359,782 delivered 2026-02-04) at $179.45. The two bases must not be combined.
6. The disclosure-withdrawal inference — the deletion is FACT, verified by diffing the share table’s row labels across FY2021–FY2025 (7 rows in FY2022 → 5 in FY2023; EM, Eurobonds and Composite removed; never restored; Composite absent from the document entirely from FY2024). The motive is NOT established: the very filing that deleted the rows (FY2023 10-K) states that Eurobond volumes rose “due to increases in estimated market volumes and our estimated market share” — i.e. MKTX said its Eurobond share was rising in the deletion year. The defensible adverse inference is narrower and rests on the post-deletion attribution shift: MKTX cited share gains as a volume driver while it had them (2021–2023) and stopped citing share at all once the rows were gone (FY2024: “mainly due to an increase in estimated market volumes”; FY2025: “driven by an increase in block trading”). The withdrawal removed verifiability; it does not by itself establish decay.
7. The Bloomberg competitive ceiling — INTERPRETATION. Bloomberg discloses no share data. Supported by MKTX naming it as a competitor across credit, rates and data in the FY2025 10-K, not by any share figure.
8. The Tradeweb FPM comparison — the trend (TW cash credit −14.3% vs MKTX −7.6%) is comparable and reportable. The levels ($126.83 vs $138.87) are NOT — the two issuers’ credit categories differ materially (MKTX’s includes EM, eurobonds, munis; TW’s excludes derivatives, China, EP flow). No conclusion rests on the level comparison.

All primary sources above are public SEC filings, earnings-call transcripts and public data, read in full.