The Goldman Sachs Group GS

Reference table

1. The quarter, and the sentence that frames it

From the Form 10-Q filed 2026-08-03:

And immediately below that table, in the same filing:

> "Our target (through-the-cycle) is to achieve ROE within a range of 14% to 16% and return on average tangible common shareholders' equity (ROTE) within a range of 15% to 17%."

That sentence is the most important thing in this dive. It is not our estimate of what is normal for Goldman Sachs; it is Goldman Sachs's estimate, published in the same document as the 25.5% result. The firm delivered 850 basis points above the top of its own through-the-cycle range and 950 basis points above the midpoint.

The knowledge base independently supplies the framework for what to do with that, from a named independent speaker three years ago:

> 2023-05-01 · neutral · conviction 65 · principle · channel business_breakdowns · speaker Mark Rubinstein · speaker_role independent

> "A no-growth bank is valued by comparing return on equity to cost of equity, yielding a price-to-book heuristic; Goldman targets a 14-16% ROE, its overriding financial target."

The claim identified both the method and the target correctly, three years before the filing we are reading confirmed both. We use exactly that method in Section 4.

Five consecutive beats, and the gap is widening:

Net revenues have risen in four of the last five quarters and the sequence — $14.6B, $15.2B, $13.5B, $17.2B, $20.3B — ends 39% above where it began.

The segment picture, from the vendor block cross-checked against the 10-Q's segment naming:

Two things follow. First, 71.1% of Goldman is a transaction business — advisory, underwriting and market-making — against Morgan Stanley, which sits alongside it in this batch, where the fee-based Wealth and Investment Management halves are roughly half the firm. That is the single clearest structural difference between the two and it is why Goldman's returns are both higher at the top of the cycle and less defensible at the bottom.

Second, Platform Solutions has effectively been eliminated — from $2.427 billion to $151 million in a year. The knowledge base characterises the exit accurately and it is worth quoting because it is the correct reading:

> 2026-01-13 · neutral · conviction 50 · channel compound_and_friends · no named speaker

> "Goldman's consumer/Apple-card push was a failed experiment (>$7B pretax losses since 2020); exiting it is disciplined — they emerge bigger and better."

Geography, from seg_geo, which ties exactly to the segment total: FY2025 Americas $36.548B (62.7%), EMEA $14.155B (24.3%), Asia $7.580B (13.0%), summing to $58.283B — the same figure as the segment table. A clean internal reconciliation.

2. Book value, capital, and what the vendor gets wrong

From the 10-Q's non-GAAP reconciliation at 2026-06-30:

The vendor gets this wrong twice, in the same direction, and the combined error is 17.9%.

bookValuePerShareTTM and tangibleBookValuePerShareTTM both read $409.00 — identical to each other. The vendor (a) divides by total shareholders' equity including $13.028 billion of preferred stock, and (b) sets tangible book equal to book, ignoring $9.271 billion of goodwill and identifiable intangibles.

Price to tangible book is 3.13x, not 2.57x. The vendor makes Goldman Sachs look 17.9% cheaper on the single most important metric for a bank, and every screen in the market that uses this field is wrong the same way.

The share count also corrects. Common shareholders' equity of $109,714M divided by book value per common share of $367.67 gives 298,404,000 shares; tangible common equity of $100,443M divided by $336.61 gives 298,395,000 — the same figure. The vendor's market capitalisation implies 295,006,000. Corrected market capitalisation: 298.4M × $1,052.98 = $314.21B, 1.2% above the vendor's $310.64B.

Now the two facts that matter most and that no ratio in the vendor file expresses.

First: tangible book value per common share grew 0.3% in six months. From $335.49 to $336.61. In the same six months Goldman earned $11,802 million applicable to common shareholders. Common shareholders' equity actually FELL, from $109,819M to $109,714M. The reason is disclosed:

> "During the second quarter of 2026, we returned a total of $5.36 billion of capital to common shareholders, including $4.00 billion of common share repurchases and $1.36 billion of common stock dividends."

Goldman is returning approximately one hundred percent of its earnings, and it is repurchasing stock at above three times tangible book value. That is accretive to earnings per share and dilutive to tangible book value per share — and it means the return to a shareholder here is dividend plus buyback plus whatever the multiple does, with essentially no compounding of the underlying asset. By contrast Morgan Stanley, in the same six months and in the same batch, grew tangible book value per share 6.4%.

Second: capital is tightening.

The 10-Q explains it: the ratios fell "primarily reflecting increases in both Credit and Market RWAs." And the same document states the firm's policy: "we will continue prioritizing deployment of capital for our clients where returns are attractive and distribute any excess capital to shareholders through dividends and share repurchases, while targeting a 50 to 100 basis point buffer above our capital requirement."

A 140 basis point decline in six months on a deliberately thin buffer is a genuine constraint on how long the $16 billion annual buyback pace can run. We flag it as a medium-term watch item rather than an alarm.

Capital returns, on the corrected basis. The Board raised the quarterly dividend from $4.50 to $5.00 per share beginning in the third quarter of 2026 — a $20.00 annualised run-rate, a 1.90% yield, against the vendor's dividendPerShareTTM of $17.00 which is a trailing blend. FY2025 repurchases were $12.360 billion; the June-quarter pace of $4.00 billion annualises to $16 billion, or 5.09% of the corrected market capitalisation. Total shareholder yield is approximately 6.99% — the highest in this batch.

3. Cash flow — ignored entirely, and why

Operating cash flow was negative $45.2 billion in FY2025 while net earnings were $17.2 billion, because changeInWorkingCapital reads −$68.9 billion. For a broker-dealer that figure is the change in the size of the trading and financing book, not a measure of profitability.

Rejected in full, without exception: freeCashFlowYieldTTM (−13.36%), priceToFreeCashFlowRatioTTM (−7.48x), priceToOperatingCashFlowRatioTTM (−8.03x), evToFreeCashFlowTTM (−13.52x), evToOperatingCashFlowTTM (−14.26x), incomeQualityTTM (−1.88), capexToOperatingCashFlowTTM (−0.055), dividendPaidAndCapexCoverageRatioTTM (−4.75), operatingCashFlowRatioTTM (−0.024). Nine fields, all arithmetically correct, all meaningless — and one of them feeds a discountedCashFlowScore of 1 out of 5 into the vendor's composite rating.

4. Valuation — priced in or room?

At $1,052.98 (corrected market cap $314.21B on 298.4M common shares; no enterprise value quoted):

Estimate coverage is good (14-17 analysts on FY2026 and FY2027, 10 on FY2028). The FY2029 row carries FOUR analysts on both lines and is excluded.

The arithmetic worth pausing on: the first half of 2026 delivered $38.51 of diluted EPS — an annualised $77.02 — and the FY2028 consensus, two and a half years away, is $76.939. Consensus is modelling no growth at all from the current run-rate through 2028, and the FY2026 estimate of $69.770 is 9.4% below the rate already achieved. Either the estimates are far too low, or the run-rate is unsustainable. The consensus rating of "Hold" across 55 analysts says the street believes the latter.

Per the data contract we tested est.ebitAvg and est.ebitdaAvg and they FAIL. FY2026 netIncomeAvg of $21.490B exceeds ebitAvg of $14.681B; the same inversion holds in FY2027 ($21.632B vs $15.084B), FY2028 ($21.793B vs $15.362B) and FY2029 ($24.485B vs $15.908B). Net earnings after tax cannot exceed earnings before interest and tax. The FY2023 row is worse still: ebitdaAvg of $27.196B against ebitAvg of $8.673B, a 3.1x gap. All EBIT and EBITDA estimate rows rejected. epsAvg is used and checks well against actuals — 4.8% low on FY2025 and 9.0% low on FY2024, consistent with five consecutive beats.

Peer context. The vendor peer set is the same polluted list supplied for Morgan Stanley — Applied Digital, Bitfarms, Bit Digital, CleanSpark, Hut 8, Marathon Digital and Riot Platforms are bitcoin miners, and three more entries are sub-$10-million microcaps. The genuine comparables are Morgan Stanley ($342.33B), Evercore ($12.37B), Stifel ($12.74B), Lazard ($4.26B), Moelis ($5.09B) and PJT Partners ($4.38B). No peer multiples are supplied, so no peer-multiple comparison is drawn. The relevant fact: Morgan Stanley's market capitalisation exceeds Goldman's by 9.0%, on a business with roughly half its revenue in fee-based wealth and asset management against Goldman's 28.6%.

4a. What today's price assumes (the inversion)

At $1,052.98 — 3.13x tangible book and 14.4x FY2027 consensus — the price embeds:

• A durable return on tangible common equity of approximately 23%. (Our derivation from the multiple.) On the standard framework — price to tangible book equals (return less growth) divided by (cost of equity less growth) — 3.13x at a 10% cost of equity and 4% growth implies 22.8%. The company's own published through-the-cycle target is 15-17%. This is by far the most fragile assumption in the price and the company itself disputes it in the same document as the result.
• Global Banking & Markets does not mean-revert. (Filing-confirmed segment size; our assumption on persistence.) 71.1% of FY2025 net revenues, growing 18.6% that year and driving a 39.5% consolidated increase in the June quarter.
• EPS reaches $73.07 by FY2027 and $76.94 by FY2028. (Consensus-derived; 17 and 10 analysts.) This is the LEAST demanding assumption in the price — the first half of 2026 already annualises above the FY2028 estimate. The nearest check is 2026-10-13, where the $16.26 consensus is 22.5% below the $20.98 just delivered.
• The buyback keeps running at roughly $16 billion a year. (Filing-confirmed pace; our assumption on continuation.) At 5.09% of market capitalisation it is half the expected return. But CET1 fell 140 basis points in six months to 12.9% against a targeted 50-100 basis point buffer, so this is not obviously safe.
• Tangible book value per share does not need to compound. (Filing-confirmed: +0.3% in six months.) A holder is buying a yield and a multiple, not an asset that grows. That works while the multiple holds and stops working the moment it does not.

4b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: EPS growth (+5.3%, from FY2027E $73.070 to FY2028E $76.939) + multiple drift (COMPRESSION, from 14.4x to roughly 13.5x, −6.3%) + shareholder yield (+6.99%) ≈ +6% to +7%.

Almost the entire expected return is the shareholder yield. That is an unusual and honest description of this security: Goldman is paying out roughly 7% a year and neither its earnings nor its tangible book is compounding meaningfully, so what is left is the multiple, and the multiple is at the high end of anything this firm has sustained.

Sensitivity, stated openly, on the filed tangible book of $336.61 per share: at 2.0x (the company's own 15-17% target implies roughly this) the shares are $673; at 2.5x, $842; at 3.0x, $1,010; at 3.13x, spot; at 3.5x, $1,178; at 4.0x, $1,346. On the FY2027 consensus EPS of $73.07: 10x gives $731, 14.4x is spot, 18x gives $1,315.

The bull case at $1,330 requires roughly 25% multiple expansion from an already-elevated level. The bear case at $740 requires no earnings failure at all — only that the return on tangible equity drifts back toward the range the company itself publishes.

4c. Variant perception (where we differ, what would surprise)

• We agree with the street's rating and differ on the target. Consensus is 22 buy, 30 hold, 3 sell — rating "Hold", the only such consensus in this batch — but the average target of $1,202.33 implies +14.2%. Our base of $1,090 is 9.3% below it. The tension in the street's own numbers (a hold rating with a 14% target) is itself informative: the sell-side likes the franchise, models a decent year, and does not want to underwrite the multiple.
• We differ materially from the vendor on the single most important metric. The vendor reports price to tangible book of 2.57x by dividing by total equity including $13.028 billion of preferred and by ignoring $9.271 billion of goodwill and intangibles. The filing supports 3.13x. A 17.9% error in the primary valuation lens for a bank is a material finding and it makes this stock look considerably cheaper on every screen than it is.
• We think the tangible-book-per-share stagnation is under-discussed and it is the cleanest distinction between Goldman and Morgan Stanley. In the same six months: Goldman's tangible book per share +0.3%, Morgan Stanley's +6.4%. Goldman is buying back stock at 3.1x book; Morgan Stanley at 4.1x book is doing far less of it. The consequence is that Goldman's shareholder return is front-loaded as cash while Morgan Stanley's accumulates in the asset, and the two require completely different multiples to produce the same total return.
• We think consensus is too low on FY2026 EPS and it does not change the verdict. $69.770 against an annualised first-half of $77.02, with five consecutive beats averaging 19.1%. We would not be surprised by FY2026 above $78. But at 3.13x tangible book on a self-declared above-target return, an earnings beat that does not raise the sustainable return does not raise the fair value.
• Positive surprise that would force a re-rate: management formally raising the through-the-cycle ROTE target above 17%, which would give the multiple a structural justification it presently lacks; Asset & Wealth Management growing materially toward half the firm; or CET1 rebuilding above 14% while the buyback continues.
• Negative surprise that would break the thesis: ROTE falling below 18% in any quarter; CET1 below 12.5%, which on a 50-100 basis point buffer would force the buyback down; a Q3 miss against the low $16.26 consensus; or a credit event in the Global Banking & Markets book, whose risk-weighted assets grew 8.7% in six months.

Synthos fair values

All three anchors are multiples of the filing's tangible book value per common share of $336.61 at 2026-06-30, cross-checked against the FY2027 consensus EPS of $73.070. We do not project tangible book forward because it grew only 0.3% in the last six months and the buyback is the reason — projecting growth would assume away the central fact.

• Bear ~$740 — 2.20x tangible book. Cross-checks: 10.1x FY2027E; 11.0x the FY2027 consensus LOW of $67.024; a 2.70% dividend yield at the raised $20.00 rate; 19.5% below the 200-day average of $919.19. The scenario: capital-markets activity normalises, Global Banking & Markets reverts toward its FY2024 base, and return on tangible common equity settles inside the firm's own 15-17% target — at which the framework gives 2.0x, or $673, so this bear case is already slightly generous. −29.7%.
• Base ~$1,090 — 3.24x tangible book. Cross-checks: 14.9x FY2027E; 14.2x our own FY2026 estimate of roughly $77; 15.6x FY2026E consensus. Sensitivity: 3.0x gives $1,010, 3.5x gives $1,178, and the street's $1,202.33 implies 3.57x. The scenario: the cycle cools modestly, ROTE settles in the high teens to 20%, the buyback continues at close to the current pace, and the multiple compresses slightly. +3.5%, plus 6.99% of shareholder yield.
• Bull ~$1,330 — 3.95x tangible book. Cross-check: 18.2x FY2027E; 17.3x FY2028E; $5 above the street's high target of $1,325. This bull case REQUIRES roughly 25% multiple expansion from an already elevated level and that is its weak leg. The scenario: the transaction cycle persists, sixth and seventh consecutive beats push FY2026 above $80, management raises the through-the-cycle target, and the market re-rates the franchise as structurally higher-returning. +26.3%.

Base is 3.5% above spot; asymmetry roughly 0.89:1 — MORE downside than upside, and the only sub-parity ratio in this batch. That, on its own, would argue for something below Hold. What holds it at Hold is a 6.99% total shareholder yield — the highest of the twelve names — which pays a holder properly for two years of waiting even if the multiple goes nowhere.

5. Knowledge base — 28 claims, one excellent named cluster, two graded misses

Raw entity hits: 28. Text matches: 4, all discarded. Independent voice-days: approximately 12.

The search covered GS, Goldman Sachs and Goldman across the entity field of all 51,928 distilled claims. Stance on the entity lane: 10 bullish, 10 neutral, 8 bearish — the most genuinely balanced distribution in this batch.

All four text matches are the Morgan Stanley collision class — Goldman as the author of a research note, not as an equity: "Goldman Sachs calls it a $100 billion market" (drones, 2018), "Goldman Sachs projects inflation declining to 3.7%" (2024), "could shift GDP growth from the 1-3% band to 3-5% (per Goldman Sachs)" (2025). All discarded. Several entity-tagged claims have the same character — "Oil could fall into the $50s (Goldman)" (2025-08-29) and "Most analyses (Fed, Goldman) show tariffs are paid mainly by American firms" (2026-02-25) — and are likewise used for nothing.

The valuable cluster is six claims from a single 2023-05-01 episode with a named independent speaker, Mark Rubinstein, and it is the best analytical material on any financial name in this batch:

> "A no-growth bank is valued by comparing return on equity to cost of equity, yielding a price-to-book heuristic; Goldman targets a 14-16% ROE, its overriding financial target."

> "Goldman's sales-and-trading intermediation spread is not fixed; when volatility is higher, bid-offer spreads widen and the opportunity to make money sitting in the middle rises."

> "Over 20 years Goldman's compensation ratio fell from ~50c to ~30c per revenue dollar as balance sheet, brand and technology made the firm — not employees — more important."

> "Goldman now runs with ~$100B of capital versus ~$35B in 2006-07; banks have never been more capitalized relative to assets, so today's larger balance sheet is less risky."

> "Goldman's 2021 record ~$60B revenue was driven by exceptionally high risk appetite; unlikely to see that level again in the short term given the cyclical, GDP-linked business."

Three of these are directly corroborated by the filing we are reading. The 14-16% return-on-equity target appears verbatim in the 10-Q. The ~$100 billion of capital is now $100.443 billion of tangible common shareholders' equity at 2026-06-30 — the figure was right to the dollar three years early. And the volatility-spread mechanism is precisely what produced a 39.5% increase in net revenues. A three-year-old episode that got the target, the capital base and the mechanism right is the strongest qualitative material in this batch, and it is one voice on one day.

The counterweight, and it is a concentration problem. Seven of the twenty-eight entity claims resolve to Jordi Visser — four under channel jordi_visser with no named speaker and three with the speaker field populated. Four of those seven, dated 2025-11-15, 2025-11-16 (×2) and 2025-11-17, are the same assertion restated: that JPMorgan and Goldman are making all-time highs, benefit most from mergers and initial public offerings, and are major beneficiaries of artificial-intelligence-driven margin improvement. Counted once.

Three claims in the lane are now gradeable and two are clear misses.

> 2025-09-07 · BEARISH · conviction 68 · channel brent_johnson · skill 0.7

> "Goldman Sachs is set to fall a lot — easily back into the mid-600s."

> VERDICT: MISS. The shares closed 2026-08-04 at $1,052.98, having reached a 52-week low of $715.95 and a high of $1,152.07. They never approached the mid-600s.

> 2026-03-14 · BEARISH · conviction 75 · channel jordi_visser · speaker jordi_visser

> "When financial stocks are below a downward-pointing 200-day MA, really bad things have historically happened; Goldman is 20% off highs and financials down 30-50%, signaling credit trouble."

> VERDICT: MISS. Goldman is now 14.6% ABOVE its 200-day average of $919.19 and up 48.4% over twelve months. The credit trouble did not arrive.

> 2023-11-01 · BEARISH · conviction 85 · channel arthur_hayes · skill 1.2

> "Big banks are functionally insolvent on unrealized HTM treasury losses; state-backed but unable to lend."

> VERDICT: FALSIFIED, as it was for Morgan Stanley. Goldman's CET1 is 12.9%, tangible common equity $100.4 billion, and risk-weighted assets grew 8.7% in six months — the opposite of "unable to lend."

One neutral claim proved prescient and deserves the credit: the 2023-05-01 observation that Goldman's "2021 record ~$60B revenue was driven by exceptionally high risk appetite; unlikely to see that level again in the short term given the cyclical, GDP-linked business." FY2021 net revenues were $58.982 billion; FY2023 fell to $45.226 billion and FY2024 to $52.164 billion before FY2025 reached $58.283 billion. Three years of not seeing that level again, exactly as described — and it is the same claim, from the same speaker, that this dive is effectively repeating about the June 2026 quarter.

Conclusion. The knowledge base is net neutral on Goldman Sachs — 10 bullish, 10 neutral, 8 bearish — which is unusual and appropriate. Its most valuable content is a single three-year-old episode with a named speaker that supplied both the valuation framework and the cyclicality warning, and both have held up. Its bearish content is timing-based and has been wrong twice. We use the framework in Section 4 and the cyclicality warning in the verdict; the lane does not move the fair value, and the balance of stances is a direct input to the Hold.

6. Data integrity — what we rejected from the vendor file and why

Eleven findings. Structurally similar to Morgan Stanley, with one error that is worse.

1. bookValuePerShareTTM and tangibleBookValuePerShareTTM are BOTH $409.00 — identical, and both wrong. This is the most consequential error in the file. The filing reports $367.67 and $336.61. Two separate mistakes compound: the vendor divides by TOTAL shareholders' equity, including $13.028 billion of preferred stock, and it sets tangible book equal to book, ignoring $7.342 billion of goodwill and $1.929 billion of identifiable intangibles. Price to tangible book is therefore 3.128x, not 2.574x — a 17.9% understatement, larger than the equivalent error on Morgan Stanley. priceToBookRatioTTM and priceToFairValueTTM (both 2.5745) are rejected and rebuilt at 2.864x and 3.128x.

2. The share count is 1.2% low — corrected from the filing. The vendor's market capitalisation divided by price implies 295,006,000 shares. The filing's common shareholders' equity of $109,714M divided by book value per common share of $367.67 gives 298,404,000, cross-checked by tangible common equity of $100,443M divided by $336.61 (298,395,000). Corrected market capitalisation: $314.21B against the vendor's $310.64B.

3. revenue is GROSS in eight of nine quarters and NET in one — an internal inconsistency. Goldman reports net revenues; the vendor's revenue field is generally gross, and net revenue is recovered by subtracting interestExpense. The identity holds exactly for the June 2026 quarter ($38,431M − $18,093M = $20,338M, matching the 10-Q), for December 2025, September 2025 and June 2025. But the March 2026 row reads revenue of $17,227M with costOfRevenue of only $315M and interestExpense of $17,082M — subtracting gives $145M, which is absurd. In that one quarter the vendor appears to have populated revenue with the NET figure ($17,227M matches the earnings-calendar actual exactly). We use net revenues throughout, derived per quarter and cross-checked against earn_cal, and we flag that the vendor's quarterly revenue series is not on a consistent basis. Consequently priceToSalesRatioTTM of 2.63x, computed on a mixed-basis trailing revenue figure, is rejected.

4. enterpriseValueTTM carries a 43.7% unexplained residual and is conceptually inapplicable — WITHHELD. The vendor's $561.244B less market capitalisation of $310.636B implies net debt of $250.608B; the 2025-12-31 balance sheet shows $445.275B. The residual is $194.7B, or 43.7% — the largest in this batch and far above the 15% threshold. Separately, enterprise value is meaningless for a firm whose $609.534 billion of "debt" funds a matched trading and financing book. evToEBITDATTM (17.88x), netDebtToEBITDATTM (7.98x), evToSalesTTM (4.76x) and enterpriseValueMultipleTTM all rejected. No enterprise-value figure appears anywhere in this dive.

5. Nine cash-flow-derived fields are meaningless — rejected in full. Operating cash flow was −$45.154 billion in FY2025 against $17.176 billion of net earnings, driven by a −$68.907 billion working-capital movement that is the change in the trading book. Rejected: freeCashFlowYieldTTM (−13.36%), priceToFreeCashFlowRatioTTM (−7.48x), priceToOperatingCashFlowRatioTTM (−8.03x), evToFreeCashFlowTTM, evToOperatingCashFlowTTM, incomeQualityTTM (−1.88), capexToOperatingCashFlowTTM, dividendPaidAndCapexCoverageRatioTTM (−4.75) and operatingCashFlowRatioTTM.

6. The vendor composite rating of C+ / 2 is half artefact — partially rejected, mechanism shown. Sub-scores: overall 2, discounted cash flow 1, return on equity 5, return on assets 3, debt-to-equity 1, price-to-earnings 2, price-to-book 1. The discountedCashFlowScore of 1 is produced entirely by the negative free cash flow in finding 5 and is rejected. The debtToEquityScore of 1 reflects a 3.57x ratio, which is what a broker-dealer is, and is rejected as a lens. Unlike Morgan Stanley's file, however, returnOnEquityScore correctly reads 5 against a 17.0% trailing return on equity, and priceToBookScore of 1 is directionally right — at 2.86x book and 3.13x tangible book this is genuinely expensive. We reject two of the six sub-scores and accept the price-to-book signal.

7. est.netIncomeAvg EXCEEDS est.ebitAvg in every forward year — EBIT and EBITDA rows REJECTED. FY2026: $21.490B against $14.681B. FY2027: $21.632B against $15.084B. FY2028: $21.793B against $15.362B. FY2029: $24.485B against $15.908B. Net earnings after tax cannot exceed earnings before interest and tax. The FY2023 row compounds it with ebitdaAvg of $27.196B against ebitAvg of $8.673B — a 3.1x gap. epsAvg is used and checks well: FY2025 estimate $48.846 against actual $51.32 (4.8% low) and FY2024 estimate $36.897 against actual $40.54 (9.0% low).

8. seg_prod drops the Investment Banking line from FY2022 onward — flagged; recent years used with the 10-Q's segment naming. The block carries an explicit "Investment Banking" line for FY2021 and earlier and none from FY2022, when Goldman reorganised into three segments. The vendor's "Global Markets" and "Investment Management" labels correspond to the 10-Q's Global Banking & Markets and Asset & Wealth Management. FY2025's three lines sum to $58.283B, which ties exactly to the seg_geo total for the same year — a clean internal reconciliation. We use the block under the 10-Q's vocabulary and note that the FY2021-and-earlier rows are not comparable to the FY2022-onward rows.

9. capitalLeaseObligations is populated inconsistently and totalDebt is not a leverage measure. The vendor's $609.534 billion comprises short-term debt $311.152B, long-term debt $296.212B and leases $2.170B. For a broker-dealer this measures the size of the funding book. The relevant leverage measures are the regulatory ones — CET1 of 12.9% against risk-weighted assets of $790.640 billion — and they are in the 10-Q, not the payload. financialLeverageRatioTTM of 17.33x and debtToEquityRatioTTM of 3.57x are correct and uninformative.

10. lastDividend: 17 is a trailing blend, not the current rate — corrected from the filing. The 10-Q states: "The Board approved an increase in our quarterly common stock dividend from $4.50 to $5.00 per share beginning in the third quarter of 2026." The forward run-rate is $20.00, a 1.90% yield, not the vendor's 1.61%.

11. quote.yearHigh/yearLow disagree with tech.hi52/lo52 — we use tech. The quote block reports $1,153.99 / $705.55; the computed block reports $1,152.07 / $715.95. Discrepancies of 0.2% and 1.5%. We use tech throughout for consistency with the moving averages and drawdown figures.

Not defects, correctly reported and independently confirmed: the gross-to-net revenue identity holds exactly in four of five quarters checked and reconciles to earn_cal; the quarterly diluted EPS series matches the 10-Q for the June quarter ($20.98) and the earnings calendar for the others; returnOnEquityTTM of 16.97% is consistent with the 10-Q's 21.7% half-year return on common equity (the difference being total versus common equity); effectiveTaxRateTTM of 20.24% is consistent with the filing; and the seg_prod and seg_geo blocks tie to each other exactly for FY2025.

Non-equity tripwire — checked and passed. GS is common stock, NYSE-listed, one class of common (multiple series of preferred exist and are separately listed — $13.028 billion of them, which is the source of finding 1). Beta 1.292; a 52-week range of $715.95 to $1,152.07 — a 60.9% spread; a variable, rising dividend just increased 11.1%; volume of 2.52M shares (roughly $2.65B of turnover). This is common equity.

7. Technicals

• Price $1,052.98. −8.60% from the 52-week high of $1,152.07; +47.07% above the 52-week low of $715.95. Position within the annual range: 77th percentile.
• Moving averages: sitting exactly on a 50-day average of $1,051.58 (+0.13%); +14.6% above a 200-day average of $919.19.
• RSI 35.5 — the weakest reading of any name in this batch — and MACD −5.63. A stock at the 77th percentile of its annual range with an RSI in the mid-thirties has been falling persistently into this session. Today's +2.52% is a bounce off the 50-day average, not a breakout.
• Maximum drawdown from peak over the trailing year: −8.60%, i.e. the current price is the low point of a shallow drawdown.
• Relative performance: 3-month +16.57% vs SPY +7.59% and QQQ +7.67%; 6-month +11.27% vs SPY +11.09% and QQQ +15.71%; 12-month +48.40% vs SPY +24.26% and QQQ +30.80%. A 24-point twelve-month outperformance on a 1.292-beta stock.
• Sentiment: 22 buy, 30 hold, 3 sell across 55 analysts — consensus rating "Hold", the only such rating in this batch. Consensus target $1,202.33 (+14.2%), median $1,224.50, high $1,325 (+25.8%), low $995 (−5.5%). A "Hold" rating with a 14% average target is internally tense and worth reading as the sell-side liking the franchise while declining to underwrite the multiple.

Today's move

GS closed 2026-08-04 at $1,052.98, up 2.52% or $25.92 from a $1,027.06 close, on 2.52M shares against a 2.12M average. No company-specific news is in this file for the date; the 10-Q was filed the previous day and carries no information beyond the 2026-07-14 release. Morgan Stanley rose 2.75% in the same session — this was a financials move rather than a company event, and both names bounced off flat 50-day averages with weak internals.

8. Insiders — routine director grants only

Every insider transaction in this file was filed 2026-07-17 for transactions dated 2026-07-15, and each is coded A-Award of 22 restricted stock units at a price of $0 — Mark A. Flaherty (director, 5,270 units held after) and John B. Hess (director, 892 held after) among them.

This is the quarterly non-employee director grant: a fixed dollar value converted at the prevailing price, awarded on the same day to each director. It is a calendar event, not a decision.

What the file contains is therefore: zero open-market purchases, zero open-market sales, and no officer transactions of any kind. Neutral, and reported as neutral. For a firm that has just reported a 45% earnings beat and raised its dividend 11%, the absence of executive activity is unsurprising given standard blackout practice around results, and it carries no signal.

9. Verdict, kill-criteria and flip conditions

Hold.

What is genuinely exceptional: June-quarter net revenues of $20.338 billion (+39.5%), net earnings to common of $6.399 billion (+84.2%), diluted EPS of $20.98 (+92.3%) against a $14.47 consensus — a 45.0% beat, the largest in this batch; return on tangible common equity of 25.5% against 13.6% a year earlier; a pre-tax margin of 42.1%; five consecutive beats averaging 19.1%; $100.443 billion of tangible common shareholders' equity; the disciplined and now essentially complete exit from Platform Solutions (revenue from $2.427 billion to $151 million); and a total shareholder yield of approximately 6.99% — the highest of the twelve names here — comprising a dividend just raised from $4.50 to $5.00 quarterly and a $16 billion annual buyback pace.

What we are declining to pay for: 3.13x the filing's tangible book value per common share of $336.61 — corrected upward from the vendor's 2.57x, which understates it by 17.9% — pricing a durable return on tangible equity of roughly 23% against a company-published through-the-cycle target of 15-17%; a firm that is 71.1% a transaction business; tangible book value per common share that grew 0.3% in six months because approximately all earnings are returned and buybacks are executed above three times book; Common Equity Tier 1 falling 140 basis points to 12.9% on a deliberately thin 50-100 basis point target buffer as risk-weighted assets grew 8.7%; and a base fair value of $1,090 that is 3.5% above spot with a bear of $740 — 0.89:1 asymmetry, the only sub-parity ratio in this batch.

The distinction that matters. Nothing in this dive suggests Goldman Sachs is a poor business or a badly-run one. The problem is entirely that the company has told us, in the same document as the result, what it thinks a normal year looks like — and the price assumes the abnormal year is the normal one. The knowledge base's named independent speaker said the same thing about the last peak three years ago and was right for three years. A holder collects 7% a year while that question resolves; a buyer is paying a peak multiple on a peak return with more downside than upside in our own arithmetic. Hold.

Pre-registered KILL criteria — what would take this to Avoid:

• Return on tangible common equity below 18% in any quarter, which would begin the reversion toward the company's own 15-17% range and take the framework value toward $750.
• CET1 below 12.5%. At a targeted 50-100 basis point buffer, that forces the buyback down — and the buyback is 5.09% of the 6.99% shareholder yield.
• The 2026-10-13 print missing a $16.26 consensus that is already 22.5% below the quarter just delivered.
• Global Banking & Markets net revenues falling below roughly $9 billion in a quarter, which would mark the transaction cycle turning.
• A credit event in a book whose risk-weighted assets grew 8.7% in six months.
• Tangible book value per common share declining outright, which would mean buybacks above three times book are actively destroying the asset rather than merely not growing it.

Pre-registered FLIP conditions — what would take this to Buy — Tactical:

• Price below approximately $880 — 2.61x tangible book and 12.0x FY2027 consensus, near the 200-day average of $919.19 — at which the base case is +24% and the dividend yield 2.27%. This is 16% below today's close.
• Management formally raising the through-the-cycle return targets above 15-17%. This is the single cleanest path to justifying the multiple, and it would be a deliberate, dated, checkable disclosure.
• Two more quarters of return on tangible common equity above 22% with CET1 rebuilding above 13.5%.
• Asset & Wealth Management growing toward 40% of net revenues from 28.6%, which would change the applicable valuation framework toward Morgan Stanley's.
• Consensus FY2027 EPS revising above $80, i.e. the street conceding that the current run-rate is the base rather than the peak.

Where GS fits in the Synthos Framework Portfolio. The financials sleeve, at 0% today with a 2% target on a fill near $880. On batch overlap this is the sharpest concentration in the twelve names: Goldman Sachs, Morgan Stanley, Citigroup and American Express are four expressions of one rate-and-capital-markets cycle, and Goldman is the highest-beta, most transaction-dependent and lowest-fee-annuity of the four. Within the pair that matters most — Goldman versus Morgan Stanley — Morgan Stanley has roughly half its revenue in fee-based wealth and asset management against Goldman's 28.6%, grew tangible book per share 6.4% against Goldman's 0.3% in the same six months, and trades at 4.08x tangible book against Goldman's 3.13x. Goldman is the cheaper multiple on the more cyclical business with the higher cash yield; Morgan Stanley is the dearer multiple on the more durable one with the compounding book. Neither is obviously the better buy today and holding both is one position. Logged as a tracked Synthos call (Hold) as of 2026-08-04 at $1,052.98.

Single biggest risk: return on tangible common equity reverting to the company's own published target. The June quarter delivered 25.5%. The 10-Q states the through-the-cycle target is 15% to 17%. On the standard bank framework — the same one the knowledge base's named speaker described in 2023 — a 16% return at a 10% cost of equity and 4% growth gives 2.0x tangible book, which on the filed $336.61 is approximately $673 a share, 36% below spot. We do not forecast that outcome and we do not need to: the point is that the gap between what the firm earned and what the firm says it expects to earn is 850 basis points, and the entire premium in the price sits inside that gap.

Most fragile assumption in the price: that a 6.99% shareholder yield funded by a buyback at 3.1x tangible book can continue. Goldman returned $5.36 billion in the June quarter while its Common Equity Tier 1 ratio fell 140 basis points to 12.9%, against a stated policy of maintaining only "a 50 to 100 basis point buffer above our capital requirement." Risk-weighted assets grew 8.7% in six months and capital fell 2.5%. If the balance sheet keeps growing at that rate, either the buyback slows or the buffer disappears. And because the buyback is 5.09 of the 6.99 percentage points of shareholder yield — and because the yield is essentially the entire expected return in Section 4b — a buyback cut is not a capital-allocation footnote here; it is most of the investment case.

Provenance & disclosures

• Traceability: 28 tagged knowledge-base claims name Goldman Sachs by entity, plus 4 text matches — ALL FOUR text matches discarded as the Morgan Stanley collision class, in which the firm is cited as the author of a research note rather than as an equity ("Goldman Sachs calls it a $100 billion market", "Goldman Sachs projects inflation declining to 3.7%", "per Goldman Sachs" on GDP growth). Two further entity-tagged claims have the same character ("Oil could fall into the $50s (Goldman)", "Most analyses (Fed, Goldman) show tariffs are paid mainly by American firms") and are used for nothing. Concentration: SEVEN claims resolve to Jordi Visser (4 channel-only, 3 with the speaker field populated), of which FOUR dated 2025-11-15 through 2025-11-17 are the same 'banks at all-time highs / AI margin beneficiary' assertion restated, counted once. The valuable material is SIX claims from a single 2023-05-01 episode with a named independent speaker, Mark Rubinstein, three of which are directly corroborated by the 10-Q read for this dive: the 14-16% return-on-equity target appears verbatim in the filing, the "~$100B of capital" figure matches tangible common shareholders' equity of $100.443 billion at 2026-06-30, and the volatility-spread mechanism is what produced the 39.5% revenue increase. Stance 10 bullish / 10 neutral / 8 bearish — the most balanced lane in this batch. Breadth 12, net conviction neutral. Three claims graded against outcomes: a 2025-09-07 call that Goldman was "set to fall a lot — easily back into the mid-600s" is a MISS ($1,052.98 close, 52-week low $715.95); a 2026-03-14 claim that Goldman below a falling 200-day average signalled credit trouble is a MISS (now 14.6% ABOVE the 200-day average, +48.4% over twelve months); and a 2023-11-01 claim that big banks were "functionally insolvent" is FALSIFIED (CET1 12.9%, risk-weighted assets +8.7%). Against those, a 2023-05-01 neutral claim that Goldman's 2021 record revenue was "unlikely to be seen again in the short term" is a HIT — FY2023 net revenues fell to $45.226B and did not regain the FY2021 level until FY2025. All quotes verbatim from stored claim text.
• Data as-of: income statement, capital ratios, tangible book reconciliation and capital-return disclosure for the quarter and six months ended 2026-06-30, from the Form 10-Q filed 2026-08-03, the day before this dive; fiscal 2025 from the 10-K filed 2026-02-25 · estimates 2026-08-04 · prices 2026-08-04, quote timestamp 1785873603 = 2026-08-04T20:00:03Z ($1,052.98, +2.52%; 50-DMA $1,051.58; 200-DMA $919.19; RSI 35.5; MACD −5.63) · knowledge-base claims 2026-08-04. Goldman's fiscal year is the calendar year. All figures come from the Synthos vendor data file for GS or from the SEC filings in the GS archive; no figure comes from memory, recall or external retrieval.
• Filing archive contents: 10-K filed 2026-02-25 (fiscal 2025, tables preserved); 10-Q filed 2026-05-01 (March 2026 quarter); 10-Q filed 2026-08-03 (June 2026 quarter — the primary source for this dive); 8-Ks filed 2026-07-20, 2026-07-21 and 2026-07-27. Note that the archive contains each document twice under slightly different filenames — an indexing duplication, not a content issue. Tables were available on all substantive documents, and the June 10-Q supplied the financial overview, the through-the-cycle target statement, the capital-ratio table, the tangible book reconciliation and the dividend and buyback disclosure used throughout.
• Where the filings contradicted or corrected the vendor (detailed in Section 6): book value per common share corrected from $409.00 to $367.67 and tangible book value per common share from $409.00 to $336.61 — the vendor reports the two as IDENTICAL, dividing by total shareholders' equity including $13.028 billion of preferred stock and ignoring $9.271 billion of goodwill and identifiable intangibles — taking price to tangible book from 2.574x to 3.128x, a 17.9% understatement of the primary valuation metric for a bank and the single most consequential correction in this dive; the share count corrected from a vendor-implied 295,006,000 to a filing-derived 298,404,000, raising market capitalisation to $314.21B; revenue identified as GROSS in eight of nine quarters and NET in the March 2026 quarter — an internal basis inconsistency that makes the trailing revenue figure and priceToSalesRatioTTM unusable; enterpriseValueTTM WITHHELD on a 43.7% unexplained residual, the largest in this batch, and on the separate ground that enterprise value is inapplicable to a broker-dealer; nine cash-flow-derived fields rejected in full, including a free-cash-flow yield of −13.36% and a price-to-free-cash-flow ratio of −7.48x; two of six vendor rating sub-scores rejected as artefacts while the price-to-book signal is accepted; est.netIncomeAvg found to EXCEED est.ebitAvg in every forward year, rejecting all EBIT and EBITDA estimate rows; seg_prod flagged for dropping the Investment Banking line from FY2022 onward; lastDividend of $17.00 identified as a trailing blend against a $20.00 forward run-rate confirmed by the 10-Q's disclosure of the increase from $4.50 to $5.00 quarterly; and quote.yearHigh/yearLow rejected in favour of tech.hi52/lo52. Where vendor and filing AGREED — worth recording: the June-quarter net revenue identity ($38,431M gross less $18,093M of interest expense equals the filed $20,338M); the diluted EPS of $20.98; the effective tax rate; and the exact tie between the seg_prod and seg_geo totals for FY2025 at $58.283B.
• Basis note: all revenue figures in this dive are NET revenues and all EPS figures are diluted earnings applicable to common shareholders, both taken from the 10-Q where available and from the earnings calendar otherwise. Return on tangible common equity and price to tangible book are the primary valuation lenses; enterprise-value and cash-flow-derived multiples are withheld or rejected throughout, with reasons stated. Goldman reports no separate non-GAAP earnings measure; tangible common shareholders' equity and the ratios derived from it are the firm's own disclosed non-GAAP measures and are used as such.
• Estimate coverage: 14 analysts on FY2026 net revenue and 16 on FY2026 EPS; 15 and 17 on FY2027; 10 and 10 on FY2028. The FY2029 row rests on FOUR analysts on both lines and is excluded from every conclusion. est.ebitAvg and est.ebitdaAvg are rejected in all forward years; only epsAvg and revenueAvg are used, and epsAvg checks 4.8% and 9.0% low against FY2025 and FY2024 actuals — consistent with five consecutive beats averaging 19.1%.
• Peer note: the vendor supplies the same polluted peer set used for Morgan Stanley — seven of the entries are bitcoin mining companies (Applied Digital, Bitfarms, Bit Digital, CleanSpark, Hut 8, Marathon Digital, Riot Platforms) and three more are microcaps under $10M of market capitalisation. The genuine comparables are Morgan Stanley ($342.33B), Stifel ($12.74B), Evercore ($12.37B), Moelis ($5.09B), PJT Partners ($4.38B) and Lazard ($4.26B). No peer multiples are supplied in the file, so no peer-multiple comparison is drawn. Morgan Stanley's market capitalisation exceeds Goldman's by 9.0% on a business with roughly half its revenue in fee-based wealth and asset management against Goldman's 28.6%.
• Fair-value caveat: the $740 / $1,090 / $1,330 anchors are multiples of the filing's tangible book value per common share of $336.61 at 2026-06-30 — 2.20x, 3.24x and 3.95x — cross-checked against the FY2027 consensus EPS of $73.070 (10.1x, 14.9x, 18.2x). We deliberately do NOT project tangible book forward, because it grew only 0.3% in the preceding six months and the buyback at above three times book is the reason; projecting growth would assume away the central fact of the analysis. Sensitivity: 2.0x gives $673 — which is what the company's own 15-17% through-the-cycle target implies on the standard framework — 2.5x gives $842, 3.0x gives $1,010, 3.5x gives $1,178, and the street's $1,202.33 implies 3.57x. Stated arithmetic, not a discounted cash flow, which would be inappropriate for the reasons in Section 3.
• Timing: second-quarter 2026 results were released 2026-07-14, twenty-one days before this dive, beating consensus EPS by 45.0% — the largest beat in this batch — and the Form 10-Q was filed 2026-08-03. The next print is 2026-10-13, 70 days away, with consensus of $16.26 EPS on $17.642 billion of net revenues, both far below the $20.98 and $20.338 billion just delivered.
• Accessibility note: no information in this dive is conveyed by colour. All emphasis is carried by bold text, table structure and explicit labelling.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.