All three horizons move to SELL, and the Donatien Pick is Stopped. The cause is a valuation re-anchoring, not a deterioration in the business — which by most measures got better. Business Quality is unchanged at 64 and Timing improved 8 points. The prior report anchored on a "compounder forward-P/E line" of 26× against a forward FY2027 multiple of 20.2×, a ratio of 0.77 and an Attractive band. That 26× line is not either of the framework's financials guardrails (deposit-taking P/E 16× / P/TBV 3.0×, or capital-light financials P/E 30×); the prior calibration contained no sector field at all, so the guardrail check could not run; and it scored a forward multiple where the framework specifies the clean one. Re-anchored on the mandated bank instantiation — justified price-to-tangible-book, because 64.7% of net revenue is net interest income and the company holds $45.5bn of deposits — the name reads 2.49× actual against 1.10× warranted, a ratio of 2.26. Decisively, the 16× deposit-taking guardrail is breached by the clean trailing multiple (44.1×), the reported multiple (38.9×) and the forward FY2027 multiple (22.2×), so the Expensive band needs no estimate at all.
SoFi Technologies is an all-digital American bank. It began in 2011 refinancing student loans and now holds a national bank charter, taking $45.5bn of customer deposits and lending them back out as personal loans, student loans and mortgages, alongside a fee-earning arm that sells brokerage, credit-card, insurance and investing products to the same 15.8 million members. A third segment, the Technology Platform built from the Galileo and Technisys acquisitions, rents SoFi's core banking software to other financial companies. What distinguishes it is the combination of a genuine bank charter — slow, expensive and discretionary to obtain — with no branches at all: deposits fund more than 90% of the balance sheet at a cost the company puts about $712m a year below its old warehouse financing, which is why its net interest margin of 5.98% is roughly double a conventional bank's.
Sector & lifecycle. Financials — a US national bank (SoFi Bank, N.A., chartered January 2022) wrapped in a consumer-fintech distribution model. Lifecycle stage: high-growth, profitable — net revenue +41.0% TTM with a 12.9% GAAP net margin in the June quarter, which is an unusual combination and the reason this name has to be scored on two metric sets at once.
SoFi is a genuine hybrid, so the framework's fintech rule is to blend banking and technology metrics weighted by the actual revenue mix. On the Q2 FY2026 income statement (quarter ended 30 June 2026), GAAP net revenue was $1,218.7m (total revenue $1,571.0m less interest expense $352.3m). Of that, net interest income was $788.2m = 64.7% and non-interest (fee, platform and loan-sale) income was $430.5m = 35.3%. Net interest income is above the 60% line, so the primary lens is banking — return on equity, return on assets, net interest margin, capital, the operating-expense ratio and credit quality — with the technology/growth lens weighted at 35.3%. Every blended figure below uses those weights. By segment, GAAP net revenue was Lending $724.8m, Financial Services $466m (+29% YoY) and the Technology Platform $84.5m.
Data-basis note (lender trap): the provider's "revenue" field of $1,571.0m is gross interest income plus fees. Every growth, margin and multiple figure in this report is struck on net revenue. Free cash flow, EBITDA and gross margin are not computed — for a balance-sheet lender they are structurally meaningless (originations held on balance sheet make operating cash flow persistently negative: −$3.65 per share TTM).
| Sub-signal (lens) | Value | Reference | Score | Rationale |
|---|---|---|---|---|
| Net-revenue trajectory (both) | +42.6% YoY in Q2 FY2026; +41.0% TTM | Financials median low-single-digit; fintech peers 15–25% | 90 | Top-decile on any lens. Guidance was raised to $4.75–4.85bn adjusted net revenue for FY2026 (+33% at the midpoint), so it decelerates but stays exceptional. |
| Profitability (blended 64.7/35.3) | GAAP net margin on net revenue 12.9% (Q2 FY2026) vs 11.4% (Q2 FY2025) | US bank median ~28%; fintech peers ~10% | 54 | Improving, but roughly half the bank-peer margin. The TTM figure of 14.8% is flattered by an abnormally low tax rate (see the earnings-quality note in §4) — the June quarter's 23.4% rate is the clean one. |
| Cash generation (bank-adjusted) | FY2026 guided adjusted EBITDA ~$1.6bn (33–34% margin); GAAP operating cash flow negative | N/A — FCF is not a bank metric | 60 | Scored on the company's own adjusted-EBITDA margin and book-value accretion rather than FCF. Note the short-seller allegation that adjusted EBITDA is overstated (§2). |
| Capital & balance sheet (banking) | CET1 18.7%; deposits $45.5bn (+$5.3bn QoQ) funding 90%+ of liabilities; debt/equity 0.31; financial leverage 5.5× | CET1 >10% is "strong"; bank leverage typically 10–12× | 82 | The strongest thing in this company. CET1 of 18.7% is close to double the strong threshold and gives the balance sheet room to absorb exactly the fair-value shock modelled in §11. The deposit build is the other half — management puts the annualised funding saving at ~$712m versus warehouse financing. |
| Credit quality (banking — the crux) | Total loan book annualised net charge-off rate 1.81% (prior year 2.12%); by product: personal 2.62% (PY 2.83%), student 0.61%, credit card 7.65%. Total on-balance-sheet 90-day delinquency 0.28% (personal loans 0.40%) | Blended consumer-lending book 1.5–2.5%; unsecured personal 4–6%; prime bank 0.3–0.5% | 72 | Better than the headline suggests, and improving on every line. Each figure is paired with its own reference: the total book at 1.81% sits inside a normal blended consumer-lender range and is down 31bp year on year; the personal-loan rate of 2.62% is comfortably below the 4–6% unsecured band. Delinquency is 0.28% across the book. This is the strongest single counter to the 14 August consumer prints. |
| Net interest margin (banking) | 5.98% (+4bp QoQ) | Typical bank 2.5–3.5% | 88 | Roughly double a normal bank — the product of consumer lending funded by retail deposits. |
Banking leg (weight 64.7%) — ROE vs operating-expense ratio: score 25. Return on tangible equity is 6.8% on ending tangible equity ($636.3m TTM net income ÷ $9.41bn), 8.7% on average tangible equity, and the provider reports 7.1% on total equity — all three are below the 10% "healthy" line and, more to the point, below this name's own 11.20% cost of equity (§4). The operating-expense ratio is 83.2% of net revenue (improving from 86.9% a year ago) against a "good" line of under 60%. On the framework's bank table that is the bottom band.
Technology leg (weight 35.3%) — Rule of 40: score 92. On the company's own basis — guided FY2026 adjusted net-revenue growth of 33% plus the guided adjusted-EBITDA margin of 33.5% — the score is 70, above the 60 "exceptional" line. (Mixing TTM growth with a guided margin would flatter it to 74.5; the like-for-like guided pair is used.)
Blend: 0.647 × 25 + 0.353 × 92 = 49. That single number is the honest summary of this company — an elite growth business running bottom-quartile bank returns. It is the main reason Quality sits in the Medium band rather than the High one, which is the decision that matters most here (§1).
Moat score = 60 (average of the five).
SoFi is gaining share where 93% of its revenue sits: members reached 15.8m, up 35% YoY with a record 1.1m added in the quarter; deposits grew $5.3bn sequentially to $45.5bn; originations were $14.8bn, up 69%.
The Technology Platform needs the full picture, not just the headline. Galileo and Technisys did $84.5m, down 23% YoY, on a 14% contribution margin — but the same release discloses that the segment grew 13% sequentially, that the year-on-year decline "includes the impact from a large client which fully transitioned off the platform prior to 31 December 2025" — a lapping effect that annualises out — and that platform accounts rose by 2m in the quarter. That is a segment stabilising after losing one large client, not a business in structural decline. The competitive pressure from Marqeta, Fiserv, Adyen and client insourcing is real, and losing a large client is evidence of it; but "shrinking 23% a year" would misread a lapping comparison as a trend. Threat level: moderate.
| Direct rival | Threat type | Share trajectory | Moat-erosion vector |
|---|---|---|---|
| Chime | Pure-play digital bank, now public | SoFi gaining on breadth (lending, investing, card); Chime holds the low-balance primary-account niche | Deposit-rate competition — both must pay up for primary-account status, which caps pricing power |
| Traditional bank apps (JPMorgan Chase, Capital One 360, Ally) | Incumbent digital substitution | SoFi gaining hard — deposits went from near zero in 2022 to $45.5bn | Incumbents have a structurally cheaper cost of funds and are closing the UX gap; this is the long-run switching-cost risk |
| LendingClub | Direct unsecured personal-loan rival, also bank-chartered | SoFi gaining on scale | Both chase the same prime-plus borrower, so competition runs through APR — a direct pricing-power constraint |
| Affirm (and Upstart) | Adjacent consumer-credit wallet | Stable — different point of sale, same borrower balance sheet | Competes for consumer credit capacity rather than for the account itself |
| Marqeta, Fiserv, Adyen — plus client insourcing | Technology Platform (Galileo/Technisys) | Mixed — one large client fully transitioned off before 31 Dec 2025 (the whole of the −23% YoY), but the segment grew 13% QoQ and added 2m accounts | Client concentration and insourcing risk; the erosion is episodic rather than continuous on the current evidence |
Net effect on the moat: Switching Costs set at 58 — the loss of one large platform client is real evidence of switchability, but sequential account growth of 2m argues against a general exodus — and Cost Advantage at 62 (the funding edge is genuine but has not yet converted into operating leverage). Threat level moderate. The competitive read still propagates to the §11 Bear and the §12 thesis-invalidation rule, but as a client-concentration risk rather than as a structural decline.
Returns (40%): 40. Return on tangible equity of 6.8% sits below the 11.20% cost of equity used in §4 — on today's returns the business is not yet creating economic value. It is, however, rising steeply (group ROE was roughly 1.5% in FY2024), which earns the framework's +10 durability credit rather than its −10 penalty.
Capital allocation (30%): 55. Split verdict. Acquiring Golden Pacific to obtain the bank charter was transformative and is the source of the funding advantage. Galileo ($1.2bn, 2020) and Technisys ($1.1bn in stock, 2022) sit inside a segment that has lost a large client and is only now re-growing sequentially — a disappointing return so far on roughly a fifth of tangible book, though not the write-off a −23% headline alone would imply.
Skin in the game (30%): 48. The CEO bought stock on the open market in March 2026 in response to the short report, which is the right signal. Against that, the diluted share count has gone from 1,065m (Q2 FY2024) to 1,352m (Q2 FY2026), +26.9% in two years; basic shares went 1,058.6m → 1,102.5m → 1,284.3m, i.e. +4.15% then +16.49%. Stock-based compensation is roughly 5.5% of net revenue, well inside the 25% red-flag line — the dilution is capital-raising and convertible conversion, not pay.
Quality = 64. Sector sub-signals (weight 52%) average 74.3; industry benchmark 49 (18%); moat 60 (15%); returns and capital allocation 47 (15%). The independent provider health rating corroborates the band: C+ / 2 of 5 overall, with return-on-equity at 2 and price-to-earnings at 1.
This is one point below the High-quality band, and the reader should know exactly what that means. At 65 the base Decision Matrix row changes from "Medium quality × Expensive valuation → SELL" to "High quality × Expensive valuation → HOLD (great business, wrong price)". So the difference between this report saying Sell and saying Hold is a single point of Quality. What holds it at 64 is not a judgement call but three measured facts: a return on tangible equity of 6.8% against an 11.20% cost of equity, an operating-expense ratio of 83.2%, and a blended industry-benchmark score of 49. What would lift it is operating leverage — the expense ratio converting the funding advantage into returns.
And the conclusion that matters does not turn on it. Whether this reads SELL or HOLD, there is no BUY at any horizon, so the Donatien-Pick Stop (§14) is unaffected either way.
Net interest income is 64.7% of net revenue and the company holds $45.5bn of customer deposits under a national bank charter, so the framework's sector instantiation applies: justified P/TBV = (normalised RoTE − g) ÷ (r − g). Working:
Warranted P/TBV = (0.120 − 0.030) ÷ (0.1120 − 0.030) = 1.10×. Actual = $18.29 ÷ $7.34 tangible book per share = 2.49×. Ratio = 2.26 → EXPENSIVE.
The estimate is not load-bearing — and that is the point. The break-even normalised RoTE needed to escape the Expensive band (ratio below 1.40) is 17.6%, and the Fair line (ratio below 1.20) needs about 20.0% — against a consensus that reaches 13.2% only in FY2029. And an independent corroboration needs no estimate at all: the framework's own bank table anchors P/TBV to ROE directly — ROE 12% → P/TBV 1.0–1.2× — which puts the 1.10× computed here mid-band. Two further checks below reach Expensive without any RoTE input.
| Cross-check | Actual | Warranted | Ratio | Band |
|---|---|---|---|---|
| Justified P/TBV (primary) | 2.49× | 1.10× | 2.26 | Expensive |
| Two-stage residual income, terminal RoTE 14% | 2.49× | 1.39× | 1.79 | Expensive |
| Clean trailing P/E vs the two-stage warranted P/E (banks g 6%) | 44.1× | 14.2× | 3.10 | Expensive |
| Clean trailing P/E vs the two-stage warranted P/E (secular-fintech g 15%) | 44.1× | 20.4× | 2.16 | Expensive |
| Sector guardrail, P/E form — no estimate of any kind | clean trailing 44.1× · reported 38.9× · forward FY2027 22.2× | 16.0× | 2.76 / 2.43 / 1.39 | Expensive on all three |
The residual-income cross-check runs a five-year RoTE path (10.0% → 11.5% → 13.2% → 14% → 14%) with full retention and a terminal residual income growing at 3%, discounted at 11.20%: it values the shares at ~$10.19, a justified 1.39× tangible book. The two-stage warranted P/E is shown on both growth buckets — the Banks bucket caps g(near) at 6% (the consistent choice, since the name is instantiated as a bank) and the secular-fintech bucket at 15% (the generous alternative); the answer is Expensive on either. Most decisive is the last row: the deposit-taking guardrail line of 16× P/E is breached by the clean trailing multiple, by the reported multiple, and even by the forward FY2027 multiple — so the Expensive band holds with no normalised-RoTE estimate, no discount rate and no growth assumption. There is no defensible parameter set on which this name is not Expensive.
Non-operating income: zero. nonOperatingIncomeExcludingInterest and
totalOtherIncomeExpensesNet are $0 in every one of the last ten quarters. There are no
equity-stake markups here, so nonop_pct_of_net_income = 0 and the mega-cap AI
earnings-quality trap does not apply.
The tax line does. The trailing-twelve-month effective tax rate is 13.8% ($101.5m on $737.7m of pre-tax income), because the earlier quarters carried the tail of a deferred-tax valuation allowance release. The June quarter's rate was 23.4%. Normalising the trailing year at 24% gives net income of $560.7m and, on the 1,352.0m diluted shares outstanding in the June quarter, diluted EPS of $0.4147 — a clean trailing P/E of 44.1×. Struck on the same share base, reported TTM net income of $636.3m gives EPS of $0.4706 and a reported multiple of 38.9×; that 38.9× → 44.1× pair is like-for-like. (The data provider's own TTM EPS of $0.49 and P/E of 37.3× use a different share base and are shown for reference only — the pair on that basis would be 37.3× → 42.4×.) Management guides to a 22% FY2026 tax rate, which would give 43.0× rather than 44.1×; 24% is used as the more conservative normalisation and both are disclosed. Clean PEG on the clean trailing basis is 1.37 (44.1 ÷ 32% consensus three-year EPS growth); on the forward FY2027 multiple the same PEG is 0.69 — both disclosed rather than one being chosen.
SoFi carries $46.6bn of loans at fair value. The personal-loan portfolio was marked at 104.7% of par at 30 June 2026, down 71 basis points in the quarter. The disclosed weighted-average discount rates are 4.97% on personal loans and 4.29% on student loans.
Those must be read against the company's own matched benchmarks, not against the 10-year Treasury. The same disclosure gives benchmark rates of 3.99% and 3.90%, corresponding to two-year SOFR for personal loans and four-year SOFR for student loans. The implied credit spreads are therefore 98bp on personal loans and 39bp on student loans — not the ~34bp that comparing a roughly two-year-duration book to a ten-year Treasury would suggest. Benchmarking a short-duration amortising book to the 10-year is a duration mismatch and overstates the concern.
What actually moved, and what it means. The company attributes the discount-rate rise primarily to a 37 basis-point increase in benchmark rates (3.62% → 3.99%), with a modest decline in the weighted-average coupon, a modest increase in the assumed weighted-average annual default rate, and a modest increase in the prepayment assumption (personal-loan CPR 25.55% → 25.77%). So the credit spread was essentially unchanged quarter on quarter, 99bp → 98bp: the mark fell because rates rose, not because the market repriced SoFi's credit. That is a materially less alarming picture than a spread-compression story, and it is the honest one.
The exposure that remains is real and is a rates exposure. A 98bp spread is thin insulation on unsecured consumer paper, the assumed default rate was raised, and the mark has further to fall if benchmark rates keep rising. On an assumed ~2-year weighted average life — consistent with the two-year SOFR benchmark the company itself uses, though the life is not disclosed, so treat the number as illustrative — a further 100 basis points on the discount rate is worth roughly 2% of the mark, about $0.9bn pre-tax, or ~10% of tangible equity. Against a CET1 ratio of 18.7% that is absorbable rather than existential, which is why it sits in the §11 Bear case rather than in a gate.
| Multiple | SoFi | Reference | Read |
|---|---|---|---|
| Price / tangible book | 2.49× | US bank median ~1.6–1.8×; own 5-yr range roughly 0.7×–4× | Expensive vs peers, mid-range vs own history |
| Price / book | 2.12× | — | Expensive vs bank peers |
| Trailing P/E (reported, 1,352.0m diluted) | 38.9× | Bank peers 11–14× | Expensive |
| Trailing P/E (clean, 24% tax, same share base) | 44.1× | Deposit-taking guardrail 16× | Expensive |
| Forward P/E on FY2027 consensus $0.8227 | 22.2× | Deposit-taking guardrail 16× | Still above the guardrail |
| Price / net revenue | 5.4× | Fintech peers 3–6× | Fair |
| Forward PEG (FY2027 multiple ÷ 32% growth) | 0.69 | <1.0 attractive | Attractive |
| FCF yield | N/A | Not a bank metric | Replaced by book-value accretion; no dividend is paid |
Implied-growth read. Reading the anchor backwards: at $18.29 the market is paying 2.49× tangible book, which the Gordon relationship only justifies at a sustainable RoTE of about 17.6–20.0% — 17.6% merely to stop being Expensive, 20.0% to read Fair. Management has spoken of a long-run ambition in that region; consensus gets to 13.2% by FY2029. The gap between those two numbers is the investment question. The framework is explicit that feeding the ambition into the model would "warrant" any price, so it does not.
The Loan Platform Business — and it is quantified. The release puts LPB revenue at $143.3m in the quarter, with $3.1bn of the $14.8bn of originations made for third parties. That is fee income with no balance-sheet capital behind it, and at roughly 12% of net revenue it is no longer a rounding error — yet it is valued inside a lending multiple.
A funding-cost call option. Every input that is hurting today reverses if the Fed resumes cutting: deposit costs fall, the fair-value discount rate falls and the mark on $46.6bn of loans rises. Management explicitly held its adjusted-EBITDA and EPS guidance on an assumption of two rate hikes; that assumption is the bear case being priced, and it is also the option you own for nothing.
Excess capital. CET1 of 18.7% against a >10% "strong" line is several billion of surplus capacity — either growth funding that needs no equity raise, or buyback capacity.
Effect: +3 within the band, and no more. Optionality is a tilt, never a re-rating. The core business is priced at 2.26× its warranted book multiple; the free upside is the reason to keep watching this name, not a reason to call it cheap.
| Street cross-check | Value | Read |
|---|---|---|
| Consensus target | $21.11 (median $21.00; high $29.00, low $16.00) | +15.4% to consensus — the "price 10–20% below consensus" band, score 72 on that sub-lens |
| Coverage / dispersion | 27 analysts; high is 1.81× low | Deep coverage; wide dispersion, so −10% confidence |
| Grades distribution | 0 Strong Buy / 9 Buy / 14 Hold / 4 Sell — consensus Hold, 33.3% bullish | Score 32 — the Street is not behind this name |
| Provider health rating | C+, overall 2/5 (DCF 1, ROE 2, ROA 3, D/E 3, P/E 1, P/B 2) | Independent corroboration of both the Quality band and the Expensive read |
Valuation = 32. The anchor sets the Expensive band (0–39) and the relative lenses can only order within it: the +15.4% gap to consensus and a sub-1.0 forward PEG argue for the upper part of the band, the Hold-consensus grades and the C+ health rating for the lower. Thirty-two, including the +3 optionality tilt, is where that lands. The consequence is mechanical: Expensive fires the Valuation Ceiling gate (§2), and Medium Quality against an Expensive valuation is a SELL row on the base Decision Matrix at every horizon.
Primary driver: the US rate regime. For most lenders that means funding cost and net interest margin. For SoFi it means three things at once, which is why the driver carries more weight here than at a normal bank: (1) the cost of $45.5bn of deposits, (2) the net interest margin on the loan book, and (3) — distinctively — the discount rate applied to $46.6bn of loans carried at fair value, which flows through reported earnings and tangible book directly. This quarter demonstrated the mechanism precisely: the personal-loan mark fell 71bp, and the company attributes it primarily to a 37bp rise in benchmark rates rather than to any repricing of its credit. Secondary driver: US consumer credit health, which sets the charge-off rate and the default assumption inside those same marks.
| Horizon | Reading | Score | Evidence and date |
|---|---|---|---|
| Historical (25%) | Neutral | 55 | Rates fell then stalled through 2025–26; SoFi's total-book charge-off rate improved from 2.12% to 1.81% year on year, and deposits replaced warehouse funding. Constructive, but the tailwind stopped. |
| Current (50%) | Headwind | 40 | Fed cuts are priced out and the live debate is hike-versus-hold (2-year 4.22% against a 3.63% funds rate; 10-year 4.70%, curve bear-steepened +48bp — macro report 2026-08-12, which also carries a live "Fed hikes in September" tail at ~44% market-implied with three hawkish dissents in July). Management held FY2026 adjusted-EBITDA and EPS guidance on an assumption of two hikes even while raising the revenue line, and the shares fell about 10% on 29 July on precisely that. On the consumer side, July retail sales printed −0.6% MoM (US Census advance estimate, 2026-08-14, against roughly +0.1% to +0.3% expected; still +5.0% YoY) and Michigan consumer sentiment fell to 51.0 from 55.2 versus a 54.5 consensus. |
| Forward (25%) | Headwind, improving tail | 42 | The FOMC minutes on 19 August are the next dated read, and the macro report names them as the trigger that would confirm the September-hike tail. If the Middle East energy shock that the Michigan survey blames for the cost-of-living fear subsides, the cut path returns and every input above reverses — that is the §11 Bull. |
Driver = 55 × 0.25 + 40 × 0.50 + 42 × 0.25 = 44 — Headwind. Amplification eligibility: a headwind must be ≤ 35 to intensify a SELL into a STRONG SELL, and this is 44. No amplification fires; the matrix output stands unchanged. The driver does not alter the three fundamental pillar scores.
Two corroborating prints on the same day with a named mechanism (energy → cost of living → spending) is more than noise, and the Michigan decline was broad across income, age and political lines. It is still one month. So it moves the driver's Current score and the §11 Bear probability; it does not flip a regime call. It is also worth saying plainly that SoFi's own credit data points the other way: the total loan book charged off at 1.81% against 2.12% a year earlier, personal loans at 2.62% against 2.83%, and 90-day delinquencies across the book are 0.28%. The consumer prints are a forward risk to marks and originations, not a description of what the loan book has done so far.
Thesis-invalidation floor. The case breaks if the personal-loan mark goes through par (100%) while the assumed default rate keeps rising — that combination would mean the fair-value cushion is gone and the reported earnings of the last two years were borrowing from the future. The mark is 104.7% and fell 71bp last quarter, though on this evidence for rate reasons rather than credit reasons. That dial is moving; the distinction between why it moves is what to watch.
Financials read Neutral / Underperform / Neutral across short, medium and long in the 2026-08-12 macro report. Pressure anchors on the medium horizon, so it is a Headwind. A headwind makes a long position Contrarian by definition, and the conviction score measures how justified fading it would be. Here it is low — 35 — and the reason is specific: the framework scores contrarian conviction off valuation washout, oversold exhaustion and a turning driver, and only one of the three is present. Valuation is Expensive rather than washed out (2.26× warranted); the driver is still deteriorating, not turning; the tape alone is genuinely oversold-and-recovering, sitting at the 19th percentile of its 52-week range with swing lows stepping up since 29 July. One of three does not justify fighting the macro. The dominant regime — an energy-shock stagflation with Fed cuts priced out, a live hike-versus-hold debate and a September-hike tail the macro report puts at ~44% — is close to the worst configuration for this particular balance sheet, because it raises the deposit cost, compresses the spread AND lifts the discount rate applied to $46.6bn of loans held at fair value at the same time. The 14 August consumer prints post-date that macro report, so its consumer read is used here with that qualification. Effect on the signal: pressure is Headwind, which is one of the two conditions for amplifying a SELL to STRONG SELL — but the driver at 44 is above the ≤35 threshold, so no amplification fires and the base signal stands unchanged.
Source: GICS sector map (XLF) — SOFI is not carried in the macro report's watchlist_forecast, so the sector signal is used and that substitution is stated here rather than left implicit · Macro report 2026-08-12
| Component | Weight | Score | Basis |
|---|---|---|---|
| Multi-timeframe trend | 30% | 58 | Monthly uptrend and daily recovering against a weekly downtrend — see §9. Mixed / transitioning. |
| Risk-reward / position risk | 20% | 40 | The stop that makes sense sits at $16.20, below the $16.47 swing low — $2.09 away, which is 2.66 ATR on a daily ATR of $0.786. Anything beyond 2.5 ATR is an unfavourable entry. Price is also within 2.2% of the first daily resistance at $18.70, which costs a further 15 points; the nearest swing-low support is 8.6% below. |
| Macro overlay | 15% | 32 | Fed hawkish with a live September-hike tail at ~44% implied; XLF Neutral / Underperform / Neutral in the 2026-08-12 macro report. The one favourable line is the curve — the 10-year at 4.70% above the 2-year at 4.22%, bear-steepened +48bp, which helps a spread lender. |
| Sentiment | 18% | 50 | Six analyst actions in thirty days and every one a maintain — Truist (Hold, 12 Aug), Mizuho (Outperform, 31 Jul), Morgan Stanley (Underweight, 30 Jul), Needham (Buy, 30 Jul), Wells Fargo (Equal Weight, 30 Jul), Truist (Hold, 24 Jul). No upgrades, no downgrades. Estimate revisions are marginally positive (FY2027 consensus EPS $0.8227 versus $0.808 at the last report). News tone is split between "record quarter" and "why did it fall 10%". |
| Catalyst density | 17% | 62 | No company catalyst inside thirty days — Q2 was reported 29 July and the next print is late October. But the FOMC minutes land on 19 August and this is a rate-driven name, so it is not a clean calendar either. |
Timing = 58×0.30 + 40×0.20 + 32×0.15 + 50×0.18 + 62×0.17 = 50 — the Neutral band (40–54). Fintech carries the medium macro sensitivity weighting (macro 15%, sentiment 18%, catalyst 17%).
Over three months SoFi is +14.2% (from $16.02 on 2026-05-14) against SPY +5.0% and XLF +13.8% — comfortably ahead of the index, level with its own sector. Over one month it is −1.4% (from $18.55 on 2026-07-14) against SPY +4.5% and XLF +3.4%, so it has lagged both since the earnings reaction. The shares sit at the 19th percentile of their 52-week range ($14.88–$32.73) — beaten down, which is information about position rather than about value.
What the tape actually says: the low of $14.88 was made on 29 July on the earnings reaction, and the identified swing lows since have stepped up in sequence — $15.65, then $16.47, then $16.72, then $17.08. Price closed at $18.29 on Friday 14 August, above the 20-day ($17.435) and 50-day ($17.489) averages but well under the 200-day at $20.82. RSI is 55.3 and the MACD histogram has been positive since early August. That is a genuine turn, and it is why one entry path is open in §12 even though the signal is a Sell.
| Date | Event | Impact | Forecast | Previous | Relevant? | Why |
|---|---|---|---|---|---|---|
| 2026-08-17 | NY Empire State Manufacturing (Aug) | Medium | 10.2 | 15.6 | Low | General risk tone only |
| 2026-08-17 | NAHB Housing Market Index (Aug) | Medium | 33 | 34 | Medium | Reads across to the home-loan segment |
| 2026-08-18 | Housing Starts / Building Permits (Jul) | High | 1.35m / 1.37m | 1.427m / 1.374m | Medium | Mortgage origination demand |
| 2026-08-19 | FOMC Minutes | High | — | — | YES — highest | The rate path is this name's primary driver: deposit cost, net interest margin AND the discount rate on $46.6bn of fair-value loans. The macro report names these minutes as the trigger for its live September-hike tail. Sets the next update date. |
| 2026-08-20 | Initial Jobless Claims (Aug/15) | Medium | 210k | 209k | Medium | Employment is the leading input to consumer charge-offs |
| Date | Event | Actual | Forecast | Surprise | Impact |
|---|---|---|---|---|---|
| 2026-08-14 | Michigan Consumer Sentiment (Aug) | 51.0 | 54.5 | −6.4% | Negative — expectations 55.4 → 50.6, current conditions 54.8 → 51.8; the survey attributes it to cost-of-living fear from the Middle East conflict |
| 2026-08-14 | US retail sales MoM (Jul) | −0.6% | +0.1% to +0.3% | Miss | Negative — to $763.6bn, still +5.0% YoY; nonstore/online −2.2%, motor vehicles −1.8%. Census advance estimate; not carried in the provider calendar, sourced centrally this run |
| 2026-08-14 | Michigan 1-Year Inflation Expectations (Aug) | 4.3 | — | vs 4.2 prior | Negative — sticky expectations argue against cuts |
| 2026-08-14 | Atlanta Fed GDPNow (Q3) | 4.3 | 5.8 | −25.9% | Negative — growth tracking cut sharply |
| 2026-08-14 | Business Inventories MoM (Jun) | 0.0% | 0.1% | Miss | Neutral |
One high-impact event inside three trading days, and it is the one that matters most: the FOMC minutes on Wednesday 19 August. Financials is a high-macro-sensitivity sector under the framework, so that release counts as an impactful scheduling event and sets this report's next update to 20 August. It is also the named trigger for the macro report's live September-hike tail (~44% market-implied, three hawkish dissents in July) — and a hike feeds this company through the fair-value discount rate as well as through funding costs. Looking back, the 14 August pair — retail sales −0.6% and Michigan at 51.0 — is a genuine negative for a consumer lender, and it post-dates the 2026-08-12 macro report, so the macro file's consumer read is used here with that qualification. Two prints in the same direction with a named mechanism is more than noise; it is still one month, so it moves the driver score and the §11 Bear probability rather than a regime call. Set against it, SoFi's own June-quarter credit lines improved on every measure.
| Timeframe | Trend | Direction | RSI | MACD | Key S/R | Breakout | Vol |
|---|---|---|---|---|---|---|---|
| Monthly | Uptrend ↑ | Bullish | 52.6 | +1.39, hist −0.98 (fading) | S: $6.41 R: $32.73 | Resistance breakout | 0.38× |
| Weekly | Downtrend ↓ | Bearish | 49.7 | −0.74, hist +0.36 (improving) | S: $14.88 R: $25.11 | Support breakdown | 0.61× |
| Daily | Recovering → | Neutral-Bullish | 55.3 | +0.24, hist +0.11 | S: $16.72 R: $18.70 | Resistance breakout | 0.48× |
| Hourly | Strong uptrend ↑ | Bullish | 49.7 | +0.03, hist −0.03 | S: $17.88 R: $18.48 | Resistance breakout | 0.02× |
| 15-min | Weakening → | Neutral | 44.8 | −0.02, hist +0.004 | S: $18.28 R: $18.45 | Support breakdown | 0.06× |
| Confluence: Mixed / Transitioning · MTF Score 58 | |||||||
Weighted at the framework's 30/25/25/12/8, the five timeframes give 58 — Mixed / Transitioning. The provider's own one-word summary reads 'bullish'; the weighted computation does not agree with it, and the disagreement is the weekly chart, which is still in a downtrend with the 50-week average at $21.72 far above spot. The honest picture is a daily recovery inside an intact intermediate downtrend: price is above the 20-day ($17.435) and 50-day ($17.489) averages but 12.2% below the 200-day at $20.82, and the 200-day is the line that would confirm a real trend change. The textbook pattern is 'higher-timeframe bearish, lower-timeframe rally into resistance' — the first daily resistance is $18.70, only 2.2% above Friday's close. The identified swing lows since the 29 July capitulation step up cleanly: $14.88, $15.65, $16.47, $16.72, $17.08. Two caveats on the intraday rows: the run date is a Sunday, so the hourly and 15-minute volume ratios of 0.02× and 0.06× are weekend artefacts and the 'strong uptrend' hourly reading should not be leaned on. And the monthly support levels of $4.24–$6.41 are pre-2023 relics — the levels that matter are the daily and weekly ones. All moving-average comparisons here use RAW closes; SoFi pays no dividend, so adjusted and raw series are identical and the dividend-adjustment trap does not apply.
Daily closes 11 Feb 2026 to 14 Aug 2026 with the 50-day simple moving average (raw, unadjusted — SoFi pays no dividend, so adjusted and raw series are identical). The $14.88 low on 29 July is the 52-week low, made on the earnings reaction; the identified swing lows since step up — $15.65, $16.47, $16.72, $17.08. The 200-day average sits at $20.82, above the top of this window.
Trigger: the energy shock that the Michigan survey blames for the cost-of-living fear subsides, the Fed resumes cutting rather than hiking, and the whole chain reverses at once — deposit costs fall, the fair-value benchmark rate falls back toward the 3.62% of the prior quarter, the 104.7% personal-loan mark recovers, and management raises the adjusted-EBITDA guidance it held back on a two-hike assumption. Members keep compounding at 35% and the Loan Platform Business scales past its $143.3m quarterly run-rate. On a tangible book of roughly $8.90 by mid-2027 this is about 2.7× tangible book — a multiple the market has paid for this name before, and still above the warranted 1.10×. Falsification: the FOMC minutes on 19 August read hawkish and the September-hike tail the macro report puts at ~44% converts.
Trigger: the operating story delivers exactly as guided — FY2026 adjusted net revenue of $4.75–4.85bn, adjusted net income near $825m — while the rate path stays hostile and no re-rating arrives. Tangible book grows toward ~$8.90, so a roughly flat share price is itself a de-rating from 2.49× to about 2.0× tangible book. That is the point: on this reading the shareholder's return comes from the multiple compressing into the book value rather than from the price. Note the base is a market path, not the intrinsic anchor — the anchor values the equity at $8.07 (Gordon) to $10.19 (residual income), and the gap between $18.00 and those figures is the risk this report is flagging, not a forecast that it closes within twelve months.
Trigger — three linked risks. (1) Rates into the marks. This is a rate exposure, not currently a credit one: the mark fell 71bp last quarter because benchmark rates rose 37bp (3.62% → 3.99%), while the credit spread was essentially unchanged at 99bp → 98bp. If the September-hike tail the macro report carries as live at ~44% converts, the benchmark rises again. On an assumed ~2-year weighted life a further 100bp is worth roughly $0.9bn pre-tax, about 10% of tangible equity — painful, but absorbable against a CET1 ratio of 18.7%. The sharper risk is the spread widening rather than the benchmark rising: 98bp is thin insulation on unsecured consumer paper, and the assumed default rate has already been nudged up once. (2) The consumer. July retail sales −0.6% MoM and Michigan sentiment at 51.0 arrive after the quarter closed; if they show up in charge-offs, the default assumption rises again and this becomes a credit story rather than a rates one. The macro report's private-credit crack tail is building rather than armed — HYG sits fractionally below both its 50- and 200-day averages on raw price — so it is carried here as a mechanism rather than as an inherited armed tail. (3) Client concentration in the platform. One large client fully transitioned off Galileo before 31 December 2025 and accounts for the whole of the −23% year-on-year decline; a second departure would turn an episodic loss into a pattern and would take the Technology Platform's stabilisation with it. Layered on top: the Muddy Waters allegations remain unresolved. At $13.00 the shares are ~1.5× a $8.60 tangible book — still above the warranted 1.10×.
Probability-weighted 12-month target: $17.70 (0.20 × $24.00 + 0.50 × $18.00 + 0.30 × $13.00), about −3.2% from Friday's $18.29 close. Probabilities sum to 100 with the base case the most probable. The weighting is deliberately base-heavy: the operating business is delivering and is likely to keep delivering, so the central case is not a collapse — it is a share price that goes nowhere while book value grows into the multiple. The 30% bear weight, above a conventional 20%, reflects that the bear's leading mechanism is a rate path the macro report itself scores as a live ~44% probability, feeding a fair-value book whose mark has already fallen 71bp in a quarter.
Forecast: Fundamental group — UNLIKELY inside twelve months. It needs price under $10.19 or a driver score back above 50. Price would have to fall 44%; on the driver, the nearer path is the Fed, and the first dated read is the FOMC minutes on 19 August. Confidence: Low. Technical group — ALREADY MET on the higher-lows branch, but fragile: it fails on two consecutive closes back below the $17.49 50-day average, which on the current $0.786 daily ATR is a little over one average day's range away. Confidence: Moderate, decaying. Catalyst group — CATALYST-DEPENDENT, next window late October with the Q3 print. It needs a post-print move above +5% on more than twice average volume; the last two prints produced −13% (29 April) and about −10% (29 July), so the base rate is poor. Confidence: Low. What would change the report itself: the valuation band, not the tape — and there are two distinct thresholds, which are worth keeping apart because they need very different things. Escaping Expensive means the warranted ratio below 1.40: at this price that is tangible book per share above roughly $11.90, or a sustainable return on tangible equity of about 17.6%. Reaching Fair means the ratio below 1.20: tangible book per share above roughly $13.89, or a sustainable return on tangible equity near 20.0%. Consensus reaches 13.2% only in FY2029, and book value compounds about 15–20% a year, so even the nearer of the two is a multi-year wait on the arithmetic route alone. The other route is a single point of Business Quality — at 65 rather than 64 the matrix row becomes HOLD rather than SELL — but that is not the knife-edge it looks like: reaching 65 needs a full point on the composite, which would take the benchmark's banking leg from 25 to roughly 34 on a return on tangible equity of 6.8% against a '>10% healthy' line and an 83.2% operating-expense ratio against a '<60% good' line, or the moat from 60 to 67, or returns and capital allocation from 47 to 54. None of those is close on the current evidence.
Forecast: Stop-loss — unlikely in the next four to six weeks at the current trajectory: $16.20 is 11.4% below spot and beneath both the 20-day and 50-day averages, so it needs a fresh shock. The live risk trigger is the FOMC minutes on 19 August. Thesis invalidation — one of six conditions live (the driver has turned), so the group is not triggered; it needs a second. The two worth watching are the fair-value credit spread widening beyond 98bp and the net interest margin breaking 5.5%, both quarterly reads, so the next genuine check is the late-October print. Profit-target — unlikely: the $21.00 median target is 14.8% above spot and RSI is 55.3.
What you are risking. The stop sits at $16.20 — $2.09, or 11.4%, below Friday's close. The bear path is $13.00, a 28.9% drawdown. You would be buying at 2.49× tangible book against a warranted 1.10× — 2.26 times the multiple the returns support — and doing it with the shares 2.2% below their first resistance at $18.70 after an 11.6% run in two weeks, which is close to the worst place on the chart to start. There is also an unresolved short-seller allegation about the accounting of the very asset that backs the book value, and a live September-hike tail that feeds straight into the fair-value marks.
What you are gaining. Immediate participation in a business growing net revenue 42% a year with charge-offs falling (total book 1.81%, from 2.12%), a $45.5bn deposit franchise behind a bank charter, a CET1 ratio of 18.7% that can absorb the shock the bear case models, and a free call option on the Fed: management held its profit guidance assuming two hikes, so a single cut re-rates the marks, the funding cost and the guidance together. The base case is $18.00 — flat — and the bull is $24.00, +31.2%. There is no dividend, so nothing accrues while you wait.
The read. The probability-weighted outcome is $17.70, about −3.2% from here. Reward to risk against the stop is 2.7:1 on the bull case only, and the weighted expectation is negative. Acting now is not worth it on these numbers — this is an assessment, not a verdict on the company.
What you are giving up. The base target is $18.00, so on the central case you forgo essentially nothing; the real cost is the 20% bull branch worth +31.2%, which is a bet on the Fed rather than on the company. You also give up the Loan Platform Business at a $143.3m quarterly run-rate and a deposit franchise compounding at $5bn a quarter. You would be selling at 2.49× tangible book — nearly 80% above the most generous intrinsic estimate of $10.19 — so you would not be selling below fair value on any reading here.
What you are protecting. The 30%-weighted path to $13.00, and a 44% decline from the 52-week high has already shown this name's drawdown behaviour with a beta of 2.204.
The read. Be precise about the mechanics: no exit trigger is live — the stop is 11.4% away, one of six thesis-invalidation conditions holds, and the profit target is 14.8% above. The Sell in this report is a valuation-band verdict on new capital, not a fired stop.
Position sizing is not computed — no portfolio allocation or role was supplied, and inventing one would be worse than omitting it. The volatility context that would feed such a calculation: beta 2.204, so a 5% position behaves like an 11% position in risk terms against the index; daily ATR $0.786, which is 4.3% of price — a normal day here moves more than a normal week in a large-cap bank; and a drawdown of 54.5% from the $32.73 52-week high to the $14.88 low, all inside the last twelve months. The §12 conviction ladder reads Half-Size (1 of 3 entry paths open), but the signal is SELL, so the ladder describes how much you would buy if you were buying — it is not a sizing recommendation here.
{
"ticker": "SOFI",
"date": "2026-08-16",
"version": "v6",
"company": "SoFi Technologies, Inc.",
"brand": "",
"currency": "USD",
"exchange": "NASDAQ",
"exchange_ticker": "NASDAQ:SOFI",
"isin": "US83406F1021",
"api_ticker": "SOFI",
"finder_ticker": "SOFI",
"finder_exchange": "NASDAQ",
"sector": "Financials \u2014 Banks (US digital bank / consumer fintech)",
"gics_sector": "Financials",
"analysis_status": "stopped",
"analysis_status_note": "Donatien Pick auto-Stopped under SKILL Step -2 B3: analysis_status moved from 'donatien-pick' to 'stopped' because this refresh returns no BUY in any horizon (short, medium and long are all SELL). The name is VISIBLE on the watchlist and auto-reactivates to 'donatien-pick' if a later report produces a BUY in any horizon. The second consumed field that moved is short_entry_confirmed, false -> true: the Technical entry group is now MET on the pullback branch, where the prior report was capped technical_pending. entry_conviction and entry_groups_met read the same on both reports (Half-Size, 1 of 3), but the open path moved from Fundamental to Technical. NOTE: the Stop does not depend on the SELL-versus-HOLD distinction \u2014 Quality 64 sits one point below the High band, and at 65 the matrix row would read HOLD rather than SELL, but there would be no BUY at any horizon on either row.",
"price_at_rating": 18.29,
"price_at_rating_note": "Close of Friday 2026-08-14, the latest print \u2014 2026-08-16 is a Sunday and markets were closed.",
"signal_short": "SELL",
"signal_medium": "SELL",
"signal_long": "SELL",
"primary_signal": "SELL",
"short_entry_confirmed": true,
"quality_score": 64,
"valuation_score": 32,
"timing_score": 50,
"driver_score": 44,
"driver_label": "Headwind",
"composite_short": 48,
"composite_medium": 49,
"composite_long": 52,
"lifecycle_stage": "high_growth_profitable",
"moat_score": 60,
"overall_confidence": 55,
"quality_confidence": 55,
"valuation_confidence": 60,
"timing_confidence": 58,
"economic_alignment_stance": "Contrarian",
"economic_alignment_conviction": 35,
"economic_alignment_pressure": "Headwind",
"economic_alignment_source": "sector-map (XLF N/U/N) \u2014 SOFI is NOT carried in the macro report's watchlist_forecast, so the GICS sector map is used",
"macro_report_date": "2026-08-12",
"revenue_mix_nii_pct": 64.7,
"revenue_mix_fee_pct": 35.3,
"segment_net_revenue_usd_m": {
"lending": 724.8,
"financial_services": 466.0,
"technology_platform": 84.5
},
"val_multiple_basis": "justified P/TBV = (normalised RoTE - g)/(r - g) \u2014 the SKILL's deposit-taking-bank instantiation (SKILL:490), required because net interest income is 64.7% of Q2 FY2026 net revenue ($788,195k / $1,218,666k = 64.68%, above the SKILL:256 60% line) and the company holds $45.5bn of deposits under a national bank charter. Actual 2.49x P/TBV on $7.34 tangible book per share vs warranted 1.10x (normalised RoTE 12.0%, r 11.20%, g 3.0%). Independently corroborated by SKILL:188, which anchors P/TBV to ROE directly (ROE 12% -> P/TBV 1.0-1.2x), placing the computed 1.10x mid-band. Cross-checks, all Expensive: two-stage residual income justified P/TBV 1.39x (terminal RoTE 14%) = 1.79; clean trailing P/E 44.1x vs a two-stage warranted P/E of 14.2x on the Banks g_near cap of 6% = 3.10, or 20.4x on the secular-fintech cap of 15% = 2.16. DECISIVELY, the deposit-taking guardrail P/E line of 16x is breached with NO estimate of any kind by the clean trailing multiple (44.1x), the reported multiple (38.9x) AND the forward FY2027 multiple (22.2x), so the Expensive band does not depend on the normalised-RoTE judgement at all. Break-even normalised RoTE to escape Expensive (ratio <1.40) is 17.6%, and to reach Fair (ratio <1.20) is 20.0%, against an FY2029 consensus of 13.2%. Note warranted_ratio 2.26 is computed on the rounded inputs 2.49/1.10; at full precision (2.4918/1.0976) it is 2.27 - immaterial, same band, well inside the linter's tolerance. The P/TBV form of the guardrail (3.0x) is NOT breached at 2.49x; the P/E form is.",
"actual_multiple": 2.49,
"warranted_multiple": 1.1,
"warranted_ratio": 2.26,
"val_band": "expensive",
"discount_rate_r": 0.112,
"risk_free_10y": 0.047,
"g_near": 0.06,
"g_term": 0.03,
"g_near_note": "Banks bucket cap (6%), consistent with instantiating the name as a bank. The secular-fintech cap of 15% is disclosed in \u00a74 as the generous alternative; both give an Expensive P/E cross-check.",
"risk_addon_pp": 2.0,
"risk_addon_basis": "beta 2.204 > 1.6 (framework's top add-on tier)",
"rates_source": "MacroDriver-state-20260812.json \u2014 10Y 4.70%, 2Y 4.22%, effective funds 3.63%, curve bear-steepened +48bp. All rate inputs taken from the macro state per SKILL Pillar-2 step 1.",
"normalised_rote_pct": 12.0,
"breakeven_rote_pct": 17.6,
"fair_band_rote_pct": 20.0,
"breakeven_note": "breakeven_rote_pct is the normalised RoTE that would take the warranted ratio below 1.40 (escaping Expensive); fair_band_rote_pct is the 1.20 Fair line. Equivalent tangible-book-per-share thresholds at this price: $11.90 and $13.89 respectively.",
"current_rote_pct": 6.8,
"tangible_book_per_share": 7.34,
"cet1_ratio_pct": 18.7,
"eps_trailing": 0.4706,
"trailing_pe": 38.86,
"eps_basis_note": "TTM net income $636.264m / 1,352.0m diluted shares (Q2 FY2026) = $0.4706; price/EPS = 38.86x. The clean pair on the SAME share base is $0.4147 / 44.1x. The provider's own TTM EPS of $0.49 (P/E 37.3x) uses a different share base and is not reproducible from the filings; on that basis the clean pair would be 42.4x.",
"nonop_pct_of_net_income": 0.0,
"nonop_note": "No non-operating or mark-to-market income: nonOperatingIncomeExcludingInterest and totalOtherIncomeExpensesNet are $0 in all ten quarters pulled. The earnings-quality adjustment here is the TAX line \u2014 TTM effective rate 13.8% (tail of a deferred-tax valuation-allowance release) versus 23.4% in the June quarter. Normalising the trailing year at 24% gives EPS $0.4147 and a clean trailing P/E of 44.1x against a reported 38.9x on the same share base. Management guides to a 22% FY2026 rate, which would give 43.0x; 24% is used as the more conservative normalisation and both are disclosed.",
"clean_pe": 44.1,
"clean_peg": 1.37,
"clean_peg_note": "On the clean trailing multiple (44.1 / 32% consensus 3-yr EPS CAGR). On the forward FY2027 multiple the same PEG is 0.69; both disclosed in \u00a74.",
"forward_pe_fy2027": 22.2,
"consensus_eps_2026": 0.59629,
"consensus_eps_2027": 0.8227,
"consensus_eps_2028": 1.05419,
"consensus_eps_2029": 1.37981,
"net_revenue_ttm_usd_m": 4305.8,
"net_revenue_growth_ttm_pct": 41.0,
"net_interest_margin_pct": 5.98,
"deposits_usd_bn": 45.5,
"nco_rate_total_book_pct": 1.81,
"nco_rate_total_book_py_pct": 2.12,
"nco_rate_personal_loan_pct": 2.62,
"nco_rate_personal_loan_py_pct": 2.83,
"nco_rate_student_pct": 0.61,
"nco_rate_credit_card_pct": 7.65,
"delinquency_90d_total_pct": 0.28,
"delinquency_90d_personal_pct": 0.4,
"credit_note": "Release Table 8. The 2.62% figure is the PERSONAL-loan rate, not the total book; the total book charged off at 1.81% (prior year 2.12%). Likewise 90-day delinquency is 0.40% on personal loans but 0.28% across the book. Each metric is paired with its own reference band in \u00a73 (blended consumer book 1.5-2.5%; unsecured personal 4-6%).",
"personal_loan_fair_value_mark_pct": 104.7,
"fair_value_loans_usd_bn": 46.6,
"fv_discount_rate_personal_pct": 4.97,
"fv_discount_rate_student_pct": 4.29,
"fv_benchmark_rate_personal_pct": 3.99,
"fv_benchmark_rate_student_pct": 3.9,
"fv_benchmark_basis": "two-year SOFR for personal loans, four-year SOFR for student loans (company disclosure)",
"fv_credit_spread_personal_bp": 98,
"fv_credit_spread_student_bp": 39,
"fv_spread_qoq_bp": "99 -> 98 (essentially unchanged)",
"fv_note": "The 71bp fall in the personal-loan mark was driven PRIMARILY by a 37bp increase in benchmark rates (3.62% -> 3.99%), plus a modest lower weighted-average coupon and a modest HIGHER assumed weighted-average annual default rate, with prepayment also modestly HIGHER (personal-loan CPR 25.55% -> 25.77%). The credit spread was essentially unchanged QoQ, so this is a RATES exposure rather than a credit repricing. Benchmarking the 4.97% discount rate to the 4.70% 10-year Treasury would be a duration mismatch against a ~2-year book.",
"operating_expense_ratio_pct": 83.2,
"competitive_share_trajectory": "gaining",
"competitive_threat_level": "moderate",
"competitive_note": "Gaining where 93% of revenue sits (members +35% to 15.8m, deposits +$5.3bn QoQ to $45.5bn, originations +69%). The Technology Platform's -23% YoY is NOT a structural decline: the same release discloses +13% QoQ growth, +2m platform accounts QoQ, and that the YoY fall 'includes the impact from a large client which fully transitioned off the platform prior to December 31, 2025' - a lapping effect now annualising out. Threat level moderate (revised from elevated on this evidence). Named rivals: Chime, JPMorgan Chase / Capital One 360 / Ally, LendingClub, Affirm, and Marqeta / Fiserv / Adyen plus client insourcing in the platform. Risk reframed as CLIENT CONCENTRATION rather than structural erosion.",
"relative_strength_vs_spy": 9.2,
"relative_strength_vs_sector": 0.4,
"relative_strength_note": "3-month, percentage points: SOFI +14.2% vs SPY +5.0% (+9.2pp) and XLF +13.8% (+0.4pp). Over 1 month SOFI is -1.4% vs SPY +4.5% and XLF +3.4%, i.e. behind both.",
"range_position_52w_pct": 19.1,
"swing_lows": [
14.88,
15.65,
16.47,
16.72,
17.08
],
"timing_detail": {
"mtf_confluence": 58,
"mtf_confluence_label": "Mixed / Transitioning",
"risk_reward_score": 40,
"macro_score": 32,
"sentiment_score": 50,
"catalyst_clustering_score": 62,
"dynamic_macro_weight": 0.15,
"rsi_daily": 55.3,
"atr_daily": 0.786,
"sma20_daily": 17.435,
"sma50_daily": 17.489,
"sma200_daily": 20.82,
"stop_distance_usd": 2.09,
"stop_distance_atr": 2.66,
"note": "Daily recovering above the 20- and 50-day averages, 12.2% below the 200-day, inside an intact weekly downtrend. Hourly/15-min volume ratios are weekend artefacts (Sunday run date)."
},
"quality_detail": {
"industry_benchmark_name": "Blended by revenue mix \u2014 bank ROE+efficiency (64.7%) and Rule of 40 (35.3%)",
"industry_benchmark_value": 70,
"industry_benchmark_score": 49,
"moat_score": 60,
"roic_percentile_vs_peers": 40,
"capital_allocation": 55,
"management_skin_in_game": 48,
"rote_ending_pct": 6.8,
"rote_average_pct": 8.7,
"roe_provider_pct": 7.1,
"rule_of_40": 70,
"operating_expense_ratio_pct": 83.2,
"cet1_ratio_pct": 18.7,
"band_note": "64 is ONE POINT below the High band. At 65 the base matrix row changes from 'Medium x Expensive -> SELL' to 'High x Expensive -> HOLD'. Held at 64 by three measured facts: RoTE 6.8% vs an 11.20% cost of equity, an 83.2% operating-expense ratio, and a blended benchmark score of 49. The Donatien-Pick Stop is unaffected either way, since neither row produces a BUY.",
"note": "Rule of 40 = 70 on the company's own guided basis (33% guided FY2026 adjusted net-revenue growth + 33.5% guided adjusted-EBITDA margin); mixing TTM growth with a guided margin would flatter it to 74.5. Bank leg scored 25 (RoTE 6.8% below the 10% line and below the 11.20% cost of equity; operating-expense ratio 83.2% against a sub-60% 'good' line)."
},
"fmp_rating": "C+",
"fmp_overall_score": 2,
"analyst_consensus_target": 21.11,
"analyst_target_high": 29.0,
"analyst_target_low": 16.0,
"analyst_target_median": 21.0,
"analyst_target_upside_pct": 15.4,
"analyst_grades_consensus": "Hold",
"analyst_grades_breakdown": "0 Strong Buy / 9 Buy / 14 Hold / 4 Sell",
"analyst_bullish_pct": 33.3,
"analyst_coverage_count": 27,
"recent_upgrades_30d": 0,
"recent_downgrades_30d": 0,
"loan_platform_business_revenue_usd_m": 143.3,
"fair_value_est": 10.19,
"fair_value_note": "Two-stage residual-income intrinsic value, the most generous of the intrinsic reads. The single-stage Gordon anchor alone gives $8.07 (1.10x x $7.34).",
"stop_loss": 16.2,
"target_price": 18.0,
"scenario_bull_target": 24.0,
"scenario_base_target": 18.0,
"scenario_bear_target": 13.0,
"scenario_probabilities": {
"bull": 20,
"base": 50,
"bear": 30
},
"scenario_weighted_target": 17.7,
"entry_groups_met": 1,
"entry_conviction": "Half-Size",
"exit_groups_live": 0,
"exit_action": "Hold",
"hard_gate_state": "caution",
"gates_triggered": [
"Valuation Ceiling (Gate 3) \u2014 actual P/TBV 2.49x vs warranted 1.10x = 2.26x, the Expensive band"
],
"gates_caution": [
"Accounting overhang \u2014 Muddy Waters short report 17 Mar 2026 (doubled down 21 Jul) alleging understated charge-offs (a true PERSONAL-loan rate of 6.1% against the 2.89% personal rate then reported), $259m of unwarranted 2025 fair-value gains, $312m of unrecorded DEBT and $251m of unrecorded liabilities; company denies and threatened legal action; a plaintiff-firm (Block & Leviton) securities investigation is open. NOT a triggered gate: no auditor qualification and no regulatory finding. Verified against the primary source rather than asserted \u2014 the Q2 FY2026 10-Q (filed 2026-08-06) discloses only Smith v. SoFi Securities (deposit-sweep interest class action, N.D. Cal., June 2026) in Legal Proceedings, with NO regulatory investigation, subpoena or enforcement action relating to loan accounting, fair-value marks, charge-offs or off-balance-sheet structures. A plaintiff-firm inquiry is not a regulatory ruling under Gate 5.",
"Share-count growth \u2014 diluted shares +26.9% over two years (1,065m -> 1,352m); on BASIC shares +4.15% then +16.49%, so the '>5% per year for 2+ years' test fails in year one. SBC is ~5.5% of net revenue, far inside the 25% arm. Recorded as a caution. The basis question is not load-bearing: Gate 4 caps at HOLD and gates never raise, so it cannot affect a SELL signal either way.",
"Fair-value marks \u2014 $46.6bn of loans carried at fair value; personal-loan mark 104.7%, down 71bp QoQ. The fall was primarily a 37bp benchmark-rate rise (3.62% -> 3.99%), with the credit spread essentially unchanged at 99bp -> 98bp; the assumed default rate was nudged up. A rates exposure rather than a credit repricing, and absorbable against CET1 of 18.7%."
],
"do_not_buy_triggers": [],
"dnb_adjudication": "Trigger 2(a) deep-expensive: the ratio of 2.26 is above the 2.0x line, but the arm carries an explicit carve-out for exceptional, proven, durable growth and this name has +42.6% YoY net revenue, +69% originations, +35% members and SIX consecutive quarters of rising operating income. The carve-out applies. On the guardrail form of 2(a), 1.5x the 3.0x P/TBV line is 4.5x and 2.49x is far below it. Trigger 2(b) expensive-plus-live-catalyst: no Structural Business Model Threat (the short-seller allegations are unproven, denied, and carry no regulatory finding). All FOUR tail risks in MacroDriver-state-20260812 were adjudicated, not just the first two: (1) 'S&P 500 concentration / AI earnings-quality unwind' \u2014 armed but SoFi is not in that cohort (non-operating income is $0 across ten quarters), dismissed; (2) 'Private-credit crack' \u2014 status 'building', not armed, so it does not qualify under the inheritance rule; carried as a \u00a711 Bear mechanism instead; (3) 'Hormuz closure escalation' \u2014 status LIVE, but it reaches SoFi only indirectly through the energy -> cost-of-living -> spending chain, which is already scored inside the Driver pillar and the \u00a711 Bear consumer leg; it is not a cohort-level de-rating catalyst for this name, dismissed; (4) 'Fed hikes in September' \u2014 status LIVE at ~44% market-implied, and it genuinely applies, but it is a macro-PATH scenario already fully carried in the Driver score (44, Headwind), the Economic-Alignment pressure (Headwind), the \u00a711 Bear leg 1 and the next-update scheduling; treating a base-case rate decision as a DNB de-rating catalyst would fire DNB on every rate-sensitive name in the universe and would double-count the same risk. Dismissed. DISCLOSURE: if the framework's intent were read the other way and the September-hike tail counted as the 2(b) catalyst, the output would be DO NOT BUY rather than SELL \u2014 in either case there is no BUY at any horizon, so analysis_status: stopped is unaffected.",
"signal_derivation": "Base Decision Matrix: Quality 64 = Medium (40-64), Valuation 32 = Expensive (<40) -> SELL at every horizon. Amplification: a SELL intensifies to STRONG SELL only when the driver is <=35 AND economic pressure is Headwind; pressure IS Headwind but the driver is 44, so no amplification. Short technical-confirmation cap and the quality-starter override are not reachable from a SELL base. Hard gates cap and never raise, so the Valuation Ceiling leaves SELL standing. No Do-Not-Buy trigger fires. ROBUSTNESS: Quality 64 is one point below the High band; at 65 the row would read HOLD, not SELL. No BUY is produced on either row, so the Donatien-Pick Stop holds under both readings.",
"next_update_date": "2026-08-20",
"next_update_basis": "FOMC Minutes 2026-08-19 (high-impact, interest_rates) + 1 trading day \u2014 Financials is a high-macro-sensitivity sector, the Fed path is this name's primary driver, and the macro report names these minutes as the trigger for its live September-hike tail",
"deltas_vs_prior": {
"prior_date": "2026-07-31",
"prior_price": 16.39,
"price_change_pct": 11.6,
"quality": "64 -> 64 (0)",
"valuation": "66 -> 32 (-34)",
"timing": "42 -> 50 (+8)",
"driver": "52 -> 44 (-8)",
"signal_short": "HOLD -> SELL",
"signal_medium": "BUY -> SELL",
"signal_long": "BUY -> SELL",
"val_band": "attractive -> expensive",
"analysis_status": "donatien-pick -> stopped",
"note": "The entire signal change is the valuation re-anchoring; Business Quality is UNCHANGED at 64 and Timing IMPROVED 8 points on a genuinely better tape. The prior report anchored on a 'compounder forward-P/E line' of 26x compared against a forward FY2027 multiple of 20.2x, giving a ratio of 0.77 and an Attractive band. That 26x line is not one of the two financials guardrails in the framework (deposit-taking P/E 16x / P/TBV 3.0x, or capital-light financials P/E 30x); the prior calibration contained no sector field at all, so the guardrail arm could not run; and it scored a FORWARD multiple where the framework specifies the clean number. Re-anchoring on the mandated bank instantiation \u2014 justified P/TBV \u2014 moves the band from Attractive to Expensive on its own, and Medium Quality against an Expensive valuation is a SELL row at every horizon. That same re-anchoring is what newly fires Gate 3, the Valuation Ceiling: the 31 July calibration's gates_triggered array was empty, and this run records actual P/TBV 2.49x against a warranted 1.10x = a 2.26 ratio, inside the Expensive band. The gate caps at HOLD maximum and a gate can only cap, never raise, so it confirms rather than causes the SELL the Decision Matrix already returns; the 16x deposit-taking P/E guardrail is independently breached on the clean trailing (44.1x), reported (38.9x) and forward FY2027 (22.2x) multiples, so the Expensive band needs no estimate. No other gate is triggered: the run records three cautions (share-count dilution under Gate 4, the accounting overhang under Gate 5, and the fair-value marks), where the 31 July calibration carried no gates_caution field at all, and do_not_buy_triggers stays empty."
}
}
The three fields that decide this report: val_band: expensive (warranted ratio 2.26), quality_score: 64 — the Medium band, one point below High — and analysis_status: stopped, the automatic consequence of no BUY in any horizon on a Donatien Pick. The name stays visible on the watchlist and reactivates to a Pick if any future report produces a BUY at any horizon. dnb_adjudication records all four macro tail risks being adjudicated (including the two marked live) and signal_derivation records the matrix chain, including the note that the Stop holds whether Quality reads 64 or 65.
1. Data provenance. Every mandatory pull is listed above with its result. One failure (get_earnings_calendar, empty) with the fallback stated. Data-basis traps checked explicitly: lender net revenue — the provider's $1,571.0m "revenue" is gross, and every figure here uses net revenue of $1,218.7m (§3); non-operating earnings decomposition (step 7b) — run, returns zero across ten quarters, and the real adjustment was found on the tax line instead (§4); degenerate analyst consensus — tested, the panel is dispersed ($16–$29), so no fallback was needed; dividends — none have ever been paid and no payout claim is made; moving averages on raw prices — SoFi pays no dividend, so the adjustment trap cannot bite; Polygon date labels — not used for any dated claim, all dates and swing lows come from get_technical_indicators or get_yahoo_prices.
2. Live-status re-verification. The Muddy Waters report and its 21 July escalation, the company's denial, and the Block & Leviton investigation were all searched fresh this run and are dated above. The material distinction — a plaintiffs'-firm inquiry is not a regulatory investigation — is the reason Gate 5 does not fire. Because that ruling rests on a negative, and an unverified negative is exactly the failure mode this audit layer exists to catch, it was tested against the primary source: the Q2 FY2026 Form 10-Q filed 6 August 2026 was read directly, and its Legal Proceedings note discloses only Smith v. SoFi Securities, with no regulatory investigation, subpoena or enforcement action disclosed against the loan accounting. No claim of the form "X is resolved" appears in this report.
3. Internal consistency. Every directional claim was reconciled against the pulled numbers, and three tensions are stated rather than smoothed. (a) The driver is a headwind on consumer credit while SoFi's own credit metrics improved on every line — resolved in §5: the consumer prints are a forward risk to marks and originations, not a description of the June quarter's loan book. (b) The provider's multi-timeframe summary says "bullish" while the weighted computation returns 58 — the disagreement and its cause (the weekly downtrend) are stated in §9. (c) The signal is SELL while the §12 exit action is Hold — the exit rules govern an existing position and none is mechanically triggered; the Sell is a valuation-band verdict on new capital, and §12 says so plainly.
4. Completeness. All fifteen sections are present and carry specific figures. §13 gives volatility context but no position percentage, because no allocation was supplied.
5. Carried-forward diff. Nothing was inherited. All five pillars, the valuation anchor, the gates, the competitive read, the scenarios and the chart series were rebuilt from this run's pulls. The valuation basis was deliberately re-derived rather than carried, and the reason is set out in the changes box and in §4.
6. Signal caps. The base matrix gives SELL at every horizon (Medium Quality × Expensive
valuation). The short technical-confirmation cap and the half-size quality-starter override both act only on a
BUY or HOLD base and are unreachable from here; accordingly no short_hold_reason and no
short_cap_reason is emitted, since both fields are scoped to a short HOLD or the sanctioned
starter. Hard gates cap and never raise, so the Valuation Ceiling leaves SELL standing. Do-Not-Buy was
adjudicated explicitly against all four macro tail risks, including the two carrying status "live" —
see dnb_adjudication in §14.
7. Calibration schema. Written through
scripts/normalize_calibration.py --write and verified with --check;
scripts/lint_report.py was run against this report and the calibration and exits clean.
Known judgement calls, flagged for the independent auditor. (a) Normalised RoTE 12.0% is an estimate built from consensus net income over rolled-forward tangible equity; the framework does not specify a convention for a bank whose returns are still scaling, and at today's 6.8% the ratio would be 5.4 rather than 2.26. It is not load-bearing: the 16× deposit-taking guardrail is breached by the clean, reported and forward multiples, so the Expensive band holds with no estimate at all. (b) Quality 64 is one point below the band boundary that changes the signal — at 65 the row reads HOLD rather than SELL. The three facts holding it at 64 are stated in §3, and the Donatien-Pick Stop is unaffected either way, since neither row produces a BUY. (c) The conviction ladder reads Half-Size on a SELL report — that is the mechanical count of met entry groups, and §12 says plainly that it measures how much you would buy if you were buying, not whether you should. (d) Gate 4 recorded as a caution on the basic-share test; the framework is silent on basic versus diluted, and the question is not load-bearing, because Gate 4 caps at HOLD and gates never raise.