How Renaissance Technologies built the best investment track record in history

Executive overview

Most investors believe markets can't be beaten consistently. Renaissance Technologies proved otherwise — generating 66% gross annual returns over 34 years, outperforming every other fund in history.

Jim Simons, a mathematician and Cold War code-breaker, realised that finding signal in market noise was the same problem as decoding Soviet communications. He built a firm of physicists, astronomers, and speech-recognition researchers who never looked at company fundamentals — only patterns in data.

The edge: a single unified model, a tiny collaborative team, and incentive structures that made staying more rational than leaving.

From math department to trading desk

• Simons studied at MIT and Berkeley, joined the NSA's civilian code-breaking unit IDA in 1964
• At IDA, he and colleagues published a paper proposing statistical signal processing for stock markets — the blueprint for Renaissance, two decades early
• Fired after publicly opposing the Vietnam War; became chair of Stony Brook's math department, which he built into a world-class research institution
• Left academia in 1978 to trade, setting up shop in a strip mall near Stony Brook
• Early trading partner Lenny Baum brought hidden Markov models — the statistical tool for predicting future states from observable patterns without understanding the underlying system
• Capital came from selling a Colombian flooring company he'd co-founded years earlier

Building the Medallion fund

• Renaissance was co-founded in 1982 with Howard Morgan, mixing currency trading and venture capital (Morgan later spun out the VC side into First Round Capital)
• A near-blowup when Baum went heavily long government bonds and the portfolio fell 40%, triggering his exit
• James Axe and Sandor Strauss moved to California, building Axcom — collecting tick-by-tick intraday data going back to the 1800s, cleaning it meticulously
• Elwyn Berlekamp joined and introduced the Kelly criterion for bet sizing; shifted strategy toward higher-frequency, shorter-horizon trades
• The Medallion Fund launched in 1988 as a joint venture between Renaissance and Axcom; Simons bought out Berlekamp in 1990 at 6x basis — Berlekamp missed out on tens of billions
• First full year: 77.8% gross, 55.4% net

The IBM hires and the one-model breakthrough

• In 1993, Renaissance recruited Peter Brown and Bob Mercer from IBM's speech-recognition group
• Speech recognition uses hidden Markov models to predict the next word from observed sound frequencies — structurally identical to predicting the next price state from market data
• Brown and Mercer had large-scale systems engineering experience that pure academics lacked
• Their key innovation: replace separate models for currencies, commodities, and equities with one unified model covering all asset classes
• A single model meant every researcher's improvement immediately benefited the whole firm — unlike every other quant fund, where teams compete and silo their work
• The fund moved into equities, where deeper markets allowed far greater scale

Performance and structure

• From 1990 onward, Medallion never had a down year; gross returns never fell below 30%
• 2000 (tech crash): +128% gross; 2007: +136%; 2008: +152%; 2020: +149%
• High volatility is when Medallion shines — the models identify panic-driven mispricing that humans create
• Sharpe ratio peaked at 7.5 in 2004 — roughly double the best competing quant funds
• Fund closed to outside investors in 1993; carry raised to 36% (2001), then 44% (2002); outside investors fully removed by 2003
• Fund capped at ~$5 billion (later ~$10–15 billion) to avoid moving markets through slippage
• Total lifetime net returns: 40% per year after fees over 34 years; ~$60 billion in carry generated

What makes it defensible

• One model: all 150–200 researchers and engineers work on a single shared codebase; improvements benefit everyone simultaneously; no internal competition
• Tiny team: fewer than 400 employees total, vs. 2,000–5,000 at Citadel or Two Sigma; median tenure ~14 years; located in East Setauket, Long Island, far from finance social circles
• Incentive structure: the 5% management fee and 44% carry function as a value-transfer mechanism — newer employees earn primarily through the GP (carry) side, long-tenured employees through the LP (investment) side; no rational reason to leave or defect
• Employees' 401(k) plans invest in the Medallion Fund itself
• Data: Strauss's decades of clean historical tick data — going back to the 1800s — is a genuine head start competitors cannot easily replicate
• Lifetime NDAs plus non-competes, reinforced by economics and social community

Trading approach

• Medallion holds thousands of long and short positions simultaneously; average hold time ~1–2 days
• Makes 150,000–300,000 trades per day, mostly in small chunks to avoid moving prices
• Positions are not high-frequency trading; the edge is in non-obvious relationships discovered by machine learning, not speed
• The models surface correlations no human would hypothesise; the firm deliberately avoids imposing human narratives on why signals work
• Leverage amplified returns: basket options allowed Medallion to control ~$60 billion of positions against ~$5 billion of equity capital
• The IRS ruled in 2021 that basket option gains were short-term, not long-term capital gains — Simons alone paid $670 million in back taxes

Politics and leadership transition

• Bob Mercer was a primary funder of Breitbart, Cambridge Analytica, the 2016 Trump campaign, and Brexit; Jim Simons is a major Democratic donor — tens of millions from both
• Simons asked Mercer to step down as co-CEO in 2017; Mercer remained a scientist at the firm
• Simons retired in 2009; Peter Brown and Bob Mercer took over as co-CEOs, growing the fund from $5 billion to $10 billion with no degradation in returns
• Post-Simons gross returns: 77.3%; net: 40.3% — better than under Simons even with higher average fees