Krynex

the math behind trading and sizing. no hype, just the equations

[Krynex @kryneeex]

https://t.co/FRf5cRCtdd

[Krynex @kryneeex]

Nassim Taleb, hedge fund manager who lost money three years straight, 8.39%, 13.81%, 3.92%, right before the strategy would have paid off 3,600%:

"We will not lose everything overnight. We will slowly bleed to death."

That is what Taleb told his own traders. This is the Kelly problem inverted. The position was sized correctly. The capital simply was not his to keep sizing with.

Empirica bought deep out-of-the-money options, betting on the kind of extreme event nobody prices correctly. In 2000, that bet paid 56.86 percent while the dot-com crash destroyed everyone else. Then came 2001: negative 8.39 percent. 2002: negative 13.81 percent. 2003: negative 3.92 percent. Three years of small, steady losses, exactly as the strategy was designed to produce. Assets under management peaked near $375 million, then investors started pulling out during the bleed, unable to distinguish a correctly sized position from a failing one.

By 2005, Empirica wound down. Taleb stepped away.

Same point as my article: sizing determines whether you survive long enough for the edge to pay off, but sizing assumes the capital stays put. Taleb's own money could absorb the bleed. His investors' money could not, because their patience was never part of the Kelly calculation. The formula sizes your position. It does not size your client's nerve.

Here is the trap. Three years after Empirica closed, in 2008, the exact kind of extreme event the strategy was built for arrived. Investors who left in 2002 and 2003 to avoid the small losses missed the payoff entirely. The fund that eventually captured it, Universa, launched in 2007, one year before the crash it was positioned for.

The lecture is free, on Cambridge's own channel, where Taleb explains exactly why average outcomes and single-path outcomes are not the same number. Almost nobody watching it changes how long they are willing to hold a losing position that was never actually losing.

[Krynex @kryneeex]

https://t.co/FRf5cRCtdd

[Krynex @kryneeex]

@becamehokage exactly, survivorship stories cut the part where nothing happens for weeks. that part is most of the story

[Krynex @kryneeex]

Hayden Adams, founder of Uniswap, who had five weeks and $65,000 left to finish a codebase that needed to be rewritten from scratch:

"There were just five weeks to go, the busiest five weeks of my life. I had been unemployed for five months, living off cryptocurrency I was fortunate enough to have bought earlier in the year."

This is the Kelly problem without a single dollar changing hands. The bankroll was not capital. It was runway.

Adams had no funding, no team, and a codebase that needed to be rewritten almost entirely from scratch. He had one asset: five months of savings that would run out regardless of whether Uniswap worked. Most founders in that position hedge. They keep applying to jobs. They diversify the bet across multiple projects, multiple safety nets, multiple exits. Adams did the opposite. He committed the entire remaining runway to a single deadline: Devcon 4, five weeks out, a launch date he could not move and could not partially attempt.

Full Kelly is only correct when your estimate of the edge is exact and your capital can absorb being wrong. Adams had neither guarantee. What he had was a $65,000 Ethereum Foundation grant, a formula he believed in, and a hard stop where the runway physically ended whether the code was ready or not.

Same point as my article: the size of the bet is not always measured in dollars staked on a trade. Sometimes it is measured in months of unemployment spent on one unproven idea instead of spread across several safer ones. Adams sized his personal runway at full conviction, all in, on a codebase that was still broken five weeks before launch.

Here is the catch. This is survivorship bias's favorite kind of story, the founder who bet everything and it worked. What the story leaves out is that Adams announced the deployment to roughly two hundred Twitter followers, and for the first several weeks, almost nothing happened. Full-size bets on an unproven edge do not usually look like conviction in real time. They look like silence.

The grant application is public. The launch announcement is still on Twitter. Almost nobody who reads it changes how much runway they are willing to commit to their own unproven idea.

[Krynex @kryneeex]

https://t.co/FRf5cRCtdd

[Krynex @kryneeex]

@0x_Ito the loan structure is the underrated part, forcing 5% simple interest with no early withdrawal for a century removed every human decision that could have broken the compounding

[Krynex @kryneeex]

@Di_Krass_ purged CV is the fix most retail backtests skip entirely, they’ll k-fold cross validate time series data like it’s iid and never notice the leak

[Krynex @kryneeex]

@isssa0x the punchline rule is the one that actually gets violated most in crypto twitter threads, everyone ends with a call to action instead of the one line people would repeat

[Krynex @kryneeex]

@cartmanx404 the 38 year gap gets more interesting when you remember Sharpe almost didn’t write about it, he found the paper by accident while searching for a dissertation topic that actually had funding behind it

[Krynex @kryneeex]

@kolloeblackking Thanks bro!

[Krynex @kryneeex]

Ray Dalio, founder of Bridgewater Associates, who borrowed $4,000 from his father after being completely right about the 1982 crisis:

"Eight years after I started Bridgewater, I had my greatest failure, my greatest mistake. I had calculated that American banks had lent much more money to emerging countries than those countries were going to be able to pay back. In August 1982, Mexico defaulted. I was right. And I lost so much money I had to let go of every person I had hired."

This is the Kelly problem stated as plainly as it will ever be stated. Being right and staying solvent are two different skills, and only one of them compounds.

Dalio had calculated the crisis correctly. When Mexico defaulted in August 1982, his thesis printed exactly as forecast. He had testified before Congress that a depression was coming. He went on national television confident enough to stake his reputation on it. Then the opposite happened. The stock market entered one of the biggest bull runs in American history. His position was sized as if the thesis could not fail. It failed anyway, not because the analysis was wrong but because timing and sizing were never part of the calculation. Bridgewater burned through nearly all its capital. Dalio had to fire every employee until the company had one person left standing: himself.

He borrowed $4,000 from his father to pay his family's bills.

Same point as my article: the formula does not measure whether your macro call was correct. Dalio's call was correct. It measures whether the size of your exposure left you enough capital to survive the gap between being right and the market agreeing with you. That gap ran eighteen years in Dalio's case, not eighteen days.

Here is the catch. Most traders read this story and hear "he was humbled, then he learned." What actually happened is narrower and less comfortable. He rebuilt the entire firm around one rule: never again size a position as if certainty were possible. Every principle Bridgewater runs on today traces back to a single afternoon where full conviction met zero margin for error.

The lecture is free, eighteen minutes, on TED's own channel. The $4,000 loan from his father is in the transcript. Almost nobody watching it changes their position sizing the next day.

[Krynex @kryneeex]

https://t.co/FRf5cRCtdd

[Krynex @kryneeex]

@tigerfl0w 66% gross before fees though, net to investors was closer to 39%, still absurd but the number that actually got reverse-engineered by Wall Street quants is the net one

[Krynex @kryneeex]

@auren_xbt the correlation part is the piece people skip, uncorrelated is doing all the work in that math, most portfolios aren’t nearly as uncorrelated as traders assume

[Krynex @kryneeex]

@thedelost the Black-Scholes connection is the part that gets skipped most, an equation built with zero finance in mind became the backbone of every derivatives desk two decades later

[Krynex @kryneeex]

@cartmanx404 exactly, that’s the whole thesis in one sentence. size buys you the right to be wrong.

[Krynex @kryneeex]

George Soros, hedge fund manager whose Quantum Fund lost $800 million in a single day on Black Monday, 1987:

"I was convinced the crash would start in Japan. That turned out to be an expensive mistake. The Dow dropped 22.6 percent in one session, the largest one-day fall in history, and I was positioned on the wrong side of it."

This is the same lesson Kelly sizing teaches, just running in the opposite direction. Being wrong is not the disaster. Being wrong at full size is.

Soros had built a thesis about where the crash would originate. He was betting on a specific sequence of events across two continents. When the sequence broke, his fund lost 30% of its value in a single trading session, one of the fastest drawdowns in hedge fund history. Most managers who lived through that day never fully recovered psychologically. Stanley Druckenmiller, who watched it happen, later said nearly every legendary name he knew from that era became nonfunctional for months afterward.

Soros did not. His fund still returned 10% that year.

The reason is not luck. It is the same reason Kelly sizing works when it is followed and fails when it is not. Soros never bet with the assumption his thesis was certain. Even at 30% down in a single day, the position size left him solvent, functional, and able to trade his way back before the year closed. A trader running Kelly at full conviction on a single-country thesis does not survive a 22.6% single-day move intact. A trader running fractional sizing does.

Same point as my article: the formula does not ask whether your prediction was right. It asks how much of your capital you were willing to lose if it was wrong, and whether that number left you enough to keep playing.

Here is the catch. Every trader believes their thesis is the one that will not need a safety margin. Soros believed the crash would start in Japan. It started in the US instead. The specific direction of the error is never the part you can predict. The size of your exposure to being wrong is the only part you control.

The math on position sizing after a catastrophic loss is public. Soros wrote about the exact mechanics of his own drawdown recovery in lectures he gave at Central European University in 2009, for free, to anyone willing to watch seventy-eight minutes of footage. Almost nobody in retail crypto has.

[Krynex @kryneeex]

https://t.co/FRf5cRCtdd

[Krynex @kryneeex]

@0xDominiqq the writing-down-every-mistake part is the actual mechanism, most people repeat the same sizing error because they never turn the loss into a rule

[Krynex @kryneeex]

@0xLupenn the gambler’s ruin setup is a nice bridge here too, same random walk math, just a different boundary condition than the smoothing problem

[Krynex @kryneeex]

@enzoxbt variance drag is exactly why full Kelly sizing is so brutal in practice, the formula maximizes the geometric mean but most people are still mentally anchored to the arithmetic one

[Krynex @kryneeex]

@tsukiema_ the odds being public is what makes it work, if it were secret nobody could verify their own attempt was legitimate, and the whole point is the proof

[Krynex @kryneeex]

@kestrelxbt exactly, that’s the whole framework in one line. size is what buys you the time to be right

[Krynex @kryneeex]

Michael Burry, hedge fund manager who predicted the 2008 financial crisis three years early:

"I was right. I forecast the bubble would burst as early as 2007. By 2005 I was certain. And I nearly went bankrupt in 2008 before the market finally moved my way."

This is the Kelly Criterion problem in its purest form. Correctness does not matter. Sizing does.

Burry had the thesis. He had the data. He had the numbers showing subprime lending was a doom machine. By 2003 he was already convinced the housing market would collapse and take the economy with it. He sized his bet accordingly — all in, conviction level 100%, bankroll allocation matching his certainty about the outcome.

What happened next is the part nobody talks about. The market continued higher. Housing prices kept appreciating. Burry's fund bled capital for three straight years because being right does not pay you while you wait for the market to catch up. By 2008, when his thesis actually printed, investors had already left. The fund survived but the damage was done. Timing had disintegrated him.

Same point as my article: Kelly Criterion teaches you that conviction and accuracy are not the same as correct sizing. Burry's conviction was 100%. His accuracy was also exceptional — he was one of the few who saw it coming. But his sizing was based on a hidden assumption: that timing would cooperate. Timing didn't cooperate. And when your bet is 100% of bankroll and time is working against you, even a correct thesis becomes a death sentence.

The fix is in Chapter 5 of the Kelly framework — fractional Kelly, portfolio monitoring, and most importantly, the willingness to size smaller than your conviction suggests. If Burry had sized at quarter Kelly instead of full Kelly on that trade, he would have survived the drawdown without the psychological panic that made him vulnerable to pressure from investors and the market.

Here is the trap. Every trader thinks they're different. That their edge is so clear, their data so clean, their conviction so justified, that they can bet bigger. Burry thought that too. And he was still nearly ruined by the gap between being right and sizing correctly. You are not smarter than he is. The math is the same.

The lecture is free. The lesson cost him hundreds of millions in opportunity cost. It cost you nothing except the willingness to bet smaller than feels right.

[Krynex @kryneeex]

https://t.co/FRf5cRCtdd

[Krynex @kryneeex]

@undefinedKi the six month cycle only works if you actually rewrite from scratch though. most people just re-add the same rules from memory, which defeats the point of clearing accumulated cruft in the first place