“Have you spoken to a normie recently? They aren’t actually normal anymore. Nobody is.” - Erik Torenberg, via a16z.

Finance TV took years of my attention and never once paid me back. The performative soundbites from guests and anchors talking past each other serve nothing but their ego and a PR machine, contributing to a mass delusion about what’s actually going on, what you need to know, and how markets really work.

Consume it at your own cost.

Equally, nobody needs another podcast bro farming our attention by ‘interviewing’ guests about their world view and life philosophy. The grift has gotten so base that Sequoia’s ex-managing partner Doug Leone just told us he got a root canal with no Novocaine to “see if I was a badass or not.”

“Am I a fake badass that has this envelope nobody could tell, or am I really a badass on the inside…I had to go home and change my shirt because I was a puddle of sweat. But I didn’t move. And so I love fear. It does wonders for me.”

The only thing excelling in the internet’s attention economy is a more elegant nonsense. And if this isn’t peak, I cannot fathom what’s next. So we have to fight back, otherwise we remain prey. And not least because I am desperate for something else to watch, engage with and actually learn from as a markets writer, thinker and investor.

Enter: the independent writers and operators on Substack who live and breathe their niche with no institutional line to toe. To hold any credibility, as measured by MarketStack, you have to possess insights of actual worth. Parrots don’t have a real audience on here. And there is no lasting prestige or job title currency. Nobody generating value for readers is trading on charm alone.

To incubate this kind of speaker network, I first built The Room, a series of private talks, to access unfiltered thinking and perspective without a public stage. I secured an impressive guest list before cancelling the entire thing because my intuition was loudly objecting.

I don’t want a one-off conversation. No matter how long and deep it is.

I want real life.

Pseudo-authenticity

In one-off interactions, the co-hosts are incentivised to flatter each other, so nothing gets prised open. And the format is saturated anyway, everyone and their mother has been copying Joe Rogan for a decade. And badly.

For me, real life and real value is now watching live-streams, on a regular basis, with an open Q&A. I learn by osmosis over time, by listening to someone who is actually building a relationship with their audience, because they are inviting them into a real process instead of performing expertise. And you get to know them, which is nice.

So I’m building MarketStack TV, an (eventually) always-on live TV network for independent writers, analysts and traders. The first anchor starts streaming next week (see below). Instead of guests visiting MarketStack, they will instead co-stream to MarketStack TV alongside their owned audience. MarketStack acts as a co-creator, producer, re-distributor and aggregator layer for contributing anchors whilst serving a single point live destination for the audience.

The launch anchor is of The Oil Report.

Tim is a Dublin-based trader with 20+ years in markets including the prop desks at Geneva Trading and Amplify in London. He knew what it costs to pull a barrel out of the ground long before he traded one and he's been at it long enough to know nobody calls the market.

So don't expect any calls, hype, or trades of the week. Tim streams the way he actually trades - thinking out loud, showing the levels he's watching, and telling you plainly when he's wrong. If you want signals, he's the wrong man. If you want to learn to read the tape, pull up a chair.

Tim will be live on MarketStack TV:

• every Monday at the open for an oil-focused macro outlook

• every Wednesday after the release of the oil market’s NFP equivalent

• and every Friday to wrap up with an outlook for the following week.

Exact times below, starting next week, Monday 7 September.

To receive push notifications ahead of the livestreams, open marketstack.tv on your phone and click to share the URL > add to Home Screen. Notifications also work on desktop. And if you’re in the Substack app and following or you will get the push notification. Subscribers will also receive an email. You can also follow and watch via MarketStack on X or YouTube.

Listen out for a growing list of anchors and shows coming to MarketStack TV. And if you’d like to work with MarketStack (me) on your own livestream slot, get in touch. Even (and maybe especially) if you have never live-streamed before. The only condition is you have a point of view and your work is authentic and not performed.

MarketStack is an independent, anonymous publication summarising publicly available commentary and views from across financial media. Nothing here constitutes financial advice or a recommendation to buy, sell or hold any security. All views are a synthesis of public information. Past performance is not a guide to future results. This publication is not authorised or regulated by the Financial Conduct Authority. The author writes anonymously in a personal capacity.

Two days ago I published a piece on the Bessent buybacks and declared myself fervently neutral on long-dated Treasuries. I said there were two scenarios that could reward duration bulls, and that neither of them was simply the buybacks as announced. The first was the unknown lengths the administration is willing to go to. The second was Michael Green’s reflexive chain, which requires Warsh to cut before any of it starts. I closed by saying that reasoning your way to a directional view against those competing mechanisms was beyond me.

Then Citrini published a third scenario yesterday and it culminates in a tactical 5y30y flattener. It requires no cut, no reflexivity, and no faith in the buybacks doing what the street says they cannot do. It runs entirely through the interaction of bank balance sheets and Treasury supply. More than that, it names a mechanism of the hawk-wrapped architecture I set out in April in Warsh broke the hawk/dove binary, based on the accord Michael Howell described.

What Howell named months earlier

Howell’s “Stealth Accord” describes the 1951 separation of monetary policy from debt management being quietly reversed through operational practice rather than legislation. Nobody signs anything; the architecture just emerges. My April piece claiming that the binary hawk/dove analytical frame was now explicitly redundant was downstream of Howell’s work. Further, I added that under this coordination the hawk/dove vocabulary was actively deceptive, because a chair can dominate the headlines with the hawk narrative while operating as a stealth dove through coordinated Treasury operations. And Citrini’s memo details that coordination and plumbing.

Private QE

Citrini’s argument is a form of what Zoltán Pozsar has labelled private QE: regulatory reform creates balance-sheet capacity inside banks, which can be deployed into Treasury bills and private credit. Treasury then accommodates that demand by shortening issuance, indirectly reducing long-duration supply. In a nutshell:

Shrink the Fed’s balance sheet, let banks create the liquidity instead, shorten Treasury issuance, squeeze crowded long-end shorts—and only worry about the structural inflation problem later.

Precisely how they position for this regime is as follows.

Read more

There is an emerging ‘peak rates’ narrative which conflates improving risk-reward asymmetry in long-duration Treasuries after the selloff and apparent crowded short positioning in TLT (the 20y+ Treasury ETF) with the assumption that Bessent’s buybacks mechanically translate into lower long-end Treasury yields…

Read more

“The exposure is ugly. It is also visible.” — Navnoor Bawa

“A circle is indifferent to being understood.” — The Unicus Investor

“A moat dressed up as a risk” — Clark Tang (via Tae Kim)

“I know that you know that I know” — StreetSignal

The whispered bar for earnings this week is $94–95B versus roughly $92B consensus. Whilst that resolves on Wednesday, the financing debate only becomes auditable when the 10-Q follows.

Here’s a synthesis of that debate from nine authors writing between 12–19 August 2026, within days of the $500 billion MoU (AI compute infrastructure financing platforms) announced on August 10th, so think of this as a one-week reaction corpus to that event. Inference tagged, as always.

Common themes

• Residual value of the GPU is the load-bearing assumption of the entire structure — every substantive post converges on it. In a guest post on Tae Kim's Substack, Clark Tang argues the collateral is “aging more like an aircraft than a smartphone,” citing CoreWeave A100s contracted through 2029 with a 25% fleet price increase in July. Bawa accepts the fungibility argument is stronger than for most equipment but objects that the useful-life assumption in the collateral schedule is supplied by the party selling the collateral, against lease guarantees running five-to-seven years and a roughly two-year product cycle. Unicus takes Burry’s depreciation claim as its explicit starting point and follows the same assumption into loan-to-value rather than earnings. De Bony frames GPU residual value as a credit risk untested through a full cycle, where a demand slowdown hits borrower and collateral simultaneously. Writers: Tae Kim (Tang), Navnoor Bawa, Unicus, Le carnet des marchés.

• The $500bn is an MoU, not money — and the writers who read the release agree on the caveats regardless of their conclusion. Pawan (bullish on the sponsors), Coastal Journal (bearish framework), de Bony, and Bawa all independently flag: non-binding memoranda, third-party capital mobilised “over time,” no agreements executed, and Nvidia’s option to backstop up to $125bn / 25% (UncoverAlpha and de Bony both carry the backstop figure, de Bony via Reuters). Writers: Pawan, Coastal Journal, Le carnet des marchés, Navnoor Bawa, UncoverAlpha.

• Private credit is the destination, and its capacity/health is the constraint. Pawan sizes the opportunity: $1tn+ migrating to private structures because public IG markets can’t absorb it. Bawa sizes the strain: $500bn against a Fed-estimated ~$1.4tn total US private credit market is ~36% of the asset class pointed at one manufacturer’s hardware. Bottarelli’s fetched fragment opens with the WSJ reporting worsening loan health at the big private credit managers. Same pipe, read as opportunity, concentration risk, and existing stress respectively. Writers: Pawan, Navnoor Bawa, Mauro Bottarelli (fragment), Le carnet des marchés.

• Index transmission. Dark Side of the Boom leads with Nvidia at ~8% of the S&P making AI risk uncontainable within tech; Bawa runs the same argument with more precision (24.9% of the index on both sides of one transaction, five of six lending sponsors inside it at a combined 1.10%). Writers: The Dark Side Of The Boom, Navnoor Bawa.

Contradictions across authors

Read more

The ticker and the author are below, as is the writer who made a successful bear call on Moderna fourteen months ago. His thesis could be worth revisiting. I then summarise in detail the authors actively covering biotech/pharma on Finance Substack, including the author whose Substack stopped after he got a job on the buy side at a biotech fund. By the way, if you cold ask an LLM who covers a sector on Substack, you’ll get a short, outdated and unreliable list. BiopharmaStack is built on ~203,000 populated MarketStack data points, and will soon be available in the Terminal’s StackMaps.

Read more

Authors are organised into access tiers:

• Tier 1 — paid, live. A return record of the paid archive and continued paid calls.

• Tier …

Read more

So I decided to look at every technical analysis of the ticker published on Substack over the last few weeks, to see what authors were saying about the levels in the run up to the move. This is what I built PositionWire for, after all.

Of course, users of ’s institutional option-flow database (which is not designed for beginners), would have also seen the following building ahead of Wednesday’s earnings:

• Aug 6-7 — steady put-selling, i.e. quietly bullish

• Aug 10 — big size, bullish lean (7,000 contracts of 120P sold (Aug 21 expiry))

• Aug 11 — the day before the pop (2,750 contracts of 240C bought, Aug 21 expiry, $1.23M premium)

Which is to say the positioning was there to be read, if you were looking in the right place.

Back to the technicians.

I was quite shocked at what I found. I am not going to name and shame the authors who got it completely wrong. Instead, I will celebrate the one author who got the technicals right and the one who nailed the earnings narrative - correctly identifying the specific information the market would react to in his preview.

So this is the author who published the only chart identifying an approaching breakout channel (in the correct direction) in his weekly, on Tuesday, the day before the pop:

Read more

The screenshots show his framework: the charts are m…

Read more

Read more

“…outright gamblers exploiting market structure.”

This, as the professor of market structure - Capital Flows - is raising cash and turning a bit more short-term conservative following a relentlessly bullish July arc, where he maintained ORCL as one of the most asymmetric setups in the market, AI compute as the next asset class absorbing flows, and PURR/HYPE, ES pointing “toward an all time high”.

Across all writers though, bearish sentiment is now coming down. Here’s the full data (30 Jul – 5 Aug vs 23–29 Jul) with the tape excluded, counted by distinct authors.

Bearish sentiment dropped 5.9pt week over week.

The mechanism is bulls arriving, not bears leaving. Bearish calls fell only 10% (347 → 313), while bullish calls jumped 29% (1,116 → 1,442). Volume expanded and almost all the new volume was long. Closes were flat at 99 → 100, so nobody was rushing for exits. In terms of author breadth, the share of active authors making at least one bearish call went 42.0% → 37.4%.

Disclosed longs jumped 71% while short volume froze.

Apple and Meta absorbed the bear flow.

Apple coverage quadrupled (11 authors → 44) and shifted into net bearish. Meta is the same shape one notch down: coverage tripled, bear share doubled, however it’s still net bullish (<50%). CoreWeave and Nebius bears grew faster than the audience: bearish contingent went 1→6 and 1→5 while staying net long as coverage grew 2.8x and 1.6x.

Newly bearish tickers on the week:

The QCOM cluster is Apple-driven — LongYield and Irrational Analysis are both writing about Apple pulling its modem business in-house, so it’s really one story.

USDJPY — 4 authors

Read more

I began writing this two days ago, put on the trade expression yesterday, and was finalising the write up literally moments before news just hit that the elites at Google are leaving to build a purpose-built RSI (recursive self-improvement) machine. So the timing is unfortunate, but here it is anyway.

I believe RSI is about to intensify in the narrative, potentially displacing 'the frontier' as the axis the market organises winners and losers around.

In overly simple terms, RSI is AI that can build more advanced versions of itself such that it gets exponentially better at narrow things that are cheap to verify (algorithms, code, maths). It could advance at its own pace even as general capability (AGI) continues improving at its existing pace rather than accelerating with it. So they are related, but not necessarily coupled.

Speculation and hype over reaching RSI and reaching it sooner than expected is the kind of narrative regime shift you can only profit on before it’s super clear cut. So the research being more neutral than bullish but with strengthening signals is precisely the approaching-inflection point to be positioned ahead of, in my opinion. And markets do not need RSI to arrive for RSI to matter. They only need belief that the probability distribution has shifted and that it becomes the dominant AI bottleneck. In addition, after Leopold, the bull case will seek to evolve.

So how close are we?

In a paper published last month, nine independent economists and researchers concluded that we are not yet experiencing RSI but the trend is rising and “Our model therefore does not rule out the possibility of self-sustaining acceleration in the near future.”

They are careful to distinguish between the definition of RSI from both ‘R&D automatability’ and an ‘intelligence explosion’ - which by the way is not the same as AGI.

The mechanism that decouples RSI from AGI is the narrow/broad capabilities split and this split is what helped me understand what RSI actually is. RSI only requires the loop from capabilities back into algorithmic improvements, it doesn't require any arrow from capabilities into economic output. So you can get runaway acceleration in narrow capability like optimising algorithms, coding, maths, while broad capability, the general human-equivalent competence people mean by AGI, lags behind, because that depends on messier bottlenecks the loop doesn't touch.

By contrast, in the Situational Awareness thesis, automation and explosion are nearly the same event: “automating AI research is all it takes to kick off extraordinary feedback loops” and explosive acceleration. But this paper’s central technical point is that full automation doesn't imply explosion if returns diminish and bottlenecks bind, though they don't think such evidence is overwhelming, so they cannot rule out an extended and rapid acceleration. Leopold’s case rests on the effort increase being so large that it swamps diminishing returns, which is a claim about elasticities he never estimates but this paper does. This tension is probably why the market conflates these two states.

So you can think of Situational Awareness as the maximalist version of the spillover possibility this paper describes. It’s the classic extrapolation-versus-measurable divide between the visionary and the scientists.

Anyway back to the timeline. The researchers think this asymmetric version (runaway acceleration in narrow capability while broad capability lags) seems plausible in the near-term, and could coexist with sluggish real-world impact. When digesting the paper’s conclusions, I wondered about historical patterns of domain experts stating a confident “not yet” shortly before the breakthrough that proved them wrong. And it’s a real pattern, often because the expert was extrapolating from the current method's limits rather than anticipating the next one. Though there is a flip side, nuclear fusion was “perpetually 20-30 years away” for seven decades running, so confident near-term misses go both directions. The RSI timeline debate could land in either camp.

“We are not there yet, and recursive self-improvement is not inevitable. But it could come sooner than most institutions are prepared for.” - Favaro and Clark (2026), When AI Builds Itself

The paper of course says nothing about tickers or sector rotation. So I asked Claude who stands to win and lose once we hit RSI.

Read more

Autotomy /ɔːˈtɒtəmi/

The behavioural mechanism where an animal sheds or discards a body part, such as a tail or limb, to escape a predator’s grasp. It’s a voluntary self-amputation that typically occurs at a specialised, predetermined breakage plane, allowing the animal to survive and often regenerate the lost part later.

My instinctive rebuttal to Tae Ki…

Read more

Nvidia’s 5 year CDS spread (the cost to insure its debt) has been steadily widening since last month, which reads as a warning the credit market sees something the equity market does not. But this note challenges that.

The move began about a week after its bond issuance (which was immateria…

Read more

Last week, I said the point of mining and structuring sentiment and positioning data was to build a longitudinal record. This is the first week there are two points, and there’s a story to tell. But first, I want to talk about one of the only shorts initiated in the prior week, as claimed, by a futures trader who front-ran the commentary crowd. The finding reveals value emerges from interrogating the data, not from the data itself. Which is why I have more late nights writing about it than sourcing it.

Back to the short.

A few hours after the Nasdaq rallied on Tuesday post-soft CPI, the author of this Chinese-language Substack began shorting the futures (NQ), into the early hours of Wednesday. He had waited to see whether long-end real rates would confirm the disinflation story. They did not. Despite the fact that it was the sharpest monthly CPI decline since April 2020, long bonds broke out higher, pricing term premium instead of cooling inflation and so breaking the reaction function “weak data → lower rates → tech rallies”. So he shorted the asset most dependent on the broken link: NQ. The first batch at 30,060, its prior high. Then at 30,040.

“Part of the market formed a benign expectation off new data, and prices started to rise — but the market’s real price-setters did not endorse that expectation, and so the price eventually dived.” — (translated)

His nerve broke on Wednesday’s PPI which produced the last bounce at 29,990, and he covered one batch. Then re-shorted at 29,950.

Read more

A Fable meta-analysis & synthesis of author-positioning and sentiment data based on a snapshot of yesterday, 13 July (270 calls from 121 authors), read within the full call book — 3,633 logged calls from 458 authors, the bulk captured since late June.

(1) What the data clusters around: long energy/oil equities

The densest cluster of authors deploying receipt-backed money right now.

Read more

One of them is setting the oil price, and the bond market’s next move depends on which.

The market just ran a controlled experiment on itself.

Last Monday, OPEC+ waved through another output increase, exports through the Strait of Hormuz were recovering, and Citi mused about $60 crude if traffic normalised. Bonds rallied, Morgan Stanley’s equity desk cred…

Read more

For readers with a particularly short attention span, I appreciate that this analysis will at first glance look like an echo of market sentiment. However, mining and structuring this data will allow me to backtest meta-signals and enables myself and Terminal users to conduct forensic analysis on the ecosystem through time with a depth that is unmatched. In addition, the model is becoming more context-aware as it builds up a longitudinal record of author sentiment and positioning at points through time. My hope is that it will for instance be able to make forward-looking predictions. But I also predict that it will bring about a number of unknown unknowns and that’s what excites me most. In the meantime, there are some genuinely insightful data points that help orientate any reader of Finance Substack, today.

So let’s get into it. The analysis is on a pool of 1,553 verified directional calls inferred from 5,190 notes and posts between 3-10 July, of which 125 are detected as actual positions from 66 authors. I start with the ticker-view before the author cut.

Read more

The Korean financial press has been tracking a shareholder rebellion at an illiquid KOSDAQ name, and it just escalated into a proxy war.

The family-owned and operated company is deeply embedded in the memory buildout, and the rebels include Korea's most famously patient value investor — a fund built in Buffett's image — and Norway's sovereign wealth fund.

Yet the story is all but invisible in the English press.

That was until quietly published an investment thesis on it, which I found through MarketStack’s under-the-radar filter in PositionWire.

Note - I’ve verified the story against the Korean financial press, though several details surface only in Archetype’s primary-source work.

Allow me to set the scene.

Imagine you’re a 75-year-old founder. You own 34.8% of a company you started from nothing in 2000. Your two sons work beside you as executives, but they own essentially none of the equity.

Your company has been a qualified supplier to SK Hynix since 2001 and Samsung since 2002. You sell them the replacement parts for the insides of the machines they use to make chips. Roughly 45% of your most recent quarter’s revenue came from those two customers alone.

And every machine that Samsung and Hynix add for the HBM buildout becomes a perpetual customer the day it switches on.

Now consider the Korean tax law. When a parent gifts listed shares to a child, the taxman doesn’t value them at the price on the day of the gift. He takes the average closing price over a four-month window symmetric around the date of the gift, and applies a rate that tops out around 50%.

Everything you spent your life creating passes to your kids through that formula. For one four-month window, at a moment of your choosing, every won of your company’s share price is a direct levy on your family’s fortune.

So you'd want the stock cheap. Not forever. Just until the window closes.

And you have every tool you need. You control the board — three of the four executive directors are you and your sons. You control the dividend — over the last three years the company earned about ₩160bn and paid shareholders ₩3.6bn. A payout ratio of 2.3%. You personally were paid more than all of your shareholders combined.

You let the cash pile up and sit. ₩230bn of it now, against a market value of ₩440bn — the market prices your entire operating business at less than the cash it holds, about 4x its annual profit. Your closest peer earns less than you and is valued at 3x more.

In the middle of the biggest memory expansion in history, your company is priced like it's dying: 9 years of profits buys the whole thing, 4 once you count the cash.

And this is exactly where you want it.

Every ratio a value investor would circle in red is, from where you sit, tax planning.

For years, nobody could do anything about it. With 34.8% voting as a bloc against a scattered register, every AGM was a formality. The low payouts, the family board, the pay rises in down years: none of it was ever voted down, because none of it could be.

Then, in the space of twelve months, three things happened.

A court took away your votes. In April 2025 Korea’s Supreme Court let stand a ruling — in a case involving Namyang Dairy — that a director who is also a shareholder can't vote his own shares even on the company-wide pay ceiling. A technical confirmation of a technical point, waiting for a test case.

Then a buyer took your discount. In July 2025 a value fund called VIP Asset Management — founded in 2003 by two university students out to prove that patient, Buffett-style investing could work in Korea, and famous for never once going to war with a company it owned — crossed the 5% disclosure line in your stock.

Within a year it had built its stake to 15.64%.

The cheapness you engineered didn’t just deter investors. It handed a sixth of your company, at a discount, to the one shareholder in Korea most likely to eventually do something about it.

And then your shareholders got a fair fight — and you lost it. Twice.

The first fight came at the annual meeting in March. You’d read the room enough to withdraw the ugliest item beforehand — a clause entitling directors removed before the end of their term, the classic anti-takeover parachute, to severance of up to 20x their average pay, on top of ordinary retirement pay — and you trimmed the pay-ceiling rise.

It didn’t matter.

With your 34.8% sterilised by the Namyang precedent, the minority holders were, for the first time in the company’s history, the only votes that counted. They killed the proposal, 69.2% against — in the first proxy season the ruling had ever applied.

You could have taken the hint.

Instead you ran the same play again.

The defeated proposal went back, nearly unchanged, to an extraordinary meeting in June — split into pieces this time, so you could vote your stake on your sons’ pay while sitting out only your own. Electronic voting, allowed in March, was switched off.

The meeting was set for a Monday morning in Gumi, the company's factory-town home, an industrial city two-three hours south of Seoul. To vote against you, a shareholder had to physically show up.

They showed up.

Read more

"I know of only one bird, the parrot, that talks, and he can't fly very high."

— Wilbur Wright, 1908.

The term of art is stochastic parrot, the name for a machine that stitches words together by probability, with no reference to meaning. Coined by a computational linguist and AI-ethics researcher in 2021.

Read more

This edition covers what happened to the moving house story under those reversed condition…

Read more