Earnings quality / accounting optics (new) Negative — nearly 20 points of index growth is unrealized mark-ups that cannot repeat, and the games have now spread across all three statements

App Economy (Jul 24/29/31): reported Big-Tech profit is increasingly non-operating. Microsoft excluded a $480M OpenAI investment gain from adjusted EPS but kept a $3.2B Anthropic gain included; Amazon's net profit carried a $53.4B Anthropic markup (after May's $65B Series H at a $965B valuation, up from $380B in February); Tesla's $1.1B net profit included a $1.0B unrealized SpaceX mark. Separately, Microsoft extends data-centre useful life 15→25 years from FY27 and shifts more leases finance→operating — which sit outside reported capex — so "the spending remains, but the accounting optics improve." Watch the deltas between adjusted EPS, GAAP, and cash. App Economy's Aug 1 sweep of 36 prints shows the one-off distortion running both ways: Ford's $0.66 becomes $0.37 without a $1.3B IEEPA tariff benefit; Apple's EPS carried $0.11 of tariff refunds; Robinhood's $0.62 becomes $0.48 ex a fund-deconsolidation gain; Starbucks' 430 bp margin expansion leaned on refunds covering three quarters of duties, with North America's own margin moving only 13.3%→13.6% (management conceding the year-to-date view is "the more normalized read"). In reverse, Meta's operating income would have risen 9% rather than fallen 8% ex $2.4B of legal charges and $1.2B of severance. 2026-JUL-31 (Horizon Kinetics Q2 2026 commentary): methodological point applied to the index's core — HK deliberately refuses the standard adjustment that adds non-cash stock compensation back into free cash flow: "if a company didn't utilize stock options, you'd think employees would demand cash instead, so it really is an operating expense — and dilutes shareholder returns even as stock." On a run-rate basis across the Mag 7 that is $108 billion. It also flags the convention of dropping negative-earnings members out of a group average as a source of misleading valuation optics (two Mag 7 members have negative FCF; one trades at 700x). 2026-AUG-03 (David Hay / Haymaker portfolio update): the "earnings are exceedingly strong" rebuttal is dismissed as "the AI frenzy and… extremely flattering accounting, bordering on circular financing" — "several of the biggest tech companies are reporting huge sales to companies in which they have received stock as part of the transaction. They have then been booking enormous gains on those shares, or warrants, as income. It's legal but it is also extremely unsustainable." The measurement fix is the house discipline: "the Price/Sales ratio [is] superior to Price/Earnings… sales are considerably less volatile than earnings [and] it's much harder to inflate revenues with accounting gimmickry" — on which "the S&P today looks significantly more overvalued… than it was even at the height of" the dot-com bubble. App Economy (Aug 4): Grab supplies three clean audits — strip one-time items (a $234M net profit was really $19M of operating profit after a $307M Superbank consolidation gain, less $183M of fair-value losses); split a guidance raise into organic vs acquired (management said outright that Superbank and Stash are part of the $4.10–4.15B raise); and check the literal scope of a regulatory overhang before believing the de-rating — Indonesia's GrabBike commission cut (20%→8%, July 1) binds only two-wheel passenger transport, so "the feared Indonesia-driven guidance cut was unwarranted." 2026-AUG-06 — David Hay (Haymaker Daily): ~75% of Alphabet/GOOG's first-half 2026 profits came from the mark-up of its stakes in, primarily, Anthropic and SpaceX (up from "almost two-thirds" on Jul 26); per David Rosenberg the S&P profit increase this year ex-energy and ex-AI would be "a much more pedestrian 4%." Singh (Aug 9): Alphabet's Q2 EPS of $9.11 was only $2.88 operating; $6.23 was non-operating unrealized mark-ups on Anthropic ($183B → $380B → $965B) and SpaceX ($800B → $1.25T → a $1.77T IPO, since −40%). Stripping Alphabet's and Amazon's gains cuts S&P Q2 growth from 47.4% to 28.0%; because Anthropic must raise again for more gains to book and SpaceX has already fallen, "the reported GAAP EPS for Alphabet and Amazon have the potential to decline in 1Q and 2Q 2027" against an unrevised $405/share (+13%) 2027 consensus — "a dangerous dead weight for 2027 growth statistics." Carlson (Aug 10, "fail of the week"): the stock-comp-is-fake defence — Keith Rabois calling Opendoor GAAP-profitable in 2021 and dismissing share-based comp as "fake," rebutted by Uber's own CFO ("There are good reasons to adjust certain items from GAAP earnings. Stock comp is not one of those items"). Carlson's screen: when a defence requires deleting a real expense, or rests on per-unit ("per home", 20 straight quarters) rather than enterprise profitability, walk. Five years on, OPEN "has literally gone nowhere." 2026-AUG-06 — CNBC Halftime (Brown), the rebuttal: answers the "earnings bubble" charge with revisions — strip out Alphabet's and Amazon's write-ups of their Anthropic/startup stakes and Q2 growth is still 31%, next quarter 28.8%. The key statistic: across the last 80 quarters the average in-quarter estimate change in month one has been −1.9%; right now it is +0.1% — analysts are being forced up by management. And all 11 sectors are growing revenue, +14% in the quarter: "you can fake earnings, you can't fake revenue." Tom Lee's parallel: organic EPS growth 21%, 2027 estimates still rising. 2026-AUG-05 (Pomboy): ~12 of ~20 points of Q1 S&P earnings growth was mark-to-market accounting: AMZN's Anthropic stake marked $8B → $75B+, with ~$16B of a ~$30B quarter from the markup (GOOGL the same) — "paper gains… nothing to do with their ability to pay salaries, fund capital expenditure, and service interest" — and the same lever produces the write-downs if AI valuations cool. 2026-AUG-14 (Haymaker): Second consecutive POW built on a reported-vs-economic gap: VST's Q2 revenue "miss" ($4.02B vs $5.46B consensus) is unrealized mark-to-market losses on forward hedges — "real in the GAAP sense and irrelevant in the economic sense" — from the same hedge program that gives near-complete 2026 earnings certainty. AUG-16 (Jay Singh, SSR): the full-season arithmetic. With 88% of the S&P reported, the 86% beat rate and the 29.2% aggregate surprise would be "the highest earning surprise reported by the index since FactSet began tracking this metric in 2008" (prior record 23.2% in 2Q20) — but Alphabet's $9.11 vs $2.88 included "a gain of 98 billion in other income primarily due to unrealized gains on equity securities like Anthropic" and Amazon's $5.75 vs $1.82 included $53.4B, "combined 150 billion of gains." Ex those two names the surprise falls 29.2% → 11% and blended Q2 EPS growth falls 50.4% → 32%. The honesty check he applies: revenue cannot be marked up, and revenue growth of ~15% is genuine. 2026-AUG-17 — David Hay (Haymaker): the S&P 500's 50% total return since March 2024 came entirely from earnings — its forward P/E actually fell from 21× to 20× — and those earnings lean on "companies such as GOOG reporting massive gains on their holdings in AI entities like Anthropic," which "represented more than 50% of its profits in the first two quarters of this year." He calls it an "immense, though unsustainable, earnings booster rocket" and expects "a looming mean reversion in the S&P 500's profit margins": "on a normalized profit-margin basis, the S&P 500 is far more overpriced than it superficially appears." 2026-AUG-17 — Jay Singh (David Lin Report): S&P Q2 earnings growth reads +50% YTD but "the real number is closer to 30%" once Alphabet's mark-to-market on its Anthropic stake is stripped — vs 23% expected on June 30. Breadth is real (88% reported, 86% EPS beats, 76% revenue beats, all 11 sectors growing revenue, index revenue +15%, energy +42.5%, IT +36%) while the magnitude sits in ~5 names ("like Micron and SanDisk"). Because earnings re-rated up faster than price the forward multiple fell 23× → 20×: "the S&P is actually not even that expensive — if you believe these metrics are sustainable." 2026-AUG-19 (Ted Oakley, Oxbow Advisors, The Real Story): the hyperscaler-specific test — debt taken in to fund the build plus deferred depreciation means reported earnings "24 months from now will [not] be the same as it is today." 2026-AUG-27: Hay: reported profits are partly fiscal — under the Levy-Kalecki profit equation “a $1 trillion jump in the federal deficit inflates corporate profits by a similar amount” (offset by trade deficits and crowding-out). With deficits at 6% of GDP vs a normal 2%, even a 0.5 multiplier yields “around 2% of artificial economic growth.” (David Hay, Haymaker Daily 2026-AUG-27) 2026-AUG-22: App Economy: tariff refunds now materially distort retail earnings — Target's $4.11 adjusted EPS included $1.65 of refunds ($994M ≈ 370bp of Q2 operating margin), Amer Sports' 730bp margin expansion was 390bp refunds — strip the windfall before judging the beat. (App Economy Insights 2026-AUG-22) 2026-AUG-30 (Jay Singh, SSR call): headline GAAP EPS growth of 118.5% against adjusted growth of 31%, because of "how big Anthropic's mark-to-market was for Amazon and Google's P&L" — gains that "need to be taken with a grain of salt." Recomputed: excluding Amazon alone takes consensus consumer-discretionary EPS growth from a record 92% to just 7%; excluding Amazon and Alphabet takes the index earnings-surprise rate from 29% to 10%. The cycle marker to track is non-operating income as a share of core operating income — ~9% at Q1 heading into low double digits, versus over 25% in the 1980s — "a metric that bears watching, as it has historically risen in later-cycle environments… [it] can lead to more volatile earnings in future quarters." And the preference stated outright: "while earnings growth is going up, free cash flow growth is going down — and free cash flow, I think, is more important than earnings," with the index "likely over-earning today relative to historical levels." Wigglesworth (Monetary Matters, 2026-AUG-16) on the "other income" line. "If you just look at the net income of some of the hyperscalers, the public companies now, look at how much is actually classified as other income, which is essentially revaluations of their investments" — MSFT, GOOGL and AMZN marking up stakes in Anthropic, OpenAI and SpaceX. "If you take that away, then some of those earnings look a little bit not bad, but definitely not as good." The mirror-image doubt runs to the private counterparties: "with OpenAI and revenue, how much of that is actually cash, like free cash?" — which is why he is waiting on the filings: "I'm really looking forward to the S-1s for OpenAI and Anthropic. That's going to be a popcorn moment for me." Add the disclosure complaint underneath it: purchase and lease commitments are disclosed but not uniformly — "they use different words in different places, so it's hard to compare one quarter to the next quarter," and some names only say they have "material upcoming payments." (Wigglesworth, Monetary Matters, 2026-AUG-16.) 2026-APR-16: the inverse case — acquisition amortisation understating cash profit: Slegers values Constellation/Topicus on Free Cash Flow Available To Shareholders and Kelly Partners on NPATA ('amortization expenses are required by accounting rules, but they're not a real cash expense… very similar to Warren Buffett's idea of Owner Earnings'), and Brookfield on Distributable Earnings. (Slegers 2026-APR-16) Slegers 2026-AUG-27: the mirror image of accounting optics — marks that understate value, and how to measure the gap. At Fairfax India, unrealised returns are 19%/yr on the mark-to-market (listed) book and 8% on the private book, while realised returns on monetised investments are also 19%: same team, same period, so the private marks are deliberately conservative — and 67% of the portfolio is not marked to market. Proved in cash at the parent: Poseidon carried at $15.50/share, a ~23% stake sold in late May at $28.30 (+85%) for $1.91bn and an $837m pretax gain. The catalyst that converts estimate into price is an IPO of Anchorage, which management says would lift mark-to-market coverage from 31% to 85%. Related on the reported-earnings side (2026-AUG-20): the accruals ratio is now one of three inputs in a mainstream quality index, on AQR evidence that companies with negative cash flow and negative earnings deliver negative returns. Slegers 2026-APR-21: restates headline multiples after stock-based compensation before calling anything cheap, and the adjustment repeatedly reverses the conclusion. Fair Isaac is introduced at "its cheapest valuation level of the past 10 years (Forward PE: 23.8x)" — then SBC equal to 25% of net income takes it to 29.6x, and the entry target is set at 25x after SBC. Same treatment two days later on Fortinet: SBC at 15% of net income moves 29.2x to 33.6x. Worth noting the discipline lapsed by 2026-MAY-07, where both names are modelled on their unadjusted forward multiples. Slegers 2026-MAY-21: the cleanest worked example of a cheapness claim dissolving under one adjustment — Fair Isaac's SBC is 22% of net income, so "FICO currently trades at a FWD PE of 25.2x. But if you take into account Stock-Based Comp, the actual valuation level is 30.7x", turning "cheapest valuation of the past decade" into a pass with a named $901 entry. Slegers 2026-MAY-28: the extreme version at Perimeter Solutions, where a founders' advisory fee (1.5% of shares a year to 2027 plus 18% of market-cap appreciation to 2031) cost $435m against $653m of revenue in 2025 and produced a reported $206.4m loss — earnings have to be rebuilt from the bottom ($181.7m normalised, 18.5x) before any multiple means anything. Note the inconsistency: the same SBC restatement is not applied to Arista's headline 37.7x (2026-MAY-14). 2026-SEP-07 (Excess Returns — Cameron Dawson clip): "financial shenanigans are cyclical — you typically don't see them as prolific in the earlier stages of a boom; you see the most at the end, at the crest, as you're going over the apex." She lists them across all three statements: Nvidia extending receivables for hyperscaler customers who are themselves extending payables (flattering their FCF, worsening Nvidia's — normalize it and hyperscaler free cash flow is "even more negative"); Microsoft reclassifying financial and operating leases out of operating and into investing cash flow so guided capex "looked to be lower than it actually was"; an FT piece that Microsoft "has done a terrible job at disclosing its AI business and hides it in with a bunch of other businesses"; the 25% OpenAI stake carried by the equity method, which will force "a lot more disclosures about these companies over time"; and one-time securities gains skewing GAAP headline earnings hard enough that "as we go into '27 we could be talking about negative year-over-year growth in a quarter like the second quarter" — worst at Google and Amazon, which report GAAP only. Her governing test: "if things were as great as you say that they are, you should not have to be playing the games that you are playing," and "accounting is not something that people care about till the peak of the cycle."