AXAxos Financial, Inc.
Industry Banks Updated 24 Sep 2026
The idea
10x forward earnings for a bank compounding book value; the acquired Capital One IRA deposits are cheap funding if they arrived, and NIM above 4.6% carries the bull case.
Events to watch
- Fiscal Q1 2027 10-Q (~2026-10-29): the deposit table (did the ~$1.3bn of IRA balances transfer?) and the purchased-loan marks
- efficiency ratio
How we got the target
Scorecard: base ~$110 at ~1.10x modelled FY2028 book. Time frame: FY2028 (Jun 2028).
Target: from the research notes. Prices: market data get_latest_stock_price, close of 2026-09-22 (fetched 2026-09-23)
What would prove it wrong
8 warning signsNet charge-offs to average loans exceed 0.50% for two consecutive quarters.
Details
FY2026 was 0.18%, from 0.13% and 0.05%. The book is 34.0% commercial real estate and 43.5% CRE-plus-multifamily, with 37.4% of the real-estate portfolio in New York and 33.5% in California, and $2.54bn of CRE — 29% of that book — matures within twelve months of 2026-06-30. This is the name's real risk and it is fully disclosed.
Net interest margin below 4.25%.
Details
4.70% now. The purchased-loan runoff alone is worth roughly 13bp a year at current balances; competition for the deposits that replace Capital One's IRAs would do the rest.
The efficiency ratio crosses 52%.
Details
43.59% → 46.84% → 49.49% over three years. The whole equity story is that a branchless bank out-earns its cost base; three years of this breaks it.
Bank-level CET1 below 10.0%.
Details
12.74% → 12.42% → 10.93% in two years, while assets grew 21%. Axos is growing its balance sheet faster than it generates capital and returns almost none ($22.0m of buyback in a year on $490m of earnings), so the constraint arrives on its own schedule.
A second deposit acquisition closes materially below its announced size.
Details
Once is attrition; twice is the acquired-deposit machine not working, and acquired deposits are where the growth is.
Shares outstanding rise in a second consecutive fiscal year.
Details
FY2026 was the first since FY2022 (56,483,617 → 56,734,049), because the buyback was cut to $22.0m while RSU net-settlement cost $30.3m and SBC ran at $44.3m. The criterion above on CET1 says when the buyback has to stop; this one says what it looks like once it has.
Unrecognized tax benefits above $150m, or a settlement of the current $97.4m below half.
Details
The balance went $19.6m → $97.4m in one year, 20% of net income and $1.72 a share, fully reserved. It resolves in both directions and neither is in any model.
The FINRA petition to vacate is denied and the remaining ~$28m is accrued in one quarter.
Details
43% of a ~$49m award is accrued today; the rest is a known, dated, binary item.
How far along
6 of 10 steps done · 4 not started
Next step
on the fiscal-Q1 2027 10-Q (~late October) read the deposit table and answer one question the 8-K does not: did the ~$1.3bn of Capital One IRA balances that did not transfer arrive later, go somewhere else, or never exist by closing? Then read the purchased-loan average balance and yield in the MD&A average-balance table — that is the FY2027 earnings bridge, and it is the number most likely to be missing from any model built off the headline.
Needs your decision
S6 product test / expert call, S7 conviction.
Failed checks
none. S0/S4/S6/S7 are not closed awaiting the cluster pass, not carried forward — cluster-banks has not run, and the skill says do not improvise cluster stages per name. This is the second name in the rotation in that position, after OMAB. S8 closed closed on 2026-09-24 — reliable, 5 rows, but with the first failed MISSED row in this backfill slot: the Capital One IRA purchase announced at ~$3.2bn and closed at ~$1.9bn, unexplained in every filing between the two 8-Ks. Gates are non-blocking, so S8 is closed at one miss in five; the miss is carried into the kill criteria below. S3's 2026-09-22 "no transcript" deviation is resolved (seven usable quarters read, dodge log opened) and its earnings-quality check is written (3 flags). Still open and now explicitly so: the §25 bonus-metric read, because the FY2026 DEF 14A has not been filed. Still open: the ⚠️⚠️ transcript-archive defect under S9.
Who runs it
The Front OfficeStep by step
- S0How it does as AI gets cheapNot started
awaiting cluster pass
cluster-banksNOT improvised. AX's watchlist cluster is
us_financials, scheduled in that slot alongside C, SYF, AMG, VBNK, INTR, BMA, PAM, OMAB and EPU, which has not run.Full notes
Liquidity, which does not need the cluster: ADV $35.1m/day (21 sessions to 2026-09-21, market data NYSE:AX) on 56.7m shares — the THINNEST name this rotation has passed and the binding constraint on any position size. Owner checks dropped 2026-09-21 (README §Stage 0 override).
- S1Where the idea came fromDone
watchlist (owner, Apple Stocks, 2026-09-20), promoted by the Tier 0 EDGAR sweep on the 8-K filed 2026-09-21 (queue.json, priority 2, as_of 2026-09-22). Under the promotion rule it replaced RYAN as the last member of this slot's batch.
Full notes
RYAN returns to a later insurance slot. That promotion is the reason this name was read, and the filing that triggered it is the run's finding.
- S2Quick deal-breaker checkDone
FY2026 (ended 2026-06-30) net income $490.4m (+13.3%), diluted EPS $8.48 (+14.1%), adjusted EPS $8.74; net interest income $1,246.9m (+10.6%); total assets $29.96bn (+20.9%); deposits $24.57bn (+17.9%); equity $3,166m. BVPS $55.81.
Full notes
Five-year share count 61.9m (FY2019) → 56.66m (FY2026), −8.5%, no issuance. Credit is still benign: NCO 0.18% of average loans, non-performing assets 0.53% of assets (from 0.71%), ACL/loans 1.34%, ACL/nonaccrual 221%. Survives comfortably.
⚠️ but every efficiency series has deteriorated two years running: NIM 4.62% → 4.90% → 4.70%, ROE 21.64% → 17.30% → 16.32%, efficiency ratio 43.59% → 46.84% → 49.49%, and Bank CET1 12.74% → 12.42% → 10.93%.
⚠️
⚠️ FY2026 non-interest income +78.2% is not repeatable as stated: a $22.0m one-off legal settlement plus Verdant operating-lease rental income whose matching $51.6m of depreciation sits in expense.
⚠️
⚠️ the purchased-loan book (the 2023 FDIC portfolio) is running off: average balance $973m → $713m, interest income $160.6m → $115.4m at a 16%+ yield. That is a disclosed, dated, ~$45m/yr headwind nothing in the guidance replaces. · capital allocation (2026-09-24, dossier-2026-09-24.md): compounds. FY2022-26 operating cash flow $1,766.7m; $225.4m of buybacks and no dividend ever = 12.8% returned, the other 87% into a balance sheet that went from $14.27bn to $29.96bn of assets and $1,401m to $3,166m of equity. ROE 21.64% → 17.30% → 16.32% (10-K), ROA 2.08% → 1.76%; BVPS $23.62 (2021-06-30) → $55.81 = +18.8%/yr with no dividend, TBVPS $37.26 → $44.60 → $50.96. Goodwill impairment: zero in five years. One material acquisition — Verdant, $474.4m of cash at 2025-09-30 for ~$1.0bn of loans and leases plus $212.6m of operating-lease equipment, ~$46.7m of goodwill, $7.8m of day-one CECL.
⚠️ THE INCREMENTAL RETURN IS 14.1% — net income +$249.7m on equity +$1,765m — against a ~18% five-year average. New capital earns four points less than old, which is the same fact as the NIM, efficiency-ratio and CET1 lines above, read from the capital side.
⚠️ BUYBACK IS THE SHOCK ABSORBER AND IT IS BEING ABSORBED: $96.3m (FY2024) → $58.2m (FY2025) → $22.0m (FY2026), 249,921 shares at $87.95 = 1.58x book / 1.73x tangible book, $126.1m left. In the same year Axos paid $30.3m to net-settle RSU taxes (337,635 shares at $89.69) — 38% more cash than it spent on buybacks — against $44.3m of SBC, so shares outstanding ROSE for the first time since FY2022 (56,483,617 → 56,734,049). Five-year diluted count is still −4.8%; the direction turned. If CET1 keeps falling the buyback goes to zero first, and then dilution is the only direction. open Bonus metric NOT read — the document does not exist yet. The FY2026 DEF 14A is still not on the index; the latest remains the 2025 proxy (0001299709-25-000174), which the 2026-09-22 pass already flagged. The §25 EPS-pay-plus-buybacks test is open rather than passed. Pointer only: Verdant origination is inside the incentive plan ("incented to help support and grow that business", call 2026-04-30 [27]).
- S3Reading the company's reportsDone
(gate-minimum)
FY2026 10-K Item 1A in full (risk ordering, the NY/CA real-estate concentration, multifamily), Item 7 MD&A (three-year ratio table, average balance sheet and yields, loan composition and maturity tables, segment results, non-GAAP reconciliation), the contingencies note (the FINRA award), BDO's audit and ICFR reports; the 2026-07-30 FY2026 results 8-K; the 2026-04-23 and 2026-09-21 deposit-deal 8-Ks; the 2025-09-25 DEF 14A ownership table; and a filing-index red-flag sweep.
Full notes
Three disclosed risks named with locators.
⚠️ Deviation: no earnings-call transcript — Equibles did not connect for this session and no Axos transcript was located on fool.com.
⚠️ the DEF 14A read is the 2025 one; the FY2026 proxy was not yet on the index. EQ: 3 flags (2026-09-24, dossier-2026-09-24.md) — (1) acquisition-related costs (footnote 2: "includes amortization of intangible assets") added back in three years of three and growing, $10.8m → $7.4m → $13.7m, i.e. a category presented as transaction-specific that has never been absent (mild: 0.9% of revenue, $0.24 of the $8.74); (2) unrecognized tax benefits $19.6m → $97.4m in one year — a $77.8m build to 20% of net income / $1.72 a share — in the year the effective rate fell 29.42% → 24.17%. The accounting is conservative (a "100% reserve" against an amended-return refund claim, tax note), so it does not flatter earnings; the flag is the tax POSITION, a two-sided item nobody is modelling; (3) ~$28m of the ~$49m FINRA award still unaccrued (~$0.49/share pre-tax) — $21.0m taken, plus $7.0m for the 2019 COR Securities matter. Clean: OCF/net income 0.68x → 1.13x → 1.14x with accruals negative in the last two years; SBC flat at 2.97/3.33/2.99% of revenue; capitalised software +18.4% against revenue +17.6%; credit IMPROVING (NPAs 0.71% → 0.53%, nonaccruals 0.79% → 0.60%) while the provision tripled ($32.5m → $55.7m → $101.1m), most of which is 1.5% CECL on $4.5bn of new loans plus $7.8m of Verdant day-one. The non-GAAP bridge removed its own $92.4m FDIC windfall in FY2024 (adjusted EPS $6.74 against $7.66 reported) — a company that strips out its own good luck.
⚠️ Presentation, not a flag: the Verdant gross-up puts lease rental in non-interest income and $51.6m of depreciation in non-interest expense, so the 49.49% efficiency ratio carries a charge a pure lender would not have.
⚠️ RESOLVED: the 2026-09-22 "no earnings-call transcript" deviation is closed — seven usable quarters were read and the dodge log below is opened from the two most recent. BUT SEE THE ARCHIVE DEFECT under S9.
- S4Rivals and suppliersNot started
awaiting cluster pass
cluster-banks. - S5Who owns the stockDone
13F by manager at 2026-06-30 (BlackRock 8,389,477 sh / $817.1m, +68,179. State Street 3,023,908, +207,211. Ameriprise −762,616. Wellington −367,267.
Full notes
Citadel +323,663), 13D/G census (all 13G, no 13D, no activist), proxy ownership (20 officers and directors = 4.85%, the best insider alignment of the three names in this slot. BlackRock 14.72%, Vanguard 10.58%). Who is on the other side: index money plus small-cap growth managers, with a genuinely owner-aligned management team. Buyback: only $22.0m of stock repurchased in all of FY2026 — this is a capital-consuming grower, not a returner.
⚠️ the Q2-26 13F quarter is incompletely ingested (235 filers vs 382); no exit read from an absence.
⚠️
⚠️ the Sixth Street row is NOT a holding — see ACGL/dossier-2026-09-22.md §5.
- S6What workers and customers sayNot started
awaiting cluster pass
cluster-banksproduct test / expert call open OPEN (user)
- S7The case for it, and pricesNot started
awaiting cluster pass
cluster-banksconviction open OPEN (user) · valuation (bear/base/bull; moved from S8 2026-09-23): 10.7x FY2026A, 10.0x FY2027E and 9.0x FY2028E model EPS at $90.63 (2026-09-21); 1.62x current book, ~1.10x modelled FY2028 book.
Full notes
Base ~$110, bear ~$44, bull ~$150. ADV $35.1m. · valuation, base case re-read 2026-09-24 against S8: no haircut — the model is built on the 10-K's own numbers, not on guidance, and the record is 3 beat / 1 met / 1 missed. One assumption note: size announced-but-unclosed deposit purchases at what closes, not at what is signed — the only S8 miss is a $3.2bn agreement that closed at $1.9bn, and deposits are what fund the loan growth the model assumes.
- S8Do the bosses keep their word?Done
5 rows (2026-09-24, dossier-2026-09-24.md): 3 beat, 1 met, 1 MISSED
reliable. All 5 belong to executives still in place (Garrabrants President & CEO, Walsh EVP & CFO throughout; an Item-number sweep of every 8-K since 2026-02-02 returns no Item 5.02); 0 departed rows.
Full notes
The shape is the finding: everything Axos promised about ITSELF it delivered or beat; the one thing it promised about a COUNTERPARTY came in at 59%. Beat: FY2026 effective tax rate 24.17% against a guided 26-27%; organic loan growth ~+16% (total HFI +21.1%, $4,545,793k net) against "high single digits to low teens"; personnel + professional-services growth +$30.3m = 13.7% of the $221.7m revenue increase against a self-imposed ≤30% cap, which survives even after stripping the $22.0m settlement and Verdant lease income. Met: the March-quarter NIM "down in line with last quarter's guidance of around 10 basis points" (call 2026-04-30 [2], management scoring itself with two stated exclusions). failed MISSED — Capital One IRA deposits: "approximately $3.2 billion" (8-K 2026-04-23, Item 1.01) closed at "approximately $1.9 billion" (8-K 2026-09-21, Item 2.01), a $1.3bn / 41% shortfall. No filing explains it — including a 10-K filed 2026-08-20, four months after the agreement and a month before the close — and no call acknowledges it. The only analyst question on those deposits (their average cost, Kelly Motta, call 2026-04-30 [22]) was answered without a number [23]. This is the same fact the 2026-09-22 run finding recorded, now scored: the organic engine delivers and the acquired-deposit machine is the unproven part. open Open, not checkable from the filings: Verdant expense "approximately $8.5 million per quarter" (call 2025-10-30 [3]) and Verdant's "$150 million to $200 million growth target" [19] — neither is separately disclosed. A disclosure gap, recorded as one.
- S9What to watch nextDone
trigger written; next earnings expected ~2026-10-29 (DERIVED from filing cadence: fiscal Q1 results went 2025-10-30 and 2024-10-30. market data's calendar needs a paid plan and Equibles was unreachable)
Full notes
Tier 0 EDGAR sweep covers the feed (cik 1299709) and has already proved it works on this name
⚠️
⚠️ TRANSCRIPT ARCHIVE DEFECT, found 2026-09-24 — do not read
_transcripts/AX/FY2026-Q4as the July-2026 call.index.jsonrecords it ascall_date: 2026-07-30,status: ok,misses: 0, and its stored sha256 matches the bytes on disk — the archive is internally consistent and the file is exactly what market data served. What market data served is the fiscal-2025 fourth-quarter call: its own turn [1] says the call covers "the quarter and fiscal year ended 06/30/2025", it quotes the "$2 billion [pipeline] as of 07/25/2025" and the 26-27% tax guide "starting in the quarter ending 09/30/2025", and it never mentions Capital One — a deal announced 2026-04-23. It is a second, differently-transcribed copy ofFY2025-Q4(38,185 vs 36,787 chars, not byte-identical, so a hash check cannot catch it). The most recent AX call in the archive is FY2026-Q3 (2026-04-30). The FY2026 full-year call of 2026-07-30 is ABSENT; nothing said on it is scored anywhere in this file.transcripts.pyverifies sha256 and HTTP status but not that the delivered call matches the quarter requested; every market data transcript in this archive names its fiscal quarter in the operator's or IR's opening turn, so a cheap guard exists. Nothing in_transcripts/was edited — files are stored as received (README).
Sources and notes
Sources 7 documents
| Short name | Document | Where it came from | Read on |
|---|---|---|---|
| 10K-FY26 | Axos Financial 10-K for the fiscal year ended 2026-06-30, filed 2026-08-20 | https://www.sec.gov/Archives/edgar/data/1299709/000129970926000060/ax-20260630.htm | 2026-09-22 |
| 8K-2026-04-23 | Axos Financial 8-K filed 2026-04-23, Item 1.01 — purchase and assumption agreement with Capital One for ~$3.2bn of IRA deposits | https://www.sec.gov/Archives/edgar/data/1299709/000129970926000027/ax-20260422.htm | 2026-09-22 |
| 8K-2026-09-21 | Axos Financial 8-K filed 2026-09-21, Item 2.01 — completion of the Capital One IRA deposit acquisition (the Tier 0 promotion trigger) | https://www.sec.gov/Archives/edgar/data/1299709/000129970926000066/ax-20260921.htm | 2026-09-22 |
| 8K-FY26-RESULTS | Axos Financial 8-K filed 2026-07-30, Exhibit 99.1 — fiscal year 2026 results release | https://www.sec.gov/Archives/edgar/data/1299709/000129970926000052/pressrelease20260630axearn.htm | 2026-09-22 |
| DEF14A-25 | Axos Financial DEF 14A filed 2025-09-25 (ownership table as of 2025-09-16) | https://www.sec.gov/Archives/edgar/data/1299709/000129970925000174/ax-20250925.htm | 2026-09-22 |
| 13F-Q2-26 | Institutional holders by manager, 13F report date 2026-06-30 (ingestion incomplete — see the edge note) | python ledger/research/tools/threespread.py holders AX | 2026-09-22 |
| SIGNALS-2026-09-22 | Tier 0 EDGAR sweep promotion of AX on the 2026-09-21 8-K | ledger/research/queue.json (promoted, priority 2, as_of 2026-09-22) | 2026-09-22 |
Links to other companies 11 links
| From | Link | To | Date | Where it says so |
|---|---|---|---|---|
| Axos Bank | acquired | Capital One IRA savings and certificate-of-deposit book acquired by Axos Bank | 2026-09-21 | 8K-2026-09-21 — Item 2.01, Completion of Acquisition or Disposition of Assets, first paragraph |
| Capital One, National Association | event | Axos Bank | 2026-04-22 | 8K-2026-04-23 — Item 1.01, Entry into a Material Definitive Agreement, first paragraph |
| Office of the Comptroller of the Currency | regulator of | Axos Bank | 2026-06-08 | 10K-FY26 — Item 7 MD&A, recent developments paragraph preceding 'On May 2, 2026' |
| AX | supplier dependency | AX | 2026-06-30 | 10K-FY26 — Item 1A Risk Factors, 'Declining real estate values, particularly in California and New York…' |
| BDO USA, P.C. | auditor of | AX | 2026-06-30 | 10K-FY26 — Report of Independent Registered Public Accounting Firm on ICFR, scope paragraph |
| Verdant Commercial Capital, LLC | acquired | AX | 2025-09-30 | 10K-FY26 — Item 7 MD&A, recent developments — Verdant acquisition paragraph |
| Jenius Bank | acquired | Axos Bank | 2026-05-02 | 8K-FY26-RESULTS — Exhibit 99.1, Other Highlights — total deposits bullet |
| Financial Industry Regulatory Authority | litigation watch | Axos Clearing LLC | 2026-06-05 | 10K-FY26 — Notes to consolidated financial statements, Commitments and Contingencies — Other Matters |
| Gregory Garrabrants | officer of | AX | 2026-07-30 | 8K-FY26-RESULTS — Exhibit 99.1, CEO quote after the fiscal-year summary |
| Gregory Garrabrants | governance structure | AX | 2025-09-16 | DEF14A-25 — Security Ownership table, group row |
| BlackRock, Inc. | holds | AX | 2026-06-30 | 13F-Q2-26 — holders table, rows 1-3; WARNING lines beneath it |
History 4 entries
- 2026-09-22 — first pass, slot
names-insurance-1, model-drafted unattended by the cloud research routine. Tier 0 promotion (8-K of 2026-09-21) replacing RYAN in the batch. S1, S2, S3 (gate-minimum, no transcript — Equibles down, none on fool.com), S5 and S9 closed; S0/S4/S6/S7 left ⬜ awaitingcluster-banks; S8 modelled and left open. Run finding: the Capital One IRA deposit purchase that triggered the promotion closed at ~$1.9bn against the ~$3.2bn it was signed for — 59% of the agreed book, a $1.3bn shortfall that neither 8-K nor the 10-K explains, on the largest transaction of a company whose FY2026 deposit growth was $3.7bn of which $2.3bn was bought from Jenius Bank. Second finding: a FINRA arbitration panel awarded ~$49m against Axos Clearing on 2026-06-05 and the company has accrued $21m, 43%, on the strength of a petition to vacate — leaving ~$28m (~$0.49/share pre-tax) unaccrued; the $21m taken is $0.36 of the $8.48 diluted EPS and the whole of the Securities segment's expense increase. A second, smaller accrual ($7.0m) relates to the 2019 COR Securities acquisition. Third: FY2026 earnings quality is weaker than the headline — a $22.0m one-off legal settlement in non-interest income, Verdant lease income arriving with $51.6m of matching depreciation, and the 16%-yielding FDIC purchased-loan book running off at ~$45m of interest income a year. Evidence and access limits in dossier-2026-09-22.md. Nothing here is human-verified. - 2026-09-22 — re-clustered, same day, by the owner's split of
cluster-financials-global(WATCHLIST_PLAN.md §22). That slot held four banks, an asset manager, an Argentine energy company, a Mexican airport group and an ETF under one instrument; AX now sits inbankswith C, SYF and VBNK, whose pass iscluster-banks. The four waiting stages above point at it. Nothing here was re-scored: this run's own S1/S2/S3/S5/S8/S9 stand exactly as written, and AX still has no scuttlebutt in the repo until that cluster pass runs. - 2026-09-23 — stage redefinition, owner decision (WATCHLIST_PLAN.md §25): S8 is now Management credibility and is not yet run for this name; the old S8 valuation figures moved to the
valuation:line under S7, and any proposed buy price or position size was removed (the pipeline no longer drafts either). No research was redone. S2 capital allocation and S3 earnings quality are also pending, in the same backfill slot. - 2026-09-24 — backfill, slot
backfill-2026-09-23-6, model-drafted unattended by the cloud research routine. Wrote the three checks added on 2026-09-23 (WATCHLIST_PLAN.md §25) intodossier-2026-09-24.mdand updated this file: S8 management credibility (5 rows, 3 beat / 1 met / 1 missed — reliable), S2 capital allocation (compounds) and S3 earnings quality (3 flags). No other stage was redone; S0/S4/S6/S7 still awaitcluster-banks. Sources: the FY2026 10-K, the deposit 8-Ks of 2026-04-23 and 2026-09-21, an Item-number sweep of every 8-K since 2026-02-02 (no Item 5.02), XBRLcompanyfacts, and the first transcripts ever read on this name — seven usable quarters from the archive. Run finding — a tooling defect:_transcripts/AX/FY2026-Q4is not the 2026-07-30 call. market data served a second transcription of the fiscal-2025 Q4 call under the FY2026-Q4 key; the file's own opening turn says "the quarter and fiscal year ended 06/30/2025" and it never mentions Capital One.index.jsonis internally consistent and the sha256 matches, so nothing mechanical catches it, and the two files are not byte-identical. The July-2026 call is therefore absent from the archive and nothing said on it is scored here. Second finding: the Capital One IRA purchase is now scored as S8's one miss — announced at "approximately $3.2 billion" (8-K 2026-04-23) and closed at "approximately $1.9 billion" (8-K 2026-09-21), with nothing in between explaining it, including a 10-K filed four months after the agreement. The 2026-09-22 pass found the shortfall; this pass establishes that everything else management promised was delivered or beaten, which is what makes the shortfall legible: the organic engine works, the acquired-deposit machine is the unproven part. Third: Axos paid $30.3m net-settling RSU taxes in FY2026 (337,635 shares at $89.69) against $22.0m of buybacks (249,921 shares at $87.95), so shares outstanding rose for the first time since FY2022. Fourth: unrecognized tax benefits went $19.6m → $97.4m in the year the effective tax rate fell 29.42% → 24.17% — fully reserved, so conservative in presentation and a large two-sided item in substance. Fifth: the §25 bonus-metric read could not be done and is recorded as open, not passed: the FY2026 DEF 14A has not been filed and the 2025 proxy is the wrong year for this analysis. Nothing here is human-verified.
Full notes, 24 Sep 2026 21 min read
AX — dossier, 2026-09-24 (backfill: S8, S2 capital allocation, S3 earnings quality)
Slot backfill-2026-09-23-6. Model-drafted unattended by the cloud research routine; nothing here is human-verified. This file holds only the three checks added on 2026-09-23 (WATCHLIST_PLAN.md §25). Every other stage stands as written in dossier-2026-09-22.md. Anything from background knowledge is marked [background].
Sources this pass. EDGAR, User-Agent ddboard-research gorkem@celebico.com (CIK 1299709): the FY2026 10-K (0001299709-26-000060, filed 2026-08-20, fiscal year ended 2026-06-30) for the consolidated statements, the three-year performance-ratio table, the MD&A non-interest income and expense tables, the non-GAAP reconciliation, Item 5's repurchase disclosure, the income-tax note and the segment note; the deposit-purchase 8-Ks of 2026-04-23 (0001299709-26-000027) and 2026-09-21 (0001299709-26-000066); an Item-number sweep of every 8-K filed since 2026-02-02 (no Item 5.02 in the window — no executive change); and XBRL companyfacts for CIK 1299709. Transcripts from the archive ledger/research/_transcripts/AX/ — the first transcripts ever read on this name; the 2026-09-22 pass had none. Equibles: 0 calls. market data: 0 calls. 3spread: 0 calls. Nothing was fetched from the open web.
⚠️ Archive defect found this pass — _transcripts/AX/FY2026-Q4 is the wrong call
index.json records that file with call_date: 2026-07-30, fetched from https://api.market data/v2/company/earnings-calls/transcript/AX?year=2026&quarter=4, status: ok, misses: 0, and its stored sha256 matches the bytes on disk — the archive is internally consistent and the file is exactly what market data served. What market data served is the wrong call. Its own opening turn says so:
Joining us today for Axos Financial Inc. Fourth quarter and fiscal 25 financial results conference call are the company's President and Chief Executive Officer, Gregory Garrabrants and Executive Vice President, and Chief Financial Officer, Derrick Walsh. Gregory and Derrick will review and comment on the financial and operational results for the quarter and fiscal year ended 06/30/2025.
— FY2026-Q4, turn [1], which continues: "Gregory and Derrick will review and comment on the financial and operational results for the quarter and fiscal year ended 06/30/2025." It is a second, differently transcribed copy of the fiscal 2025 fourth-quarter call (2025-07-31), which the archive also holds correctly as FY2025-Q4. The two files are not byte-identical (38,185 vs 36,787 characters, 25 sentences shared verbatim) because they are two transcription passes over one audio recording; they discuss the same quarter, quote the same "$2 billion [loan pipeline] as of 07/25/2025", give the same "26% to 27%" tax guide "starting in the quarter ending 09/30/2025", and neither mentions Capital One, which was announced 2026-04-23 and would have dominated a July-2026 call.
Consequences, and they are the reason this is written at the top of the file rather than in a footnote:
- The most recent AX call in the archive is FY2026-Q3 (2026-04-30), not FY2026-Q4. The FY2026 full-year call of 2026-07-30 is absent. Nothing said on it is scored below.
- A pass that reads
FY2026-Q4without checking will date July-2025 statements to July-2026, which on this name means attributing a pre-Verdant, pre-Capital-One outlook to the quarter after both. The guides in that file are already scored here under their correct date. transcripts.pyverifies sha256 and HTTP status but has no check that the delivered call matches the quarter requested. A cheap guard exists: the operator's and IR's opening turns name the fiscal quarter and year in plain text on every market data transcript in this archive.- Nothing in
_transcripts/was edited. The README's rule is that files are stored as received, and they are; the finding belongs in the research record, which is here and inscorecard.md.
Quote discipline: every transcript quote was located by exact substring in the archived JSON bytes and copied from it; filing quotes were matched against a whitespace-collapsed text extraction of the EDGAR HTML. Turn numbers [n] are newline positions in the market data content string. Fiscal years end 30 June; FY2026 means the year ended 2026-06-30.
S8 — management credibility
How Axos guides: narrow, specific, and mostly about its own cost base. There is no EPS, revenue or ROE guidance in any of the seven usable calls. What management does put numbers on is a standing organic loan-growth range, an effective tax rate, a personnel-cost discipline expressed as a ratio, quarter-ahead net-interest-margin direction, and the size of announced transactions. Five of those are checkable against the FY2026 10-K and two 8-Ks.
Who promised, and who is still here. market data labels turns by name only; roles are read off EDGAR. Gregory Garrabrants, President and Chief Executive Officer, and Derrick K. Walsh, Executive Vice President and Chief Financial Officer, are both signatories of the FY2026 10-K (power-of-attorney block, 0001299709-26-000060) and have held those roles across the entire window. An Item-number sweep of every 8-K filed from 2026-02-02 to 2026-09-21 returns no Item 5.02 — no departure, no appointment. 0 departed rows; the record belongs entirely to the people still running the company. On this name that cuts against them as much as for them, because the one miss is theirs and nobody has left to absorb it.
Say vs do (guides made 2024-10-30 → 2026-04-30, plus two filed commitments)
| # | Promise (period · metric · figure) | Made where · by whom | Result · locator | Score |
|---|---|---|---|---|
| 1 | FY2026 (from the quarter ending 2025-09-30) · effective corporate tax rate ≈26–27%, down from 29–30% | call 2025-07-31 [3] · Derrick Walsh (EVP & CFO): "Starting in the quarter ending September 30, 2025 and going forward, we expect our corporate tax rate to be approximately 26% to 27%, an improvement of 3 percentage points from the previously 29% to 30%." | 24.17% — FY2026 10-K MD&A: "our effective tax rates were 24.17%, 29.42% and 29.19% for the fiscal years ended June 30, 2026, 2025, and 2024". Below the guided range, which for a cost line is the favourable direction | beat — but read it with earnings-quality flag 3 below |
| 2 | next 12 months from 2025-07-31 · organic loan growth at "the midpoint to high end of our high single digits to low teens" range, excluding the FDIC portfolio and any acquisitions | call 2025-07-31 [3] · Walsh: "we believe that we will be able to grow loan balances organically at the midpoint to high end of our high single digits to low teens year-over-year growth target over the next 12 months, excluding the impact of the loan portfolio purchased from the FDIC or any other potential loan or asset acquisitions" | +21.1% total — loans held for investment $21,551,697k (2025-06-30) → $26,096,037k (2026-06-30); net growth of $4,545,793k against $1,818,225k the prior year (10-K performance-ratio table). Verdant contributed "approximately $1.0 billion of loans and leases" at 2025-09-30 (10-K), so organic is roughly +16% — above the top of the original range even after removing the acquisition | beat; the range was raised mid-year to "low to mid-teens" (call 2025-10-30 [19], Garrabrants: "we raised our guidance on loan growth"), which is an upward revision, announced |
| 3 | FY2026 · personnel and professional-services cost growth not to exceed 30% of the growth in net interest income plus non-interest income | call 2025-04-30 [12] · Gregory Garrabrants (President & CEO): "we're targeting next year that personnel expenses will go up no more than 30% of the combination of net interest and noninterest income"; restated on 2025-07-31 [9]: "What we had previously set as a target that we wanted to ensure that our personnel and professional services cost growth did not exceed 30% of our net interest income and noninterest income growth" | 13.7% — salaries and related costs $297,955k → $323,046k and professional services $37,572k → $42,806k, i.e. +$30,325k, against revenue (net interest income plus non-interest income) of $1,258,838k → $1,480,493k, +$221,655k. The cap was $66,497k (10-K non-interest income and expense tables) | beat — and it survives a strict read: stripping the $22.0m legal settlement, and even a further ~$70m for Verdant lease income, still leaves the cap above the actual |
| 4 | quarter ending 2026-03-31 · net interest margin down ≈10bp | guided on the call of 2026-01-29 and scored by management on the call of 2026-04-30 [2] · Gregory Garrabrants (President & CEO): "Excluding the impact from the prepayments of FDIC-purchased loans and 2 fewer days in the quarter ended March 31, our net interest margin was down in line with last quarter's guidance of around 10 basis points." | in line, on management's own reading, with two named exclusions stated in the same sentence | met |
| 5 | Capital One IRA deposits — "approximately $3.2 billion", expected to close in calendar 2026 | 8-K filed 2026-04-23 (0001299709-26-000027), Item 1.01, signed for the registrant: "Axos Bank... entered into a purchase and assumption agreement... with Capital One, National Association... to acquire individual retirement accounts ("IRAs") with an aggregate balance of approximately $3.2 billion" | approximately $1.9 billion — 8-K filed 2026-09-21 (0001299709-26-000066), Item 2.01: "The Bank acquired approximately $1.9 billion of IRA deposits from Capital One and received cash for the aggregate deposit balance, less a negotiated premium." 59% of the announced size; a $1.3bn shortfall. Neither 8-K explains it and the FY2026 10-K, filed between the two, does not address it | missed |
Not scored (open, and not checkable from the filings).
- Verdant non-interest expense "approximately $8.5 million per quarter" (call 2025-10-30 [3], Walsh). Three quarters of ownership in FY2026 implies ~$25.5m. The 10-K does not disclose Verdant expense separately: it attributes $51.6m of the $143.0m expense increase to operating-lease depreciation "following the Verdant acquisition", part of the $42.8m G&A increase to "higher loan and lease servicing expenses following the Verdant acquisition", and part of the $25.1m salaries increase to "including the impact of the Verdant acquisition". The guide was explicitly about operating expense, not lease depreciation — on the same call Walsh said "the income from the operating leases will be recorded in noninterest income" — so it cannot be scored either way without a segment disclosure that does not exist. Open, and a disclosure gap.
- Verdant "$150 million to $200 million growth target" (call 2025-10-30 [19], Garrabrants: "We put that $150 million to $200 million growth target out there. I think they can exceed that"). Not separately disclosed. Open.
- Allowance of "approximately 1.5%" on all new loans and leases (call 2025-10-30 [3]). Total ACL/loans is 1.34% (10-K), which is a portfolio average and not the same measure. Not scored.
Tally: 5 rows — 3 beat, 1 met, 1 missed. All 5 belong to executives still in place; 0 departed rows.
The shape matters more than the count. Everything Axos promised about itself, it delivered or beat: the tax rate, the loan growth, the cost ratio, the margin. The one thing it promised about a counterparty came in at 59%. That is the pattern to carry forward, and it is consistent with what scorecard.md already says about the business: the organic engine works and the acquired-deposit machine is the part that has not been proved.
Two notes on the miss, because it is the row that decides the verdict:
- It is not disclosed as a miss anywhere. The April 8-K says $3.2bn, the September 8-K says $1.9bn, and nothing in between — including a 10-K filed 2026-08-20, four months after the agreement and one month before the close — tells a reader that the expected size had changed. A holder who read only the announcement would have carried $3.2bn into a model for five months.
- The one analyst question on the record about those deposits was not answered either. Call 2026-04-30 [22], Kelly Motta (KBW) asked for "the average cost of those" Capital One deposits; the answer [23] covered funding capacity, marketing expense and NIM philosophy and gave no cost. See the dodge log.
Dodge-log roll-up (§3.1). scorecard.md carried no dodge log — this is the first pass on this name with any transcript at all. It is opened from the two most recent usable calls, FY2026-Q2 (2026-01-29) and FY2026-Q3 (2026-04-30); the file labelled FY2026-Q4 is the defective duplicate described at the top.
- Sizing Verdant's fee income — asked in two consecutive quarters, no number either time. Q2 (2026-01-29): Gary Tenner (D.A. Davidson) asked for "a follow-up on the Verdant impact on the fees and expenses" [33]; Walsh answered with the "$200 million a little over $200 that we classified as operating leases last quarter" [34] — a balance, not a run rate. Q3 (2026-04-30): Kelly Motta asked how to think about operating leases versus on-balance-sheet and whether to expect further fee growth [26]; Walsh: "it is kind of tough to tell... I can't give you a specific number or reference as to how that fee income will grow... I wouldn't, I guess, model it too significantly from that standpoint, given it is only 1/6 of the origination volume" [27]. Two quarters, two analysts, one topic, no figure — and the topic is precisely the line
scorecard.mdS2 flags as making FY2026 non-interest income (+78.2%) unrepeatable as stated. Management's framing is honest uncertainty rather than evasion, and the 1/6 datum is genuinely useful, but under §3.1 the count is what it is: a thread to pull. - The cost of the Capital One deposits — dodged once, on the deal that then closed 41% short. Call 2026-04-30 [22]/[23], as above. One quarter, so a note rather than a finding — but it is a note on the same transaction as S8's only missed row, and the next chance to test it is the fiscal-Q1 2027 call (~2026-10-29), which is already what
scorecard.md's ➡️ next action is waiting for. - Not a dodge, and worth recording for the tone. Asked a broad question about how Axos manages so many initiatives at once, Garrabrants opened "I'll dodge it then I'll answer it" (2026-01-29 [24]) — and then answered it. Self-labelled, then delivered.
Verdict — reliable. Four of five scored guides landed at or better than the figure, the two most demanding of them (a self-imposed cost ratio and an organic growth range) were beaten with room, and no executive has left. It is not conservative, because the single largest commitment in the window — a $3.2bn deposit purchase — closed at $1.9bn with no explanation in any filing and no acknowledgement on any call; and it is nowhere near over-promises at one miss in five. Consequence for S7: no haircut to the base case, which is built on the 10-K's own numbers rather than on guidance. One assumption note is added: any model that funds future loan growth with announced-but-unclosed deposit purchases should size them at what closed, not at what was signed.
S2 — capital allocation
All figures $000s, fiscal years ended 30 June. Sources: XBRL companyfacts (CIK 1299709) — OCF NetCashProvidedByUsedInOperatingActivities, buybacks PaymentsForRepurchaseOfCommonStock, acquisitions PaymentsToAcquireBusinessesNetOfCashAcquired, SBC ShareBasedCompensation, GoodwillImpairmentLoss, StockholdersEquity, CommonStockSharesOutstanding, WeightedAverageNumberOfDilutedSharesOutstanding — plus the FY2026 10-K's Item 5 repurchase disclosure and performance-ratio table.
| FY2022 | FY2023 | FY2024 | FY2025 | FY2026 | 5y total | |
|---|---|---|---|---|---|---|
| Operating cash flow | 216,622 | 196,706 | 305,477 | 490,331 | 557,597 | 1,766,733 |
| Buybacks | 0 | 48,963 | 96,286 | 58,203 | 21,980 | 225,432 |
| Dividends | — | — | — | — | — | 0 — Axos pays none |
| Acquisitions (cash, net of cash acquired) | 54,597 | 5,531 | 0 | 0 | 474,448 (Verdant) | 534,576 |
| Stock-based compensation | 21,242 | 26,100 | 35,194 | 41,971 | 44,255 | 168,762 |
| SBC as % of revenue | — | — | 2.97% | 3.33% | 2.99% | |
| Goodwill impairment | 0 | 0 | 0 | 0 | 0 | 0 |
| Diluted weighted-average shares (m) | 59.52 | 59.69 | 57.51 | 56.86 | 56.66 | −4.8% |
| Shares outstanding, year end (m) | 59.78 | 58.94 | 56.89 | 56.48 | 56.73 |
Where the cash went: into the balance sheet, which is the right answer for this business and is also the constraint. $225.4m returned over five years against $1,766.7m of operating cash flow — 12.8%, and no dividend has ever been paid. The other 87% funded asset growth: total assets $14.27bn (2021-06-30) → $29.96bn (2026-06-30), +110%, and equity $1,401m → $3,166m, +126%. For a bank compounding book value at 19% a year that is the correct allocation, and scorecard.md already names the price of it: Bank-level CET1 fell 12.74% → 12.42% → 10.93% over the same three years. Axos is growing the balance sheet faster than it generates capital, and the buyback is the shock absorber — $96.3m (FY2024) → $58.2m (FY2025) → $22.0m (FY2026), falling every year as the growth accelerated.
Buybacks at what price — and what outran them. FY2026: "the Company repurchased a total of $22.0 million or 249,921 common shares at an average price of $87.95 per share and there remains $126.1 million of availability under the program as of June 30, 2026" (10-K Item 5). Against book value per share of $55.81 at the same date that is 1.58x book, and against tangible book of $50.96, 1.73x.
The number beside it is the one worth carrying: in the same fiscal year the company retained 337,635 restricted-stock-unit shares "converted to cash at the average rate of $89.69 per share to fund the grantee's income tax obligations" (10-K Item 5) — $30.3m, i.e. Axos spent 38% more cash net-settling employees' equity-award taxes than it spent buying back stock. With $44.3m of stock-based compensation against $22.0m of repurchases, the arithmetic showed up where it always does: shares outstanding rose in FY2026 for the first time since FY2022, 56,483,617 → 56,734,049. The five-year count is still down 4.8% on a diluted weighted-average basis; the direction has turned.
What the acquisitions were worth. One material acquisition in the window: Verdant Commercial Capital, $474.4m of cash at 2025-09-30 for "approximately $1.0 billion of loans and leases (including $211.0 million of PCD assets) and $212.6 million of equipment under operating lease arrangements" (10-K). It added roughly $46.7m of goodwill (segment goodwill $97,673k → $144,330k) and carried $7.8m of day-one CECL provision through the income statement. Goodwill impairment across five years: zero, and nothing from the 2022 or 2023 tuck-ins has been written down either. The FY2024 FDIC loan purchase — bought at a discount, booked as a $92.4m gain on purchase — is the other side of the same capability and is the highest-return capital deployment on this five-year record.
What the bonus is paid on. Not read this pass, and the reason is on the record: the 2026-09-22 dossier flagged that the available proxy was the 2025 DEF 14A (filed 2025-09-25) and that the FY2026 proxy "was not yet on the index". It still is not — the most recent DEF 14A for this registrant remains 0001299709-25-000174. The §25 EPS-pay-plus-buybacks test therefore cannot be run on the fiscal year being analysed, and this pass does not run it on a stale one. What the FY2026 10-K does say, in Walsh's own words on the call of 2026-04-30 [27], is that "both the management team from an incentive standpoint and our business operations back office support are incented to help support and grow that business" — i.e. Verdant origination is inside the incentive plan. That is a pointer, not a measure. This is the one part of the check that is open, and it is open for a stated reason: the document does not exist yet.
Returns (bank form — return on equity and book value per share compounded with dividends, per SKILL.md; there are no dividends).
| FY2024 | FY2025 | FY2026 | |
|---|---|---|---|
| Net income ($000s) | 450,008 | 432,908 | 490,372 |
| Return on average common equity (10-K) | 21.64% | 17.30% | 16.32% |
| Return on average assets (10-K) | 2.08% | 1.82% | 1.76% |
| Book value per share | $40.26 | $47.46 | $55.81 |
| Tangible book value per share (10-K) | $37.26 | $44.60 | $50.96 |
Book value per share $23.62 at 2021-06-30 ($1,400,936k over 59,317,944 shares) → $55.81 at 2026-06-30: +18.8% a year compounded, with no dividend paid, so that is the entire return on book. Tangible book grew 17.0% then 14.3% in the last two years.
The incremental return is the finding. Net income rose $249.7m over the five years ($240.7m → $490.4m) while common equity rose $1,765m ($1,401m → $3,166m): an incremental return on equity of 14.1%, against a five-year average ROE of about 18%. New capital is earning materially less than old capital, and every efficiency series in scorecard.md S2 says the same thing from a different angle — NIM 4.90% → 4.70%, efficiency ratio 46.84% → 49.49%, ROA 1.82% → 1.76%, ROE 17.30% → 16.32%. The growth is real and it is getting more expensive.
Verdict: compounds. 18.8% a year of book value per share with no dividend and no equity issued, zero goodwill impairment in five years, an acquisition record that includes the FDIC portfolio, and a share count down 4.8%. The two things that qualify it, neither enough to change the verdict: the incremental return (14.1%) is four points below the average, and FY2026 is the first year in four in which the share count rose, because the buyback was cut to $22.0m while RSU net-settlement cost $30.3m and SBC ran at $44.3m. If CET1 keeps falling, the buyback is the first thing that goes to zero, and then dilution is the only direction.
S3 addendum — earnings quality
Sources: the FY2026 10-K non-GAAP reconciliation, MD&A non-interest income and expense tables, performance-ratio table, income-tax note and premises note; XBRL companyfacts.
- The non-GAAP bridge. Flag 1 — and it is the mild kind. FY2026 runs net income $490,372k → favourable legal settlement $(22,000)k → FINRA arbitration matter $21,000k → acquisition-related costs $13,689k → Verdant provision for credit losses $7,765k → income tax effect $(4,944)k → adjusted earnings $505,882k, i.e. $8.74 against $8.48 of diluted GAAP EPS. The legal pair is honest: a $22.0m gain is removed in the same bridge that adds back a $21.0m loss, and the net of the two is $(1.0)m. What earns the flag is the line beneath them — "acquisition-related costs", which footnote 2 says "includes amortization of intangible assets", added back in three years of three and growing: $10,843k (FY2024), $7,408k (FY2025), $13,689k (FY2026). That is the §25 test-1 shape exactly: a category presented as transaction-specific that has never been absent. It is small — 0.9% of revenue, $0.24 of the $8.74 — and Axos has made an acquisition in three of the last five years, so it is logged as mild. The bridge's larger merit should be recorded beside it: the FY2024 $92.4m FDIC gain on purchase was removed, taking adjusted EPS to $6.74 against $7.66 reported. A company that strips its own windfall out of the adjusted number is not managing that number upward.
- Stock-based compensation. Clean, with the note from S2. FY2026 SBC $44,255k = 2.99% of revenue and 7.9% of operating cash flow, and flat as a share of revenue across three years (2.97% / 3.33% / 2.99%). The FCF-after-SBC test does not apply to a bank. What does apply is whether buybacks offset it, and in FY2026 they did not: $22.0m of repurchases plus $30.3m of RSU net-settlement against $44.3m of SBC, with shares outstanding up 250,432 (S2).
- Cash conversion. Clean. Operating cash flow over net income 0.68x (FY2024), 1.13x (FY2025), 1.14x (FY2026). Accruals ((net income − OCF) / average total assets): +0.67%, −0.24%, −0.25% — positive in FY2024, when the $92.4m non-cash FDIC purchase gain ran through income, and negative since. The series behaves exactly as the non-GAAP bridge says it should.
- Working capital — the bank equivalent. Clean on the outturn, watch the provision. SKILL.md points banks at distributable earnings against GAAP; Axos does not report a distributable-earnings measure, so the readable substitute is the credit build. Provision for credit losses $32,500k (FY2024) → $55,745k (FY2025) → $101,105k (FY2026) — a tripling in two years — against net charge-offs of 0.05% → 0.13% → 0.18% of average loans and an allowance that rose $290,049k → $347,375k. Part of the FY2026 step is mechanical: $7.8m of Verdant day-one CECL plus 1.5% allowance on ~$4.5bn of net new loans is most of it. Credit itself improved — non-performing assets 0.71% → 0.53% of assets, nonaccruals 0.79% → 0.60%. Nothing here is being held back or released to make a number.
- Costs moved onto the balance sheet. Clean. Capitalised software $159,878k → $189,265k, +18.4%, against revenue +17.6% — in line, not outrunning it. The large new balance-sheet item is the $212.6m of equipment under operating lease from Verdant, and it is not a cost deferral: its rental income is in non-interest income and its $51.6m of depreciation is in non-interest expense, both disclosed and both in the same year. It is a gross-up, and the thing it distorts is the efficiency ratio rather than earnings —
scorecard.md's 49.49% carries $51.6m of depreciation that a pure lender would not have. - Sector equivalent — the tax line. Flag 2, and it is the one to carry. The effective rate fell 29.42% → 24.17%, which S8 scores as a beat, and the 10-K attributes it to "lower state and local income taxes, including the effect of the changes in the State of California tax laws passed in June 2025, as well as higher income tax credits". Alongside it, in the income-tax note: "As of June 30, 2026 and 2025, unrecognized tax benefits totaled $97.4 million and $19.6 million, respectively, that, if recognized, would favorably impact the effective tax rate." That is a $77.8m increase in one year, to a figure equal to 20% of FY2026 net income and $1.72 a share. The direction of the accounting is conservative, not aggressive — these amounts are reserved, so they do not flatter reported earnings, and the note says why: "Amended returns were filed to preserve claims for refund. Due to the uncertainty involved in this claim, management recognized a 100% reserve against it during the fiscal years ended June 30, 2025 and 2026." The flag is about position, not presentation: a bank that has quintupled its unrecognised tax benefits in a year while its effective rate fell five points is taking tax positions whose resolution is a real two-sided item nobody is modelling.
- The unaccrued half of the FINRA award. Flag 3, carried forward from the first pass and still true in the FY2026 10-K. A FINRA arbitration panel awarded roughly $49m against Axos Clearing on 2026-06-05; the company has accrued $21.0m — 43% — on the strength of a petition to vacate, leaving about $28m (~$0.49 a share pre-tax) unaccrued, and it added a further $7.0m accrual for a matter arising from the 2019 COR Securities acquisition. Together they are the whole of the $42.8m general-and-administrative increase alongside Verdant servicing costs.
EQ summary: 3 flags — acquisition-related costs (largely intangible amortisation) added back in three years of three and growing, $10.8m → $7.4m → $13.7m (mild); unrecognised tax benefits up $77.8m to $97.4m, 20% of net income and $1.72 a share, in the year the effective rate fell 5.25 points; and ~$28m of a ~$49m FINRA award still unaccrued. Clean: cash conversion 1.14x with negative accruals, SBC flat at 3.0% of revenue, capitalised software growing in line with revenue, credit metrics improving while the provision builds, and a non-GAAP bridge that removed its own $92.4m windfall in FY2024.
Full notes, 22 Sep 2026 16 min read
AX — Axos Financial, Inc. — dossier, 2026-09-22
Slot names-insurance-1 (Tier 0 promotion). Model-drafted unattended by the cloud research routine (Opus 5), following .claude/skills/stock-sourcing-pipeline/SKILL.md. Nothing here is human-verified. Sourced facts carry the document they were read from; [background] marks background knowledge and [search-summary] marks search-derived material. Private: this directory never reaches dist/.
Access this run is recorded once in ACGL/dossier-2026-09-22.md and applies identically here: EDGAR and 3spread reachable, Equibles unreachable (connector failed to connect), market data prices yes / calendar paid, fool.com reachable but carrying no Axos transcript.
Why this name is here. queue.json promoted AX on 2026-09-22 at priority 2 for an "8-K filed 2026-09-21 (current report)". Under the skill's promotion rule a promoted name with no scorecard.md replaces the last member of the slot's batch, so AX took RYAN's place in names-insurance-1; RYAN returns to a later insurance slot. Two other names remain promoted and unhandled: PUMP (priority 2) and SNDK (priority 4). The skill also says a promoted name was promoted for a reason and that reason is the lead for the run. It was.
§1 — The finding: the deal closed 41% smaller than it was signed, and no filing says why
Three filings, one transaction, in date order:
| Date | Filing | What it says the deal is |
|---|---|---|
| 2026-04-22 (8-K filed 04-23) | Item 1.01 | IRAs "with an aggregate balance of approximately $3.2 billion" |
| 2026-06-08 | (recited in the 10-K) | OCC approves it |
| 2026-08-20 | 10-K FY2026, Item 7 | still "approximately $3.2 billion of deposits … expected to close in calendar year 2026" |
| 2026-09-21 (8-K filed same day) | Item 2.01 | "The Bank acquired approximately $1.9 billion of IRA deposits from Capital One" |
$1.3 billion, 41% of the agreed book, did not arrive, and neither the completion 8-K nor anything else on the index quantifies, explains or even acknowledges the difference. The completion filing is four sentences long.
The ordinary explanation is benign and probably right: deposit purchase and assumption agreements are customarily struck on a balance measured around signing, with the actual transfer sized at closing, so five months of IRA attrition — customers who move rather than be transferred to a bank they did not choose — reduces what changes hands. Nobody is at fault in that story. But the size is what makes it a finding. A 5% or 10% slippage is attrition; 41% over five months on a book of IRA savings and CDs — among the stickiest retail deposits that exist, since moving one is a custodial transfer with tax paperwork — is a different order of number, and the company chose not to address it in the filing that announced the close.
Why it matters to this name in particular. Axos's FY2026 headline deposit growth was $3.7bn, +17.9%, and the results release says so in a bullet that also says $2.3bn of it was bought (the Jenius Bank consumer book, closed 2026-05-02). Organic deposit growth was therefore about $1.4bn, roughly 6.7%. Acquired deposits are the funding growth story at this bank, and on its largest such transaction the machine delivered 59% of what was announced. That is not a credit event or an accounting concern. It is a question about the durability of the growth algorithm, and it is the single thing the next 10-Q can answer.
Note also what the premium is: unknown. Both 8-Ks say the Bank receives cash for the deposits "less a negotiated premium" and neither discloses the rate, so the cost per dollar of what actually arrived cannot be computed from the public record.
§2 — The second finding: a $49m award accrued at $21m
Buried in the contingencies note, not in the release, not in the MD&A narrative:
"Other Matters. On June 5, 2026, a FINRA arbitration panel rendered an award of approximately $ 49 million against Axos Clearing, a downstream subsidiary of Axos Financial in the FINRA arbitration forum styled Acquarulo et al. v. Axos Clearing LLC ."
The matter arose from clearing services Axos Clearing provided to a third-party introducing broker-dealer "which is alleged to have committed certain sales practice violations"; Axos Clearing asserts it had no duty to supervise that firm's sales practices, has filed a petition to vacate, and:
"Axos Clearing has filed a petition to vacate the award and, based on its assessment of the likelihood of potential outcomes, the Company has accrued a liability of $ 21 million in “General and administrative expense” in the Consolidated Statement of Income for the year ended June 30, 2026."
So $21m of a ~$49m award is accrued — 43% — and the remaining ~$28m (≈$0.49 per share before tax) rests on the vacatur succeeding. That is a disclosed, quantified, binary contingency with a court date attached, and it is not mentioned in the earnings release except as an expense the CFO backs out.
The $21m that was taken is not small either: it is $0.36 of the $8.48 diluted EPS (the company's own non-GAAP bridge says so), and it is the entire explanation of the Securities Business Segment's 23.1% rise in non-interest expense. In the same expense bullet sits a second legal accrual — $7.0m "for developments in a matter related to the Company's acquisition of COR Securities in fiscal year 2019". Two litigation accruals in the clearing business in one fiscal year is worth a line in the next pass; it is not yet a thesis.
§3 — S2 kill test
| Fiscal year ended 30 June | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
| Net interest income $m | 607.2 | 783.1 | 961.4 | 1,127.8 | 1,246.9 |
| Non-interest income $m | 113.4 | 120.5 | 222.7 | 131.1 | 233.6 |
| Net income $m | 240.7 | 307.2 | 450.0 | 432.9 | 490.4 |
| Diluted EPS | $3.97 | $5.07 | $7.66 | $7.43 | $8.48 |
| Total assets $bn | 17.40 | 20.35 | 22.86 | 24.78 | 29.96 |
| Deposits $bn | 13.95 | 17.12 | 19.36 | 20.83 | 24.57 |
| Equity $m | 1,643 | 1,917 | 2,291 | 2,681 | 3,166 |
| Diluted weighted shares m | 59.52 | 59.69 | 57.51 | 56.86 | 56.66 |
(XBRL companyconcept, CIK 1299709. FY2024's $7.66 EPS includes a $92.4m gain on the FDIC loan purchase, $1.57/share, which is why FY2025 looks like a decline and is not one.)
Share count over five years: 61.90m (FY2019) → 56.66m (FY2026), −8.5%, entirely through modest repurchase, no issuance anywhere. Clean.
The ratios, which tell a different story from the earnings line (10-K Item 7, three-year selected-ratio table):
| FY2026 | FY2025 | FY2024 | |
|---|---|---|---|
| Return on average common stockholders' equity | 16.32% | 17.30% | 21.64% |
| Net interest margin | 4.70% | 4.90% | 4.62% |
| Efficiency ratio | 49.49% | 46.84% | 43.59% |
| Equity to assets | 10.57% | 10.82% | 10.02% |
| Axos Bank CET1 / RWA | 10.93% | 12.42% | 12.74% |
| Net charge-offs / average loans | 0.18% | 0.13% | 0.05% |
| Non-performing assets / total assets | 0.53% | 0.71% | 0.51% |
| ACL / loans held for investment | 1.34% | 1.36% | 1.34% |
| ACL / nonaccrual loans | 221.07% | 170.23% | 229.84% |
Earnings up 13%, and returns, margin, efficiency and bank capital all moving the other way. Assets grew 20.9% in a year while the Bank's CET1 fell 149bp; Axos is growing its balance sheet faster than it generates capital and returns almost none of what it earns ($22.0m of buyback against $490.4m of net income).
Three things that make FY2026 earnings lower-quality than the headline. All three are in the filings and none is in the release's framing:
- Non-interest income +78.2% to $233.6m is attributed to three things: Verdant operating-lease rental income, "A $22.0 million legal settlement in our favor reached in March 2026" — a one-off, and the company's own non-GAAP bridge removes it — and rental income from a commercial office complex the Company bought for ~$125m in January 2026 and intends to occupy itself. The Verdant lease income arrives with its own cost: depreciation and amortisation rose from $29.0m to $80.6m, +$51.6m, "primarily due to depreciation on equipment under operating leases following the Verdant acquisition".
- The FDIC purchased-loan book is running off, at a 16% yield. Average balance $973.1m (FY2025) → $712.7m (FY2026), interest income $160.6m → $115.4m, yield 16.51% → 16.19%. That is a $45m/year disclosed and dated headwind on a pre-tax base of about $750m, and another $40-50m is likely in FY2027 on the same trajectory. Nothing in the reported guidance replaces it; the new loan book yields 7.44%.
- $28m of the FINRA award is unaccrued (§2).
Survives or dies: survives, with room and without drama. A 10.57% equity/assets ratio, a Bank CET1 of 10.93% against a 7% requirement [background], non-performing assets at 0.53% of assets and falling, an allowance covering nonaccruals 2.2x, and $490m of annual earnings against no holding company funding stress. The question this name poses is not solvency; it is whether a 16% ROE franchise is on its way to a 13% one.
§4 — S3 gate: the three biggest risks management itself discloses
Axos's Item 1A is grouped rather than ranked, so these are chosen by the size of the disclosed exposure, not by position — and each is quantified by the company.
- Geographic real-estate concentration. "At June 30, 2026, approximately 37.4% and 33.5% of our real estate loan portfolio was secured by real estate located in New York and California, respectively." (Item 1A, "Declining real estate values, particularly in California and New York, could reduce the value of our loan and lease portfolio…") Seventy-one per cent of the real estate book in two states, on a balance sheet where real estate secures the majority of loans.
- Multifamily. "At June 30, 2026, our multifamily residential loans were $2.5 billion or 9.5% of our loan portfolio." (Item 1A, "Many of our mortgage loans are multifamily residential loans and defaults on such loans would harm our business.") New York multifamily has its own rent-regulation story [background] and the company's own disclosure puts the concentration and the geography in adjacent risk factors.
- Interest rates — placed first in the whole item. "Changes in interest rates could adversely affect our performance." is the opening risk factor of Item 1A, under "Risks Relating to Macroeconomic Conditions". For a bank running a 4.70% NIM off floating-rate commercial assets funded with retail deposits, this is the single largest swing factor in the P&L and management ranks it first.
The loan book, and the timing risk inside it (Item 7, loan portfolio composition and maturity tables):
| At 30 June | FY2026 $m | % | FY2025 $m | % |
|---|---|---|---|---|
| Single family — mortgage & warehouse | 4,581.8 | 17.6% | 4,395.3 | 20.4% |
| Multifamily and commercial mortgage | 2,484.3 | 9.5% | 2,940.7 | 13.6% |
| Commercial real estate | 8,867.9 | 34.0% | 6,937.2 | 32.2% |
| Commercial & industrial — non-RE | 9,495.6 | 36.4% | 6,795.5 | 31.6% |
| Auto & consumer | 666.5 | 2.5% | 483.0 | 2.2% |
| Total loans held for investment | 26,096.0 | 100% | 21,551.7 | 100% |
The mix moved hard in one year: CRE +27.8%, C&I non-RE +39.7% (including Verdant), while multifamily shrank 15.5%. And the maturity table adds the part a composition table hides: of the $8,867.9m of commercial real estate, $576.0m matures in under three months and $1,964.7m within a year — $2,540.7m, 29% of the CRE book, refinancing inside twelve months, with none of it scheduled beyond five years. Short-dated CRE concentrated in New York and California is precisely the exposure BDO's critical audit matter points at.
Auditor and red-flag sweep. BDO USA, P.C., PCAOB ID 243, San Diego. Sole critical audit matter: the allowance for credit losses, $347,375 thousand at 2026-06-30, and specifically management's "significant judgments regarding the weighting of the economic forecast scenarios within the Commercial Real Estate (including construction lending) and Com[mercial]" portfolios. ⚠️ The FY2026 internal-control opinion excludes Verdant: "Verdant Commercial Capital, LLC constituted 6% of total assets, as of June 30, 2026, and 7% of net revenues for the year then ended", and management did not assess it either. That is a permitted first-year scope exclusion, not a finding — but 7% of FY2026 revenue sits outside both ICFR opinions and the next 10-K will be the first to cover it.
Filing index back to 2010: one NT 10-K, filed 2022-08-30 (FY2022 late filing) — four years ago and not repeated; SEC comment-letter correspondence in 2018 and again in 2023, the most recent exchange closing 2023-04-27; no 10-K/A and no 10-Q/A. Of the three names in this slot, AX is the only one with a late-filing notification in its history, and it is old.
§5 — S5 ownership
13F holders at 2026-06-30 (threespread.py holders AX): BlackRock 8,389,477 sh / $817.1m (+68,179); State Street 3,023,908 / $294.5m (+207,211); Charles Schwab IM 751,934 (+19,853); Ameriprise 654,362 (−762,616); Invesco 617,610 (+71,304); Emerald 536,162; Wasatch 515,576 (−100,805); Geneva 477,009 (−131,072); Wellington 470,586 (−367,267); Goldman Sachs 453,375; Citadel 423,753 (+323,663); Kennedy Capital 426,294 (+194,907).
⚠️ Quarter incompletely ingested — 235 filers against 382; both Vanguard entities (4.91m and 3.09m shares), American Century (2.51m), FMR (2.44m), Dimensional (2.18m), Geode (1.96m), Davis (1.54m), Franklin, Northern Trust and Morgan Stanley carry no filing in it. No exit may be read from an absence.
⚠️⚠️ The Sixth Street row (751,934 shares, $73.2m, "NEW", identical to Schwab's) is not a holding — the filing defect documented in ACGL/dossier-2026-09-22.md §5 and now warned on mechanically by threespread.py.
13D/G: all 13G, no 13D, no activist. A 13G/A filed 2026-03-26 sits in the same window as the Vanguard internal realignment documented in the APH pass (names-ai-capex-3), which is the known cause of Vanguard disappearing from 13F tables across issuers — worth remembering before reading Vanguard's absence here as anything.
Proxy (DEF 14A 2025-09-25, as of 2025-09-16): BlackRock 8,329,879 = 14.72%, The Vanguard Group 5,985,476 = 10.58%, on 56,595,223 shares outstanding. "All current directors and executive officers as a group (20 persons) (8)" hold 2,746,375 = 4.85% — the best insider alignment of the three names in this slot (ACGL 3.3% across 17, KNSL 5.6% across 14 but 4.0% of that is one person). ⚠️ this is the FY2025 proxy; the FY2026 one was not on the index when this pass ran.
Shelf / ATM / buyback. A 424B5 prospectus supplement was filed 2025-09-17 off an existing registration, so a shelf exists and has been used. Buyback: $22.0m of common stock repurchased in the whole of FY2026 against $490.4m of net income — under 5% of earnings. This is a capital-consuming grower and the Bank's CET1 says why it has to be.
§6 — S8 valuation and sizing (proposals only — ⏳ OPEN (user))
Price $90.63 at 2026-09-21 (market data NYSE:AX). Shares 56.734m → market cap $5.14bn. Equity $3,166.4m → BVPS $55.81, P/B 1.62. FY2026 diluted EPS $8.48 → P/E 10.7x (10.4x on the company's $8.74 adjusted EPS). ADV $35.1m over the 21 sessions to 2026-09-21 — the thinnest name this rotation has passed.
Assumptions, named: earning assets grow ~10% in FY2027 on the $1.9bn of acquired IRAs plus organic loan growth and ~8% in FY2028; NIM 4.70% → 4.55% → 4.50% as the 16%-yielding purchased-loan book runs off and acquired deposits reprice; non-interest income flat at ~$230m in FY2027 (the $22.0m settlement does not repeat, Verdant lease income annualises); efficiency ratio held near 49.5%; provision normalising toward a 0.25% charge-off rate plus reserve build on growth; tax 28%.
| Fiscal year ended 30 June | FY2026A | FY2027E | FY2028E |
|---|---|---|---|
| Average earning assets $bn | ~26.5 | ~29.0 | ~31.5 |
| Net interest margin | 4.70% | 4.55% | 4.50% |
| Net interest income $m | 1,246.9 | ~1,320 | ~1,418 |
| Non-interest income $m | 233.6 | ~230 | ~245 |
| Non-interest expense $m | 732.7 | ~765 | ~820 |
| Provision $m | ~65 | ~75 | ~85 |
| Net income $m | 490.4 | ~511 | ~560 |
| Diluted shares m | 56.66 | ~56.3 | ~55.8 |
| EPS | $8.48 | ~$9.08 | ~$10.03 |
| Year-end BVPS | $55.81 | ~$64.6 | ~$74.3 |
At $90.63: 10.7x FY2026A, 10.0x FY2027E, 9.0x FY2028E, and 1.22x the modelled FY2028 book.
| Basis | Value | |
|---|---|---|
| Bear | the NY/CA commercial-real-estate refinancing wall bites — charge-offs to 0.75%, provision to $200m+, EPS falls to ~$5.50, and the market pays 8x a bank with a credit problem | ~$44 |
| Base | FY2028E EPS $10.03 at 11x — the multiple it trades on now, on a franchise growing 8-10% with a mid-teens ROE | ~$110 |
| Bull | acquired-deposit funding works as advertised, NIM holds above 4.6%, efficiency stops deteriorating; FY2028 EPS ~$11.50 at 13x | ~$150 |
Base is +21% over ~2 years, which — for the first time in nine names in this rotation — is a positive expected return at the traded price. It is also the widest bear case, because a bank is the one business model in this repository where the downside is not a de-rating but a capital event, and 29% of a $8.9bn CRE book refinances within a year.
Proposed buy price and size: ⏳ OPEN (user). The constraint the owner needs: ADV $35.1m/day. At 10% of daily volume a position takes three days per $10m to build or exit. This is the first name in the rotation where liquidity, not conviction, is likely to set the size.
§7 — S9 monitoring
- Next earnings ~2026-10-29, DERIVED from filing cadence. Fiscal Q1 results 8-Ks went out 2025-10-30 and 2024-10-30, with the 10-Q the same day. Not a scheduled date — market data's calendar needs a paid plan and Equibles did not connect.
- Filing feed: the Tier 0 EDGAR sweep covers CIK 1299709 and has already demonstrated its value on this name — the 8-K that produced this run's finding is invisible in every periodic filing.
- Leading indicator between quarters: weekly H.8 commercial bank deposit data and the advertised rates on Axos's own high-yield savings products [background on the mechanism; no rate was read this pass]. A bank that has just replaced $1.9bn of purchased IRAs and needs to fund a 10% asset growth rate will show it in price before it shows it in the NIM.
- The specific trigger that converts this to a buy: two consecutive quarters in which the efficiency ratio stops rising and the net interest margin holds above 4.55%, with charge-offs still under 0.25% and Bank CET1 stabilised at or above 11%. That combination would say the FY2026 deterioration was the cost of digesting Verdant, Jenius and Capital One rather than the start of a trend — and at 10x earnings the market is not paying for it.
§8 — Evidence that is thin, named
- No earnings-call transcript. Equibles did not connect; no Axos transcript was located on fool.com. This is the third name in the slot without one and the first where none exists at the fallback rung either. Management commentary here comes entirely from the results release and the filings.
- S0, S4, S6 and S7 are not done —
cluster-financials-globalhas not run, and the skill forbids improvising cluster stages per name. So there is no scuttlebutt of any kind for AX in this repository: no Glassdoor, no channel check, no outside commentary, no board claim row. That is the largest single gap on this name and it will stay open until that slot runs. - The ~$1.3bn Capital One shortfall is unexplained, not explained-away. This pass establishes the two numbers and that no filing reconciles them. The benign reading (five months of IRA attrition between signing and closing) is stated as the likely explanation and is not sourced to any document.
- No short interest (Equibles down), no analyst-coverage count (standing gap).
- The premium paid on both deposit acquisitions is undisclosed, so neither deal's economics can be computed.
- DEF 14A is the FY2025 one; compensation structure not extracted, ownership table only. Sixth consecutive name slot to leave the comp read.
- The FY2027 model's deposit-cost assumption is not sourced. Axos does not disclose the rate paid on the acquired IRA book, and the FY2026 average-balance table predates it.
Process support, not investment advice.