Investing Classics: Jim SimonsJim Simons — The Man Who Solved the Market

Simons bet on a belief opposite to mainstream value investing. He did not ask "can this business be understood?" but "does this price series contain a statistically repeatable bias?" He treated the market as a noisy signal system, decoded with science rather than narrative. Medallion's ~39% net-of-fees annualized over 30 years was achieved after charging a 5% management fee plus a 44% performance cut — the highest in the industry — which hints at the staggering gross. More telling still, the fund deliberately shrank: it limited outside money from 1993, expelled all outside capital around 2005, and capped itself near $10 billion. Today it is open only to employees.

One dollar placed in Medallion in 1988, compounded at 39% net, becomes tens of thousands of dollars by 2018; the S&P 500 with dividends grew less than twentyfold over the same span. Even Buffett never reached that annualized speed. But note: Buffett manages hundreds of billions; Simons's machine can swallow only about $10 billion — capacity is the other side of this coin that can never be flipped over.

Medallion's brilliance is almost impossible to extrapolate. It relies on ultra-short holding periods, very high turnover, enormous leverage, and a closed, small capacity. Anyone trying to "copy" it slams into the capacity wall and the information wall. Mistaking "it existed" for "I can replicate it" is the most common misreading.

• Am I mistaking "this strategy existed" for "I can capture its returns"?
• What is the capacity ceiling of the method I admire — will my capital dilute the edge away?
• Can I survive the leverage and turnover it depends on in the worst case?
• Will this edge vanish as more people discover it?

Simons was himself the author of Chern–Simons geometric theory and a Cold War code-breaker for a defense agency. He carried the code-breaker's mindset into markets: finding repeatable signals inside vast noise. His team was scientists, not finance graduates. The core discipline is called "no override" — Simons said "We never override the computer": once a model is live, you do what it says no matter how smart or foolish it feels at that moment. Because human intuition, ego, and the "story that makes sense" are precisely the sources of systematic bias, and must be kept out of execution.

Traditional funds recruit MBAs and traders; Simons did the reverse, asking applicants only "have you done good science?" Speech-recognition experts Robert Mercer and Peter Brown came from IBM and applied algorithms for handling linguistic noise to price series — with nothing to do with knowing stocks. It was exactly this "outsider" lens that let them see what Wall Street insiders could not.

"Don't override the model" holds only when the model has a real, continuously tested statistical edge; otherwise it is "blindly obeying a bad model." For an individual investor with no model, the transferable lesson is not "don't override the model" but "don't let stories and emotions override the discipline you set in advance."

• Are my trades pre-set rules, or in-the-moment emotional reactions?
• Have I mistaken "a story that makes sense" for "a verifiable edge"?
• By what objective standard do I confirm my method truly works rather than merely got lucky?

This is casino logic — and the real-world version of the Kelly formula (echoing Day 17). A 0.75% edge on a single trade is meaningless; but repeat it hundreds of thousands of times, each as independent as possible, each position small enough that even a losing streak isn't fatal, and randomness gets ground down by the law of large numbers until the edge becomes a near-certain result. The point was never "how much you win per trade," but three things holding at once: is the edge real, are the bets numerous and independent, and is each position small enough not to be killed by random swings. Medallion's ultra-short holding periods and tens of thousands of daily trades are exactly the machine that magnifies that 50.75% into tens of percent a year.

The cautionary counter-example is LTCM in 1998 (see Day 9): two Nobel laureates whose statistical edge may have been real, but who used ~25× leverage and whose many positions were in fact different shadows of the same "liquidity-convergence" bet. When Russia defaulted and correlations jumped to 1, hundreds of "independent" bets blew up together, and the fund was wiped out in two months. Same statistical idea, worlds apart in risk control.

Three ways a tiny edge fails: ① the edge is fake (overfitted to historical noise, gone out of sample); ② the bets are not independent (seemingly diversified, actually many copies of one macro bet, blowing up together); ③ the position is too large (a few losses in a row wipe you out before the law of large numbers can act).

• Does my "edge" still hold out of sample, on data I did not tune it on?
• Are these bets truly independent, or one bet wearing several different coats?
• In the worst losing streak, would I be liquidated or wiped out — unable to "survive" to see the edge pay off?

The strongest evidence comes from Renaissance's own public products. In 2020, the employee-only Medallion soared roughly 76%, while the same firm's outsider-facing RIEF lost about 19–20%, and RIDA and RIDGE lost about 31%. Same geniuses, same building — mediocre-to-bad when the money belongs to outsiders. The reason is those four walls: the outside funds hold longer, run large capacity and low leverage, and cannot access Medallion's short-horizon, high-leverage signals. How extreme is that leverage? From 2005 to 2015 Medallion used "basket option" structures with Deutsche Bank and Barclays to achieve roughly 9:1 leverage; in 2021 its insiders settled with the U.S. IRS for about $7 billion — one of the largest tax cases in history. This is not a world a retail investor can touch.

Simons himself stressed luck and humility repeatedly: he said that each morning he walked into the office wondering not "am I smart today?" but "am I lucky today?" A man who built the greatest record on record still guarded against crediting the result to himself — and that clarity is worth carrying home more than any formula.

The real misuse is treating Simons's legend as a call to "do quant at home too." The vast majority of retail quants pay the transaction costs, hold overfitted models, lack institutional data and leverage, and underperform a plain index fund over time (echoing Day 7, Day 28). What transfers is never his strategy, but his attitude: respect the data, admit ignorance, let discipline pin down emotion.

• Do I want Simons's "strategy" (uncopyable) or his "attitude" (copyable)?
• Facing a dazzling track record, do I first ask "is it copyable, how large is its capacity, on what leverage does it run"?
• If even he asked daily "am I lucky today?", am I treating my own results too much as "skill"?