Financebotresearch desk研究台

AXP › analyze

AXP · Analyze

ACCUMULATE Financial Services

Date: 2026-09-23 | Price: $301.96 | 52wk Range: $290.97 – $387.49 First analysis — no prior report exists for this ticker.

1. The one-sentence version

American Express is down 22% from its high because the market is pricing a consumer-credit deterioration that has not appeared in American Express's own credit data — net write-offs are 2.0% and flat, roughly half the large-bank average, while delinquencies sit at the low end of a three-year range.

That is the whole call. Everything below tests whether the credit data is trustworthy, whether the franchise that produces it is durable, and whether 17.2x forward earnings is a good price for it.

2. Data quality — what had to be thrown out first

AXP is a lender, a network and an acquirer in one entity, and standard vendor fields break on it. Per [[principle-primary-source-beats-vendor]], the figures this verdict rests on come from the 8-K and press releases, not the screener.

Vendor field Value Status
PE(fwd) 14.98 prices forwardEps $20.16 ❌ FY2027, not FY2026. The true current-year multiple is 17.1x on epsCurrentYear $19.02→$17.65, which ties exactly to company guidance of $17.30–17.90. Fourth-plus firing of [[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]] in this repo.
GrossMargin 0.62, OpMargin 0.20 — ❌ Synthetic. Do not use. A lender/network has no meaningful COGS line. roic.ai correctly returns null for both; Yahoo manufactures them. New pitfall — see §11.
totalRevenue $70.91B snapshot field ❌ Disagrees with the sum of Yahoo's own quarterly statements ($75.95B TTM) by 6.6%. The statement table is right; the snapshot field is not.
Tangible Book Value = Stockholders Equity ($34,280M, Q2'26) — ❌ Silently drops the goodwill subtraction when the goodwill line is missing for that quarter. Real TBV is $29.3B (equity $34.28B − $4.96B goodwill/intangibles) → TBVPS $43.41.
EV $219B, EV/Rev 3.09 — ❌ Meaningless for a lender — debt is raw material, not a claim on enterprise value.
D/E 172 — ❌ Structurally meaningless. Use equity/assets 11.1% and CET1 10.4%.
Revenue $72.23B (FY2025) statement table ✅ Correct. Ties exactly to AXP's reported "total revenues net of interest expense." [[pitfall-vendor-revenue-is-net-construct-for-banks]] does not bite here, because AXP's own headline metric is already the net construct — but the basis is named explicitly, as that note requires.
Graham IV $137.18 — ⚠️ Weak lens, do not lead with it. Graham's 22.5 multiplier embeds a 1.5x book ceiling. A business earning 34% ROE / 3.79% ROA legitimately clears 1.5x book by a wide margin. Reporting "121% overvalued" from this would be false precision.

3. Fundamentals

TTM through Q2 2026 (sum of the four quarterly statements; verified against the 8-Ks):

Metric TTM FY2025 FY2024 FY2023
Revenue net of interest expense $75.95B $72.23B $65.95B $60.52B
Net income $11.45B $10.83B $10.13B $8.37B
Diluted EPS $16.48 $15.38 $14.01 $11.21
Diluted shares 679M (Q2'26 avg) 696M 713M 736M
ROE ~34% 33.8% 34.5% —
ROA — 3.79% 3.80% —
  • Growth is real and broad. Revenue +10% FY2025 (+9% FX-adj), Q2'26 +10%. Net income CAGR 13% (3y), revenue CAGR 11% (3y). Billed business Q2'26 $455.8B, +9% — the highest FX-adjusted spend growth in three years.
  • Share count −3%/yr, sustained. 736M → 713M → 696M → 679M. This is a real per-share tailwind, not a one-off.
  • Capital returned: $8.7B in the 12 months to 2026-03-31. Dividend $3.80/yr, raised +15.9% in 2026; payout only 21%.
  • Capital position is unconstrained: CET1 10.4% (target 10–11%), and the 2026 DFAST left AXP at the 2.5% SCB floor — the statutory minimum — through 2027-09-30. There is no regulatory brake on capital return.

⚠️ FCF is structurally meaningless here and is excluded. fin.py reports FCF $16.0B with a −5.9% 3yr CAGR, which reads as deterioration. It is an artifact: for a card issuer, operating cash flow is dominated by provision add-backs while loan-book growth sits in investing. Per [[pitfall-vendor-revenue-is-net-construct-for-banks]]'s own closing note, a lender's cash-flow line from a vendor is not usable. Net income, EPS, ROA and book value carry this analysis instead.

Dividend-grower overlay

3yr 5yr 8yr 10yr
Dividend CAGR 16.7% 12.9% 11.6% 11.1%

Accelerating, not decaying. Payout 21% of earnings leaves enormous headroom. The one blemish is 2021 (flat at $1.72) — a Fed capital restriction during COVID, not a company decision, and not a cut.

⚠️ DDM is not usable on AXP — dividend growth (11–17%) exceeds any sensible discount rate, so the model diverges. DYT is also weak: the 1.26% yield against a 1.10% five-year average implies only ~13% undervaluation, and part of the yield expansion is price-driven. Neither is load-bearing here. Said plainly rather than reported as a number.

4. Credit — the thesis, and the reason to own it

For a lender this is the analysis. Everything else is commentary.

Period 30+ DPD Net write-off rate
Q1'24 – Q4'24 1.2–1.3% 1.9–2.1%
FY2025 (full year, 8-K) — 2.0%, flat YoY
Q4'25 1.3% 2.1%
Q2'26 1.2% 2.0%, flat YoY
Jun–Aug 2026 (monthly) 1.1% 1.4–1.7%

Peer comparison, Q2 2026 — this is the number that matters:

Issuer Card net charge-off rate
AXP (US Consumer) ~2.0%
JPMorgan Card Services 3.34%
Capital One domestic card 4.71%
100 largest banks (FRED) 3.70%
All commercial banks (FRED) 3.82%

AXP charges off at roughly half the large-bank average and well under half Capital One's. Provisions fell 21% YoY to $1.1B in Q2'26, and the reserve rate is 2.7% versus 2.9% in 2019 — i.e. reserved more conservatively than pre-COVID. Loan balances grew +9%, exactly matching billed business +9%: the book is growing with spend, not ahead of it. That is the signature of a lender not reaching for yield.

Two honest caveats, both material:

  1. The advantage is partly a mix artifact. A large share of AXP's billed business is charge-card / pay-in-full volume that never becomes a loan at all. The NCO gap to Capital One is therefore not a like-for-like underwriting comparison. AXP's customers are genuinely more affluent, but the metric flatters the gap.
  2. Part of the Q2 provision decline was a reserve release, not pure credit improvement — a lower-quality source of the beat. Net write-offs themselves were flat, not down.
  3. ⏳ The monthly write-off series does not cleanly reconcile to the quarterly 2.0%. June's 1.4% print looks anomalous against a 1.9–2.2% two-year band, which means the bearish "write-offs rising to 1.7%" framing circulating in commentary is probably an artifact of comparing to that outlier. Flagged as unresolved — see §10.

Net read: credit is stable-to-improving and best-in-class, and the market is not paying for it.

5. Moat

Rating: 4.1 / 5 — wide, brand-led, actively contested, and holding.

Source Rating Basis
Intangibles — brand / status 4.5/5 The core asset. Platinum went $695 → $895 (+29%) in Sept 2025, and the CFO reports renewal rates "not only through the roof, they were flat year-over-year" on the ~25% of the US Platinum book repriced by Q1'26. A 29% price rise absorbed without measurable attrition is the definition of an intangible moat.
Switching costs 3.5/5 Real but behavioural, not contractual. Membership Rewards points transfer to 15+ partners — the currency is not captive.
Network effects 3/5 AXP is the small network — 11.1% of US card purchase volume vs Visa 61.1% / Mastercard 25.8%. The effect that works is the niche premium-cardholder ↔ premium-merchant loop (Resy, Tock, TheFork).
Cost advantage 2/5 AXP deliberately runs the industry's highest cost structure. The genuine edge is underwriting: ROA 3.79% against a 1.0–1.3% typical bank — the cleanest moat number in the file, since 34% ROE is flattered by ~9x leverage.
Efficient scale 2/5 Weakest. Premium cards are not a natural monopoly, and two rivals are now inside the segment.

The closed loop: what it actually buys

AXP is issuer + network + acquirer. It captures the full ~2.2%+ discount rate instead of splitting interchange with a third-party bank — and that is the mechanical link to the brand: the whole take is what funds the $3,500 Platinum benefit stack. It also yields both sides of the transaction record (funding Amex Offers and the underwriting behind the 2.0% write-off rate) and direct merchant contracts.

But the closed loop is neither necessary nor sufficient for the moat. Discover ran one for 40 years with ~2% share and no pricing power; Chase has none and just took Sapphire Reserve to $795 successfully. Judgement: the brand is ~60% of the moat and the closed loop ~40%, with the loop acting mainly as the funding mechanism that lets the brand be sustained.

Adversarial stress-test

  • JPMorgan — 7/10, the underrated threat. JPM owns the primary banking relationship for ~80M US customers and can upgrade a 24-year-old into Sapphire at 30 for near-zero marginal cost, while AXP must buy that customer. AXP's millennial/Gen-Z win is a paid-acquisition win; JPM's is a default-relationship win, and JPM's is cheaper forever. JPM also funds with near-zero-cost checking deposits against AXP's rate-paying deposits plus $57.8B of wholesale debt. Chase moved first — Sapphire Reserve $550→$795 in June 2025, three months before AXP's refresh. That removes "only Amex can raise price" from the thesis; this was an industry-wide repricing.
  • Capital One + Discover — 4/10, the overrated threat. COF closed the $35.3B deal in May 2025 and now owns a network. But Discover is ~2% of US volume with thin, Diners-dependent international acceptance — the wrong rails for premium T&E spend, and a decade of merchant negotiation to fix. The real logic is interchange capture on debit and mass-market credit, an attack on Visa/Mastercard, not on the Amex brand. Evidence the $395 tier is straining: from 2026-02-01 Venture X authorized users pay $125 and free guesting ended. It caps AXP's breadth, not its depth.
  • Apple / Google / fintech — 3/10. The wallet layer commoditises the rail. AXP does not sell a rail; it sells an $895 subscription. Tokenising a Platinum behind Apple Pay does not reduce the fee.
  • Agentic commerce — 6/10, long-dated and genuinely novel. If an AI agent picks the instrument and optimises for cost, the rewards card loses top-of-wallet by construction. AXP's counter (closed-loop trust, fraud and dispute data; an agentic insurance product) is plausible but unproven. The near-term protection is that the agent serves a consumer who already paid $895 and will instruct it to earn against that. Not a 5-year problem; possibly a 10-year one.

Stress-test verdict: the answer is not "easily" — a real moat exists. But two well-capitalised rivals are attacking right now with credible products, and AXP's response has been to spend more. A moat that requires escalating spend to hold is a moat under pressure, even while it holds.

Merchant acceptance — the old bear case is closed; the new one is price

Acceptance is effectively solved: 99% of US places that take credit cards, and 170M+ merchant locations globally at YE2025, ~5x the 2017 level. Parity is substantially achieved.

But parity was bought with rate, and AXP stopped disclosing the average discount rate — last reported ~2.39% in Q3 2019, i.e. the single best moat-pricing metric went dark exactly as it began compressing. It must now be triangulated from the gap between billings growth and discount-revenue growth:

Period Billed business Discount revenue Implied rate/mix drag
FY2025 ~+8% +6.3% ~1.5–2pp
Q2'26 +9% +9% ~0–0.5pp

Compression is real and persistent at roughly 1–2%/yr, but it narrowed sharply in 2026 as the Platinum refresh pushed spend into high-rate T&E (travel bookings +22%, Resy spend +20%). That is the closed loop earning its keep — AXP can offset rate decay by changing what gets bought on the card, a lever Visa and Mastercard do not have. ⚠️ Inference, not disclosure.

Evergreen: 4/5. The franchise — charging affluent people a subscription to be taken care of, since 1850 — is about as durable a demand curve as exists. Marked down one point because the revenue line that is 52% of the company is the one with structural decay.

6. The mix shift — and why the common bear framing is wrong

The standard bear case is "AXP is quietly becoming a lender, so it deserves a lower multiple." The data does not support it.

Line FY2016 FY2025 FY25 growth
Discount revenue $18.68B · 58.2% $37.40B · 51.8% +6.3%
Net card fees $2.89B · 9.0% $9.99B · 13.8% +18.3%
Service fees & other n/d $7.47B · 10.3% +10.4%
Net interest income ~$6.5–7.0B · ~21% (inferred) $17.36B · 24.0% +11.7%

Net interest income's share moved ~21% → 24% in a decade — about +3pp — and has been flat for two years (23.6% → 24.0% → 23.7%). The large shift is discount revenue → net card fees: −6.4pp and +5.6pp since FY2016.

That distinction is the single most important structural finding in this report:

  • Discount revenue is a toll on merchants, priced by forces AXP does not control, and it is the slowest-growing line in the company.
  • Net card fees are a subscription — recurring, repriced annually, 32 consecutive quarters of double-digit growth, guided to exit 2026 in the high teens, with proven 29% price elasticity. Over 70% of new accounts are on fee-based products.

AXP is migrating from a transaction-toll business toward a subscription business with a transaction toll attached. Subscription revenue with demonstrated pricing power deserves a higher multiple, not a lower one. The mix shift is moat-positive, and the market appears to be pricing the opposite.

7. Regulation — the perceived risk is inverted

The Credit Card Competition Act (S.3623, reintroduced 2026-01-13) explicitly exempts three-party systems. The bill's mechanism is a routing mandate — forcing large issuers to enable a second unaffiliated network per card so merchants can route to the cheaper one. That attacks the four-party open loop. American Express is simultaneously network, issuer and acquirer, charging a single discount rate rather than an interchange fee; there is no separable interchange leg to route around.

So the headline regulatory risk — which now carries a presidential endorsement, genuinely raising its odds versus prior cycles — damages Visa, Mastercard and issuing banks while leaving AXP's structure untouched. It is a rare case where the regulatory tail is favourable on a relative basis.

⚠️ The indirect exposure is real but unquantified: if the bill compresses industry-wide merchant economics, AXP's rate premium faces comparison pressure, and AXP's own 10-K names "pricing for card acceptance" and "increased surcharging, suppression or other differential acceptance practices" as risks. Magnitude is inferred from structure, not measured.

Also live: the CFPB vacated its $8 late-fee cap in April 2025, then reopened the question with a July 2026 RFI — typically the first step toward rulemaking. AXP is far less late-fee-dependent than subprime-tilted issuers, so this lands on COF/SYF harder. No current DOJ or FTC action found.

8. Risks — named honestly

  1. Delta concentration is the largest single risk and it is bigger than most investors carry in their heads. AXP paid Delta $8.2B in 2025, guiding ~$9B in 2026 (+10%) and discussing $10B long-term. $9B is ~12.5% of AXP's entire FY2025 revenue paid to one counterparty. The contract runs to end-2029, so renewal is a 2028–29 event, not a 2026 one — and Delta has re-architected its route network around the relationship, which cuts both ways. But Delta's remuneration is compounding at ~10%/yr against discount revenue growing at 6.3%. The partner is capturing economics faster than AXP is creating them on that book. Meanwhile AXP is pruning other co-brands (Amazon, an SME portfolio), which makes the remaining book more Delta-concentrated, not less.
  2. A known, dated headwind hits the very next print. The Amazon and second small-business co-brand portfolio transfers carry a ~2.5pp drag on net interest income landing in Q3 2026 (CFO). This is disclosed, not a surprise — but the market reacted −6% to a Q2 beat, so the reaction function is unforgiving.
  3. The moat now costs more to hold. Card Member services expense +50% to $1.949B on the Platinum benefits; variable customer engagement expense guided to 44–45% of revenue. Management raised FY2026 revenue guidance to 10% while holding EPS guidance at $17.30–17.90, explicitly choosing reinvestment over flow-through. Treat that as the price of the moat — but if it becomes permanent, steady-state earnings growth is structurally below revenue growth.
  4. Commercial is the laggard. Billed business +5%, and management does not expect a meaningful 2026 recovery. Fintechs (Ramp, Brex) took the expense-software layer; AXP's "Center" response is reactive and late. This is the one front where AXP is clearly behind.
  5. Insiders are not buying. Broad, continuous officer selling (>$63M trailing twelve months) with one genuine open-market purchase in eighteen months — director Angelakis, 3,700 shares at $269.89 on 2025-03-07, below today's price. Much of the selling is mechanically paired with option exercises and the January grant cycle, which lowers its signal value considerably — every "buy"-shaped line in the Form 4 record is a $0.00 RSU grant or an exercise-and-sell ([[pitfall-yahoo-insider-purchases-counts-rsu-grants]]). But nobody stepped in at a 52-week low.
  6. Discount-rate decay with the disclosure switched off (§5) — a slow, certain, unobservable erosion.
  7. The market has a view and it has been directionally right all year. AXP is +0.6% over six months against Visa +19.2%, Mastercard +12.3% and SPY +17.7%. Over one month it trades with the lenders (SYF −11.7%, AXP −10.5%, COF −9.3%) rather than the networks (V −5.5%, MA −6.7%). Every one of ~30 analyst targets sits above spot, including the lone Sell — a consensus that has been wrong for months.

9. Valuation

At $301.96, with FY2026 guided EPS of $17.30–17.90:

Model Input Output
P/E on company guidance $17.60 midpoint at 17–20x $299-352
P/E on FY2027 consensus $20.16, discounted one year at 10%, at 17–20x $311-367
P/TBV TBVPS $43.41, ROTCE ~39% $278-386 — ⚠️ model is highly sensitive to the growth/cost-of-equity spread; brackets the answer, does not anchor it
Bogle expected return 1.26% yield + ~12–13% EPS growth, multiple flat ~13-14%/yr
DDM g (11–17%) > r ❌ Diverges — not usable
DYT 1.26% vs 1.10% 5yr avg ⚠️ Weak; ~13% implied, partly price-driven
Graham IV $137.18 ⚠️ Weak lens — embeds a 1.5x book ceiling on a 34%-ROE business
Analyst mean 30 analysts, 14 Buy / 15 Hold / 1 Sell, range $315–450 $378.29

Its own history is the most useful anchor. Per roic.ai: AXP ended FY2025 at 23.9x P/E and 7.58x P/B; FY2025 averages were 20.24x and 6.43x; FY2024 averages 17.27x and 5.67x. Today it trades at 18.3x trailing / 17.2x forward and 5.95x book (6.96x tangible) — below both years' year-end marks and at the low end of the two-year average band, while earnings grew ~11%. That is a ~23% multiple compression against rising earnings.

Fair value: $305-365, central ~$335.

I am deliberately ~11% below the Street's $378.29. The Street's targets have been above the price all year and kept being cut toward it; and my range does not extend credit for the FY2027 consensus until the Q3 print shows the co-brand NII headwind absorbed.

Entry $280-305. Spot $301.96 sits inside the top of this band. Trim 23x ttm (≈$379 on TTM EPS $16.48).

⚠️ The trim is written on ttm, not fwd, deliberately. Yahoo's forwardPE prices FY2027 EPS ($20.16) — a bare 23x fwd would render $464, roughly 54% above spot and far above any defensible fair value, telling you to hold a full position through a large overvaluation. See [[pitfall-multiple-trim-inherits-the-broken-vendor-field]]. trailingPE 18.33 ties exactly to the verified TTM $16.48, so 23x ttm ≈ $379 and rises mechanically as earnings grow — by FY2026 year-end, on ~$17.60 TTM, the same multiple is ~$405.

10. Verdict

ACCUMULATE — conviction 7.5. Start a position; stage the rest around the 2026-10-23 print.

This is a genuine dislocation rather than a stumble. American Express has de-rated from 23.9x to 17.2x forward in nine months on a consumer-credit fear that its own credit data does not corroborate — write-offs flat at 2.0% against a 3.7–3.8% industry, delinquencies at the low end of a three-year range, reserves set more conservatively than 2019, and loan growth exactly matching billings growth. The business underneath is compounding revenue at 10%, EPS in the low teens, shrinking the share count 3% a year, earning 34% on equity and 3.79% on assets — roughly three times a typical bank. Berkshire, at ~22% of the company, did not touch a share in the Q2'26 13F while trimming Capital One and Ally in the same quarter — the clearest available statement that a credit-sensitive holder distinguishes this book from the rest of the sector.

The moat is wide and, in the part that is growing fastest, strengthening: the mix is shifting from a merchant toll toward a subscription with demonstrated 29% price elasticity and 32 straight quarters of double-digit growth. And the most-feared regulatory outcome, the CCCA, structurally exempts three-party systems — it is a relative positive, not a risk.

What holds conviction at 7.5 rather than higher: the premium segment is contested for the first time in AXP's modern history, and Chase moved first on price — so the repricing validates the segment, not AXP's uniqueness. The moat is being defended with escalating spend that management has chosen not to let reach EPS. Delta is ~12.5% of revenue and compounding faster than the revenue funding it. The discount rate is decaying with the disclosure switched off. And no insider has bought at a 52-week low.

A portfolio-specific passage was removed from the public build.

Break triggers: - Net write-off rate above 2.5% for two consecutive quarters, or 30+ delinquency above 1.6%. - Net card fee growth falls below 10% — this breaks the 32-quarter streak and the subscription thesis with it. The single most thesis-critical number in the file. - The billings-growth-minus-discount-revenue-growth gap widens beyond ~3pp for two quarters — the only visible proxy for the discount rate now that disclosure has stopped. - Any major co-brand loss, or Delta renegotiated on materially worse terms. - FY2027 EPS guided below ~$19.50, breaking the mid-teens algorithm. - Marketing/acquisition spend accelerating together with card-fee growth — the tell that the flywheel is being bought rather than earned.

Upgrade conditions (toward 8.0 and a larger position): credit holding near 2.0% through a genuine consumer slowdown · card-fee growth staying high-teens while VCE expense normalises below 44% of revenue · the Q3 print absorbing the co-brand NII headwind without a guidance cut · any open-market insider purchase · price into $270-285.

Recheck: 2026-10-23 (Q3 FY2026, 7:00am ET).

11. What this analysis did NOT establish

  • The monthly net write-off series does not reconcile to the quarterly 2.0%. June's 1.4% looks anomalous against a 1.9–2.2% two-year band. I treated the bearish "write-offs rising" framing as weakly supported, but I could not corroborate the June figure against a primary 8-K. Since credit is the thesis, this is the most important open item.
  • No hard Platinum churn or attrition number exists publicly. The strongest claim in the bull case rests on a CFO sentence ("renewal rates… flat year-over-year") covering ~25% of the US book. The existing-base repricing only began Dec'25/Jan'26, so the evidence is genuinely too young to read.
  • FY2021 revenue line items could not be retrieved, so the decade mix comparison in §6 anchors on FY2016 and FY2024/25. FY2016 net interest income is an inferred residual, not read from the filing.
  • The magnitude of AXP's indirect CCCA exposure is inferred from business-model structure, not measured. The exemption from the routing mandate is textual and solid; the second-order discount-rate pressure is a judgement.
  • The Delta "~4% of card spend" figure — which drives the implied 12–15%-of-billings estimate — comes from a secondary source and is unverified. Treat it as order-of-magnitude.
  • No 2026 buyback authorization was found; the standing plan is the 120M-share March-2023 authorization. Remaining capacity should be verified against the 10-Q before relying on the 3%/yr pace.
  • Berkshire's exact Q2'26 share count was not retrieved — only that AXP remained the #2 holding at 17.1% of the portfolio and was absent from the quarter's trim list. Last hard number: 151.6M shares (22.2%) at 2026-03-31.
  • Segment billed-business splits and the 65% millennial/Gen-Z figure come from one secondary summary of the investor deck, not the deck itself.
  • Litigation: absence of evidence. Searched DOJ/FTC/CFPB and found nothing current; not proof of none.

Sources