← All articles
28 September 2026

Why Markets Climb on Fear: Nobody Left to Fold

Two poker hands face down on a table beside a stack of chips

Trading securities is often compared to playing Texas Hold’em poker. The analogy is straightforward. A good poker player understands positioning on the table, the strength of hole cards, drawing probabilities and pot-to-bet ratios, tends to have a fairly good memory of the betting sequences that led to the action in front of them, and last but not least places discipline above all.

All of these traits find their counterparts in trading: understanding the market macros, analysing company fundamentals, calculating the win rate of a trade, profit and loss ratios, risk factors, and a fairly good memory of historical and recent market events and price action. And it goes without saying that discipline is paramount in trading. A comparison drawn this way is on the technical aspect. However, playing good poker is a multifaceted mastery, and usually it is the skill beyond the technical that saves the day. Enter playing fear.

Playing fear

A serious player earns most of their profit from hands that never reach a showdown. They bet, everybody folds, and they take the pot without showing anyone what they were holding. Frequently they are holding nothing.

The term for this is fold equity. It is the value of a bet that comes purely from the chance that other players give up, and it is entirely separate from the value of the hand you are holding. A hopeless hand can be a profitable bet if the story is frightening enough and the other players have enough to lose.

The refined version is the semi-bluff. You bet with a hand that is behind right now but could improve later. Now there are two ways to win. The others fold and you take the pot immediately, or they call and you make your hand on a later card. Neither route requires you to be ahead at the moment you put the money in. That is why it is the most dependable aggressive play in the game.

There is an arithmetic constraint on all of this. Say the pot holds 100 and I bet 100 into it. You have to put in 100 to win 200, so calling pays if you are good more than a third of the time. A 34% win rate is enough. Turn it around and it binds me too: if you fold more than half the time, my bet with nothing pays for itself.

So the table settles into a pair of frequencies. I should be bluffing about a third of the times I bet that size, and you should be calling about half of them. Neither number depends on courage, on reading my face, or on whether my story sounds convincing. They fall out of the pot-to-bet ratio, and a player who drifts away from them in either direction is handing money to the other side.

That is playing fear, and the rest of this piece is about the same two frequencies operating in a market.

How a fold actually works

It is easiest to see how that plays out in a single stock.

Take a mid-cap that does not exist. It trades at $40, with 50m shares in the float and 1m traded on a normal day. Three long funds hold 3m shares between them. The retail tail keeps its stops at the obvious levels, under $36 at the prior low and again at $34 where the 200-day runs.

A hedge fund shorts 3m shares, 6% of the float, and makes the bet visible. Visibility comes either from a published thesis or from a persistent offer sitting in a thin book. In poker terms, the falling price is the bet and the size is the message.

At $36 the stops trigger. 1.5m shares of forced selling hit a book that clears 1m on a good day, and the price gaps through $34 to $33. That takes out the 200-day, and now the long funds have a different problem: a risk committee looking at a position down 15% in a name that just broke the level everyone watches. They sell their 3m. Meanwhile, nothing about the fundamentals of the company changes.

The fund has to buy its 3m back, and the people capitulating are the only ones offering size. The folds did not just move the price. They are also the way out of the trade. So the fund bids into the selling, which is the one thing that can slow it. Of the 4.5m shares coming out of frightened hands, 3m go straight back to the fund that frightened the holders out, and that bid is why the fall stops at $30 instead of running further.

What is left at $30 is a holder base nothing can frighten. The stops are cleared, the long funds are out, and the float sits with index money and whoever bought the capitulation. The pot has been collected.

This is when the short case looks strongest on the chart, and short interest rises as new funds put the trade on at $30 because it is working. Then earnings come in fine. There is nobody left who has to sell, so it takes very little buying to move the price, and it goes $30, $33, $37.

So a fold in a market is a trade at a price the folder would have refused earlier. A position cannot be surrendered. It can only be transferred, which means every fold needs a buyer on the other side of it. Each fold removes a future seller. Collect them all and the next player to bet has nothing left to win.

The climb is paid for by the folds

A whole market does the same thing on a longer clock. Nobody needs to be running a raid. A macro scare empties the book without help, and the folds are the same folds. What changes is that at index level there is a proxy for the fear, in the price of protection, and it has 36 years of history.

Take every trading day since 1990 and sort them by how frightened the market was at the previous close.

VIX at the previous closeShare of all sessionsAnnualised returnShare of total compound growth
below 1532%7%27%
15 to 2031%5%20%
20 to 3029%6%20%
above 308%42%34%

8% of the sessions carried a third of everything the S&P 500 has made in 36 years. The calm two-thirds of the record, where almost all of investing is experienced, compounded at 5% to 7%. That top row also carries four times the volatility of the bottom one, so it concentrates the risk as much as the return.

A bet doesn’t win the pot

“Markets rise on fear” treats all fear as one thing. Look only at markets already down more than 10%, so the bad news is real in every case, and split them by how violent the selling got.

Market more than 10% below its highMeanMedian
VIX in the twenties+2.8%+9.3%
VIX above 30+22.7%+22.9%
A calm market near its high+11.2%+12.1%

Only the top row is robust. Take out 2000–02 and 2007–09 and it barely moves, at +26.7%. The twenties do not survive the same test: without those two crises they return +16.9%, better than the calm baseline, and on medians they were never far from it anyway. So the claim has to be narrower than the table first suggests. Violent selling has been followed by strong returns in every era of the record. Moderate fear is just where the historical left tails happen to live.

A bet does not win a pot. A fold wins a pot. Until the capitulation the capital that would fund a recovery is still in the position, hoping, and the pot belongs to no one.

The same test kills a more comfortable idea. Depth of decline, on its own, tells you nothing about what comes next. A market down 20% has no better forward return than one at an all-time high. Cheapness is not fold equity. Only the violence of the selling tells you whether anyone has left.

Which is the inversion at the centre of this. The raider’s edge was fold equity he had not yet collected. The buyer’s edge at the bottom is fold equity that is gone. You are not betting that people will fold. You are betting that they already have.

The best bluff ever made

The purest demonstration of all this was not in equities at all.

In July 2012 the euro was being priced for breakup. Spanish and Italian borrowing costs had reached levels that made staying in the currency union arithmetically unsustainable, and the market was, in effect, betting that the ECB either could not or would not stop the breakup.

Mario Draghi spoke in London and said the bank was ready to do whatever it took, and that it would be enough. In September the ECB announced the facility meant to deliver on that, Outright Monetary Transactions.

Not one bond was ever bought under it.

Peripheral yields fell hard over the following months on a programme with no transactions in it. He described a bet, the other side folded, and the cards were never turned over. The bluff worked, and the part worth keeping is that a threat nobody ever executed did more for those markets than most of the money that was actually spent.

The semi-bluff still running

Draghi’s was a pure bluff. He held nothing and never had to show it. The semi-bluff is the version with a draw attached, and the largest one in the market today is the AI build-out.

The hyperscalers have committed hundreds of billions of dollars, years in advance of the revenue that would justify them. The draw is the straightforward half: the demand may arrive, and if it does the spending was correct.

The fold equity is the other half, and it works on the people who think the spending is a mistake. A sceptical manager has one way to express that view, which is to hold less of the largest and best-performing part of the index than their benchmark holds. It is expensive from the day they put it on and stays expensive for as long as the spending continues. They are paying rent on a thesis that cannot be settled for years. Most cannot afford it, so they come back into the position while still believing the spending is excessive. The capex did not have to be justified. It only had to outlast the people who doubted it.

Michael Burry illustrates it. In November 2025 he argued that the hyperscalers were stretching the useful life of their chips to understate depreciation, and put a figure on the profit that manufactured. Days earlier he had taken Scion off the SEC register, describing what remained as “essentially a friends and family fund.” He stayed a sceptic by removing the clients who could ask why he was behind, and most managers do not have that option.

Listed companies face the same squeeze from the other direction. A company that declines to match the spending is repriced as one that has conceded the race. Not betting becomes more expensive than betting.

You are also the one facing the bet

Everything above is written from the aggressor’s chair. Most investors spend their careers in the other one, being bet at.

Poker’s answer is the minimum defence frequency. You do not have to be right about any individual bet. You have to be in the hand often enough that the biggest wins are not all happening without you.

For investors in the market, it requires the same tenacity. Sell every frightening tape and you are the player who folds to every bet, which is handing free money to whoever is betting. It also forfeits the third of the market’s return that arrives in the weeks when nobody wants to be invested. Call every bet and you do worse still, because you will eventually call the one where they have it. Dip-buying through 2008 at a VIX of 20 was ruinous, and the same instinct above 30 was not.

The market pays for the hands nobody plays. So when a tape is frightening, the question is not whether the fear is justified. It usually is. The question is how much capital the fear has already moved out of the way, because that is what the next move is made of. When there is nobody left to fold, the price has to advance the slow way, on cards actually turned over.

Method notes

Data. ^VIX and ^GSPC daily closes via yfinance, 1990-01-02 to 2026-09-25, 9,251 common sessions. VIX history begins in 1990, so 1987 is out of sample. All cuts are in-sample and descriptive. Forward windows overlap heavily, so no significance tests are claimed and none are quoted.

The regime table. S&P 500 daily returns bucketed by the previous close’s VIX. Total growth over the period is 21.5x.

BucketDaysShareAnnualisedShare of log gain
VIX <152,96432.0%7.3%27%
VIX 15–202,86230.9%5.4%20%
VIX 20–251,82619.7%6.8%15%
VIX 25–308649.3%4.1%5%
VIX >307347.9%42.4%34%
  • Day counts sum to 9,250 against the 9,251 sessions above, because bucketing uses the prior close and the first session of 1990 has no classifier. The unrounded log-gain shares are 27.01, 19.53, 15.44, 4.50 and 33.52, summing to exactly 100; the whole-point versions sum to 101.
  • The table in the body merges 20–25 and 25–30 into one 20–30 row (2,690 days, 29% of sessions, 6% annualised, 20% of log gain). That merge smooths over the sharpest feature here: 25–30 is the weakest bucket of the five at 4.1% annualised and it sits immediately below the strongest. The function is steeply non-monotone right at the threshold the article leans on, which argues for violence of selling rather than level of fear.
  • Annualised volatility runs 9.0% in the calmest bucket to 40.3% in the highest. The >30 bucket is a high-return, high-variance state.
  • Bucket membership is not investable as stated. The classification uses the prior close, so it is implementable, but the buckets are short, fragmented and clustered in a handful of episodes.

The bet-versus-fold split. Forward 12-month S&P return by VIX bucket, split on whether the index was already more than 10% below its running high.

CellnMeanMedian
VIX <20, shallow4,273+11.2%+12.1%
VIX <20, drawdown >10%1,351+9.0%+9.9%
VIX 20–30, shallow1,360+8.8%+12.0%
VIX 20–30, drawdown >10%1,283+2.8%+9.3%
VIX >30, drawdown >10%645+22.7%+22.9%
VIX >30, shallow87+19.2%+21.3%
  • Era robustness, where the two cells part company. Excluding 2000-01-01 to 2003-03-31 and 2007-07-01 to 2009-12-31: the weak cell goes from +2.8% to +16.9% (n=628), above the ex-crisis calm baseline of +12.1% (n=4,374); the strong cell goes from +22.7% to +26.7% (n=284). The strong result survives both crises being removed and the weak result does not exist without them. “Dead money in the twenties” is a statement about 2000–02 and 2007–09, not about a VIX range.
  • On medians the contrast is 2.8pp (+9.3% against +12.1%), not the 8.4pp the means imply.
  • The >30 / shallow cell returns +19.2% on 87 sessions, close to >30 / drawdown. High VIX carries most of the signal with or without the 10% screen, so that screen does less work than the framing implies.
  • Stage reading tested, not supported. Splitting each cell by whether VIX rose or fell over the prior 5 sessions barely separates them: 20–30 gives +2.0% rising against +3.4% falling, >30 gives +21.2% against +24.8%. These are cross-sectional levels and carry no information about position within an episode. An earlier draft read the two cells as sequential stages of one episode; that reading is withdrawn.
  • Both cells use the same day’s VIX close. Lagging it one day to match the regime table gives +2.8%, +22.7% and +11.2%, unchanged.
  • Episode clustering, which is why none of this is tradeable as stated. The 645 sessions in the strong cell fall into roughly a dozen market events: 1990–91, 1997, 1998, 2000–03, 2008–09, 2010, 2011, 2015, 2018, 2020, 2022, 2025. The run from 2008-09-15 to 2009-05-18 is 170 sessions, 26% of the cell. Median drawdown at entry is −30.6%, and 76% of the cell sits below −20%. 600 overlapping observations from a dozen events is a dozen observations.
  • 2008–09 cuts both ways, which is why the body distinguishes the levels. Inside the >30 / drawdown cell, 187 sessions from 2008-07-01 to 2009-12-31 returned +27.5%, above the cell average. The same period inside the 20–30 / drawdown cell is 183 sessions at +2.3%.
  • Negative control: forward 12-month return by drawdown depth alone is flat. Below −20%: +9.7%. −20 to −10%: +8.7%. −10 to −5%: +10.7%. −5 to −1%: +10.3%. At highs: +11.3%. Depth carries no forward information in this sample; the conditioning does the work.

2026, and why it is not used. An earlier draft opened on the first-quarter 2026 selloff: VIX peak 31.0 on 27 March, index low 6,344 on 30 March, recovered to 7,743 by 25 September and +13.1% on the year. It is a clean illustration of the pattern and it is cut, because the drawdown at that low reached only 9.1% and VIX closed above 30 on just two sessions. The episode fails the >10% screen the tables use, its forward 12-month return is not yet observable, and it therefore appears in neither table. The software argument behind that fall is covered in The Meter and the Lease.

The worked hand. The mid-cap in “How a fold actually works” is constructed, and says so in the text. Parameters are chosen to be unremarkable for a US mid-cap: $40 share price, 50m float, 1m shares average daily volume, a 3m-share short at 6% of float and 3 days of volume, stops at the prior low and the 200-day, three long funds holding 3m between them, a risk-committee trigger at −15%, and a recovery path of $30, $33, $37. The price-impact function is not modelled; the moves are asserted to illustrate the sequence.

  • It is an imbalance-of-urgency sketch, not a balanced ledger. Every share sold is bought by somebody, so nothing nets to a supply surplus. The initial 3m short sale has an unnamed buyer, and so does the part of the 4.5m of forced supply the short does not take. The claim is that forced sellers transact with more urgency than discretionary buyers, which moves the price, and that the short’s own bid is the largest single source of demand at the low. Both are interpretations, as is the claim that the covering arrests the cascade.
  • No real name is intended, and rising short interest after a technical breakdown is asserted as a common pattern rather than quantified.

Poker arithmetic. Fold equity is the portion of a bet’s expected value that comes from the probability opponents fold, independent of hand strength. A semi-bluff is a bet with a hand currently behind that retains equity to improve, winning via folds now or the draw later.

  • Frequencies, worked as in the body. Pot 100, bet 100: the caller risks 100 to win 200 and needs to be good more than 1 time in 3; the bettor risks 100 to win 100, so a bluff breaks even above a 50% fold rate. The equilibrium pair is a 50% minimum defence frequency and a betting range about one-third bluffs, a 2:1 value-to-bluff ratio. Smaller bets raise the required defence and lower the bluff share. Standard indifference results, quoted for their meaning and not used in any calculation above.
  • The decomposition behind fold equity is EV = f·P + (1−f)[e(P+B) − (1−e)B] for pot P, bet B, fold probability f and showdown equity e. Fold equity is the f·P term, which contains no e. In a market P is itself a function of f, since the folds create the move, so the market usage is an analogy and not the same quantity.

Limits.

  • VIX is one proxy for how much capital has left the hand, and it is an imperfect one. It measures the price of protection rather than the quantity of it. Direct positioning data (short interest, fund cash, futures positioning, dealer hedging) would test the mechanism properly and is not used here.
  • The Draghi episode is drawn from the public record: the 26 July 2012 London speech, the OMT framework announced 6 September 2012, and the fact that no purchases were ever made under it. The reading of it as a bluff is ours.
  • The AI capex passage is an interpretation of observed corporate behaviour. It is not a decomposition, and no attempt is made to separate the fold-equity component from the draw.
  • Burry: Scion Asset Management’s SEC registration was terminated on 10 November 2025, and he described the remaining vehicle to Bloomberg as “essentially a friends and family fund.” His depreciation argument was made publicly in November 2025, estimating that extended useful lives would understate hyperscaler depreciation by roughly $176bn across 2026–28. Nvidia put options of about $187m notional were disclosed as of 30 September 2025. Nothing is claimed about what he holds now; later reports of his positioning are contradictory and are not used. The reading of the deregistration as escaping benchmark pressure is ours, and he has given other reasons for it.
  • Nothing here is a signal. It is an account of where equity returns have historically been concentrated and why the structure makes sense. Any of these relationships can persist for a decade and then stop.

Not investment advice.