Medallion.
One fund. 30 years. ~39 % net annualized through 1998, 2008, 2020. The cleanest evidence in finance that the combinator, not any single signal, is the moat.
A code-breaker's bet on the opposite of conviction.
Founded 1982 by Jim Simons, an ex-NSA cryptographer and former chair of Stony Brook's mathematics department. The Medallion fund itself came online in 1988. Simons' staffing model was deliberately strange. No MBAs, no Wall Street veterans: physicists, cryptanalysts, computational linguists, signal-processing PhDs. People used to looking for very small structure inside very large noise.
The strategy nobody outside the firm fully sees, but the shape of it is published: thousands of weak, short-duration signals, each with a tiny edge over random, each backtested for decades. Holding periods measured in hours to a few days. No single signal is big enough to matter, and that's the design.
The combinator layer is where the work lives: a machine that decides which signals to weight, when to weight them, how to net them, and when to refuse to act because the ensemble disagrees with itself. The output isn't one high-conviction trade. It's a continuous stream of small bets, fused, ranked, and risk-bound.
30 years of an ensemble that never breaks.
Each crisis that broke a single-recipe firm became an outperformance year for Medallion. The ensemble layer is why.
Simons founds Renaissance Technologies
After leaving his post as chair of Stony Brook's math dept and a stint at the NSA's IDA, Simons hires Lenny Baum and James Ax. Early trading systems are intuition-heavy and lose money for two years.
Medallion fund launches
Pivots from discretionary to fully systematic. The combinator architecture (many weak signals fused) replaces the single-model approach that didn't work.
strategy reset · systematic onlyCompounding decade · ~39 % net
The ensemble keeps adding signals: equity stat-arb, futures momentum, FX mean reversion, micro-structure patterns. Holding periods shrink. The fund refuses outside money beyond a hard cap.
capacity-bound by designLTCM dies. Medallion makes 57 % net.
The exact regime shift that destroyed LTCM (every spread correlating to 1) is just another input to Medallion's combinator. No individual signal is big enough to hurt. The ensemble adapts in real time.
+57 % net · in the year LTCM lost 92 %External investors pushed out
Performance fees raised, then investors offered redemptions. By the early 2010s Medallion is fully internal: partners and employees only. The fund cap is real because the alpha is capacity-bound.
GFC year · +82 % net
While long-only strategies post their worst year in 70 years and quant funds get hit by the August 2007 deleveraging, Medallion's ensemble produces its best result of the decade.
best decade-year · in the worst macro yearCOVID year · +76 % net (Medallion) · −20 % (RIEF, public funds)
The ironic split: Renaissance's public-facing funds (RIEF, RIDA, RIDGE), which run different, more constrained models, lose double-digits. Medallion compounds again. Same firm. Different combinator layer.
same firm · different combinators · 96-pt spreadSimons dies at 86 · the firm runs on
Jim Simons passes away May 10, 2024. The fund continues, because the combinator was the moat all along, not any founder's intuition. The architecture is the firm.
We have no individually brilliant signals. We have a system that, signal by signal, catches things slightly more often than it doesn't, and a combinator that knows which signals to listen to today.Paraphrasing the Medallion thesis · per Zuckerman, "The Man Who Solved the Market"
What Medallion got right.
Read at the combinator layer: Medallion is the archetype. Many weak signals (none of them moats individually), short holding periods (no single signal can run the firm over), and a fusion machine on top that does the actual work.
The trap Medallion avoided.
Most discretionary and many quant shops do this: find a strong signal, lever into it. When it works, returns are spectacular. When the regime shifts, there's nothing above to override.
Thousands of small bets · fused.
No signal carries the firm. The combinator does. Holding periods are short enough that a bad signal corrects in hours, not quarters. The ensemble is the alpha.
Three things Medallion proved.
Weak signals + a combinator beat strong signals alone.
Medallion has never claimed it has the world's best individual signals. The thesis is the opposite: signals are commodities. The ensemble is what compounds.
↪ Read the taxonomyCapacity is downstream of architecture.
Medallion caps at ~$10B because the combinator layer is engineered around short-horizon weak signals, which don't scale linearly. The cap proves the alpha lives in the architecture, not in size.
↪ See the architectureThe combinator is the moat, not the founder.
Simons retired in 2010 and died in 2024. The fund kept compounding. A moat that outlives its founder is, by definition, in the system, not the person. That's what makes the combinator layer different from "having a great PM."
↪ View the infographicFour firms. One pattern.
Build the ensemble, not the trade.
Compounding lives in the combinator.