The Most Crowded Trade in the World
This afternoon, Nvidia closed at a record $208.27 - up +4.3% on the day, market value back above $5 trillion, the Philadelphia Semiconductor Index riding an eighteen-session winning streak. The spark, improbably, was Intel posting its best single day since 1987.
Every buyer of that close is making the same bet: that they haven't missed it.
The base rates say most of them have, or will. Hendrik Bessembinder's census of every US stock since 1926 found only 42.6% beat one-month Treasury bills over their lifetimes, and the single most common lifetime outcome for a listed company rounds to a loss of 100%. Even when investors pick funds instead of stocks, Morningstar's Mind the Gap study finds they earned 6.3% a year over the decade through 2023 from funds that returned 7.3% - about 1.1 points a year lost to buying after rallies and selling after crashes.
So this last lesson in the series isn't another concept. It's the assembly manual: a loop with four steps - read the market, find the idea, evaluate it, trade it - where every step exists because skipping it is expensive in a documented way.
We used Barebone to pull the receipts behind each step: a $15 billion business that became $197 billion in three years, a smart-money paper trail worth six to eleven points a year in the academic record, and a shut strait that made the obvious trade a loser.
Step One: Read the Market
The market runs on two clocks, and confusing them is how you buy the right stock at the worst time.
On the short clock, prices are driven by stories and positioning - Graham's voting machine, which we tested at the start of this series. On the long clock, they're driven by cash flows. The cleanest evidence comes from Vanguard, which tested more than a dozen popular signals against US returns since 1926. The best one - the cyclically adjusted price-to-earnings ratio, which compares price to ten years of inflation-adjusted earnings - explained about 40% of the variation in the next decade of real returns (an R² of 0.43). Over the next year, Vanguard found valuations were poor predictors, and nothing else they tested was reliable either.
In other words: nothing dependable predicts the next twelve months. The price you pay predicts the next ten years.
The one macro number worth respecting on both clocks is interest rates, because rates are the gravity every asset price is computed against - when the risk-free alternative pays more, future profits are worth less today. The textbook case ran live in 2022: the Federal Reserve raised its policy rate from near zero to 4.25 - 4.50% in nine months, and the S&P 500 fell -19.4% while the Nasdaq lost -33.1%.
Then 2023 staged the counterexample. The Fed kept hiking, to 5.25 - 5.50% by July - and the S&P 500 rose +24% while the Nasdaq gained +43%.
Same tightening cycle, opposite outcomes, because by January 2023 the hikes were expected. As the oil tape proved earlier in this series: prices move on surprise, not news - rates included. What moves stocks is rates relative to what the market already believed.
The market you're reading is two markets. Know which clock your trade lives on.
Step Two: Find the Mismatch
Strip the noise from history's biggest winners and most share one signature: a sudden mismatch between demand and supply.
The cleanest modern specimen: ChatGPT launched on November 30, 2022, and demand for AI compute went vertical against a supply of advanced accelerators that essentially one company could provide at scale. Nvidia closed launch day at a split-adjusted $16.89. Its data center segment was a $15 billion-a-year business.
Watch what the mismatch did to that line, fiscal year by fiscal year, per Nvidia's results filings:
| Fiscal year |
Data center revenue |
Change |
| FY2023 |
$15.0B |
baseline |
| FY2024 |
$47.5B |
+217% |
| FY2025 |
$115.2B |
+142% |
| FY2026 |
$197.3B |
+71% |
A business thirteen times larger in three years. The stock followed the cash: from $16.89 on ChatGPT day to today's $208.27 close - a twelvefold move. That is what the loop is hunting for at this step: not a good company, but a moment when demand outruns supply so violently that the cash flows have to be repriced.
But spotting the mismatch is not the same as knowing what to buy. On February 28, 2026, the United States and Israel struck Iran and the Revolutionary Guard moved to shut the Strait of Hormuz, the 21-mile channel carrying roughly a fifth of the world's oil. A textbook supply shock: Brent went from $71.32 on the eve of the war to over $85 within a week, and through $100 by March 8.
The "obvious" trade - big American oil stocks - returned almost nothing:
| Week one of the war |
Move |
| Brent crude |
+17.8% |
| US energy stocks (XLE) |
+1.0% |
| S&P 500 |
-0.7% |
| MSCI South Korea |
-10.7% |
The mismatch was real, but it didn't bind on Exxon - diversified majors hedge, refine, and carry political risk of their own. It bound on the scarce asset nobody could build quickly: tankers, which suddenly had to haul replacement barrels across far longer routes. Frontline, the big crude-tanker owner, entered the war already up +73.9% on the year as positioning built.
The supply-demand question isn't "what's the theme?" It's where, specifically, does the shortage bind - and the answer is usually one layer away from the headline.
Where Ideas Surface First
The script for finding these mismatches used to be: wait for the research report. The modern answer is two feeds, with opposite failure modes.
The fast feed. A tweet takes seconds; a sell-side initiation takes weeks. GameStop's January 2021 squeeze was organized in public, on Reddit, around a fact sitting in public filings - short interest near 140% of the float - and the stock rose +1,625% in a month. Speed is real. So is the noise: the same feeds that surfaced GameStop early surface a hundred dead-end pumps a week, and the only filter is the verification work most posters never do. And markets can be slow even when the document is public - DeepSeek's training-cost paper sat online for a month before the market read it properly and erased $589 billion of Nvidia in a single day, in January 2025.
The slow feed. Three groups of investors are legally required to show you their trades, with a lag - and the academic record on each is specific. (An "abnormal return" below means the return above what the market or a matched benchmark delivered.)
- Politicians. Senators' stock purchases beat the market by about 85 basis points a month - roughly ten points a year - in the 1993 - 1998 sample studied by Ziobrowski and co-authors. They saw policy first.
- Insiders. Executives buying their own stock earned abnormal returns of more than +6% a year in the 1975 - 1996 record (Jeng, Metrick and Zeckhauser). Insider sales, notably, predicted nothing - people sell for taxes, houses, and divorces, but they buy for one reason.
- Super-investors. A portfolio that simply copied Berkshire Hathaway's holdings after they became public beat the S&P 500 by +10.75% a year from 1976 to 2006 (Martin and Puthenpurackal). More broadly, fund managers' single highest-conviction positions - their "best ideas" - outperformed by 2.8 to 4.5 points a year, even as their diluted full portfolios didn't (Cohen, Polk and Silli).
Notice the gray bar. After the 2012 STOCK Act forced congressional trades into rapid daylight, the edge evaporated: a Journal of Public Economics study of every senator and representative from 2012 to 2020 found their picks performed no better than random - slightly worse before transaction costs. Disclosure created the copy trade and then killed it.
The feeds tell you where informed money is looking. They produce candidates, not conclusions - every documented edge above started decaying the day it was published.
Step Three: Ask Three Questions
Every candidate the feeds produce gets the same interrogation, which compresses the middle of this series into one table:
| Question |
The test |
What it filters out |
| Great company? |
Prints cash today and credibly prints more tomorrow |
The modal stock, whose lifetime return rounds to -100% |
| Great price? |
Trades below the value of its future cash, discounted to today |
Great businesses prepaid for decades in advance |
| Great timing? |
Short term: don't fight the crowd. Long term: below value, let time work |
Right thesis, wrong year |
The first question is a filter - necessary because a small minority of companies create essentially all of the market's net wealth. The second is one equation: a stock's worth is the cash it will generate, discounted back to the present; pay less than that and time is your ally, pay multiples of it and even a winner can owe you a lost decade. The third is the two clocks again, applied: momentum and sentiment govern the entry, value governs the outcome.
Three yeses are rare. That's the point. A workflow that approves most ideas isn't a workflow; it's a permission slip.
Step Four: Trade It Like Plumbing
The last step is deliberately boring, and the boredom is load-bearing.
Execution: we ranked the brokerage apps earlier in this series, and the differences - fees, order routing, what the platform nudges you toward - compound quietly. Structure: the evidence-backed shape for most people is a passive core with a small active sleeve, roughly 80/20, so the hunt for mismatches rides on top of a base that doesn't depend on you being right. Maintenance: rebalance on a calendar, not a feeling, and size every active position so a single wrong answer can't end the game. There will be wrong answers.
Nothing in step four finds you a winner. It's what keeps the first three steps running after your first loss.
Where the Workflow Breaks
An honest manual lists its own failure modes. This one has three.
Edges decay in daylight. The Senate edge died with the STOCK Act. Insider and 13F signals are studied by thousands of quants running the same regressions, and the filings arrive with lags - up to 45 days for fund disclosures - so you are always buying yesterday's conviction. Treat the studies above as proof the feeds can carry signal, not as a guarantee of the same numbers going forward.
Mismatches are obvious mostly in hindsight. For every ChatGPT-Nvidia pairing there were a dozen plausible supply-shock stories that died quietly, and the Hormuz tape shows that even a correctly called shock can pay nothing on the asset everyone reaches for. The workflow narrows the funnel; it does not see the future.
The operator is the weak link. That 1.1-point-a-year Mind the Gap penalty wasn't charged by a fund or a broker. It's the cost of running a weighing-machine portfolio with a voting-machine brain - abandoning the loop at exactly the moments it's worth most.
A workflow is not an edge. It's the discipline that keeps you solvent and systematic long enough for your edges to matter.
What This Means
Run the loop manually and it looks like this: a sentiment gauge for the mood, the Fed's statements for rates, Reddit and X for the fast feed, EDGAR for filings and insider forms, a charting platform for the tape, a data site for the multiples, and a spreadsheet to stitch it together. Hours per idea, every single idea - which is why most people skip straight to step two, buy the headline, and become the base rate. Compressing that loop into minutes is, frankly, why we built Barebone.
But the tooling matters less than the order of operations. Read the market before you trust the idea. Find the mismatch before you study the company. Evaluate before you size. Trade it like plumbing.
The market will eventually serve up another mismatch as violent as November 2022. The only question today's record close asks is the one this series has been building toward: when it arrives, will you be running the loop - or buying the eighteenth green candle?
Data: Barebone | Sources: NVIDIA Q4 FY2026 results (SEC EDGAR), Ziobrowski et al. (Journal of Financial and Quantitative Analysis, 2004), Belmont et al. (Journal of Public Economics, 2022), Jeng, Metrick & Zeckhauser (Review of Economics and Statistics, 2003), Martin & Puthenpurackal (2008), Cohen, Polk & Silli, Best Ideas (1991 - 2005 sample), Vanguard, Forecasting Stock Returns (2012), Bessembinder (Journal of Financial Economics, 2018), Morningstar Mind the Gap (2024) | Data as of April 24, 2026
