The Medallion Fund: History, Returns and Reality

In 1989, the Medallion Fund lost 4.1%. At one point the drawdown was close to 30%, the trading system had become difficult to understand and Jim Simons stopped it.

He told investors they could have their money back. Most stayed. For six months, Simons, Henry Laufer and Elwyn Berlekamp studied the problem. Then they switched the system on again.

In 1990, Medallion returned about 55% after fees.

That reversal matters more than the legend that followed. Medallion did not emerge fully formed from a mathematician's equation. It began with discretionary trading, suffered a serious model failure and was rebuilt by a succession of people who disagreed, left, returned and improved one another's work.

Before Medallion

Jim Simons had already lived a substantial career before finance. He earned his mathematics PhD from Berkeley in 1961, worked as a codebreaker at the Institute for Defense Analyses and became chair of mathematics at Stony Brook in 1968. With Shiing-Shen Chern, he developed the geometric invariants now known as Chern-Simons theory.

The Simons Foundation's account dates the decisive paper, Characteristic Forms and Geometric Invariants, to 1974. In 1976, Simons received the American Mathematical Society's Oswald Veblen Prize. In 1978, aged 40, he left academia to start the investment business that was renamed Renaissance Technologies in 1982.

The early operation was not the modern image of systematic trading. Simons and his colleagues traded currencies and commodities with a mixture of models and judgement. Leonard Baum, Simons's former IDA colleague and a co-developer of the Baum-Welch algorithm, brought statistical methods to currency data. James Ax, a number theorist who had studied with Simons at Berkeley, extended the model-driven work.

Medallion was launched in 1988 under Ax. The name referred to the mathematics prizes won by Ax and Simons. Its first year was positive but unspectacular. Its second exposed the problem with a system that had accumulated complexity faster than understanding.

The six-month rebuild

Simons's own oral history with the American Institute of Physics is unusually direct. Ax's system was losing money. Simons believed he knew why and ordered trading to stop. He rehired Laufer, whose work favoured shorter-term trading, and brought in Berlekamp from California.

Berlekamp was not a hedge fund lifer. He was a Berkeley professor known for coding theory and combinatorial games. He bought a large share of Ax's trading operation, became its chief executive and worked with Simons, Laufer and Sandor Straus on the system.

His Berkeley business biography says they resumed trading in late 1989. Calendar 1990 produced a 55% net return after a 5% management fee and 20% performance fee. Berlekamp then sold his stake to Simons for six times what he had paid 16 months earlier and returned to academic work.

The record is sometimes retold as Berlekamp and Ax jointly inventing the winning system. The chronology is less neat. Ax led the version that ran into trouble. Berlekamp took control after Ax's departure and helped rebuild it with Simons, Laufer and Straus. Straus then carried the revised system forward.

Published reconstructions put Medallion's net returns at 39.4% in 1991, 34% in 1992 and 39.1% in 1993. The precise decimals come from reported fund records, not from Renaissance's current website. What can be said safely is that the post-rebuild results were strong enough to turn a rescue into an institution.

The episode belongs beside the collapse of Long-Term Capital Management in any history of quant finance. Both involved exceptional mathematicians. Only one responded to early model failure by pausing, simplifying and rebuilding before scale made the loss terminal.

Brown, Mercer and the research machine

Peter Brown and Robert Mercer arrived from IBM in 1993. Their work at IBM had used statistical methods in speech recognition and language processing. At Renaissance they helped turn model research into a larger software and data operation.

No public source reveals Medallion's live features or code, and claims that outsiders have identified its exact signals should be treated as fiction. Renaissance has, however, disclosed its philosophy.

In a joint statement to a US Senate subcommittee on 22 July 2014, Brown and colleagues wrote that Medallion's predictions were profitable only slightly more often than not and could be overwhelmed by external events. The model compensated by producing many recommendations, using the law of large numbers to reduce variability.

That is a better explanation than "an AI that always wins". A small expected edge can matter if it is repeated across many sufficiently independent opportunities, sized sensibly and executed cheaply. The hard work lies in proving the edge is not a backtest accident, controlling dependence between bets and preserving it after costs.

Renaissance's current profile gives a glimpse of the machinery around that idea: about 300 employees, 90 PhDs, more than 50,000 compute cores and a research database growing by over 40 terabytes each day. It also says the firm applies a uniform data-driven approach across its funds. It does not say that all funds use the same signals or produce Medallion-like returns.

Simons retired as chief executive in 2010. Brown and Mercer became co-chief executives; Brown continued alone after Mercer stepped down in 2017. Simons handed the board chair to Brown in 2021 and died on 10 May 2024, aged 86.

For profiles of the people around this history, see our greatest quant traders collection.

The return number problem

Gregory Zuckerman's 2019 biography The Man Who Solved the Market is the source behind the figures repeated most often: 66.1% average annual returns before fees and 39.1% after fees from 1988 to 2018. The publisher's description says the firm made more than $100 billion in trading profits.

These are reported historic estimates for a private fund. They are not a current, audited return series published by Renaissance. More subtly, an average annual return is not automatically the compound return earned by one untouched dollar.

The distinction matters because casual retellings inflate the arithmetic. Treating the reported 39% net figure as a steady compound rate over the 31 years from 1988 to 2018, the naive calculation gives roughly $25 million:

$1,000 x (1 + 0.39)^31 = approximately $25,000,000

That is still a hypothetical, not a result any investor achieved. Medallion did not allow one outside investment to compound untouched for 31 years. It stopped accepting new outside capital, bought out the remaining external investors by 2005 and capped fund size by distributing profits.

There is also a technical debate over the reported average. A 2024 paper by Qiang Liu used the fund sizes and trading profits reconstructed in Zuckerman's appendix and argued that 66.1% and 39.1% should not be labelled compounded annual growth rates. Capital caps and distributions break the simple wealth path. Other analyses using the annual series produce different geometric means.

The defensible wording is therefore precise: Zuckerman reported arithmetic average annual returns of about 66% before fees and 39% after fees for 1988 to 2018. We should not extend that number through 2026, call it a current CAGR or use it to manufacture an investable growth chart.

One consistency point is clear. Medallion has not been positive in every calendar year since inception. It lost about 4.1% in 1989. Reports say it avoided another annual loss through 2018, but post-2005 returns are not available to ordinary outside investors.

Capacity, fees and closed doors

Medallion's size was constrained even as its profits grew. Short-horizon signals have finite capacity: trading more can move the price, consume available liquidity and turn an edge into market impact. Distributing gains protects the opportunity set but prevents the simple compounding assumed in fantasy charts.

The fund also became an employee vehicle. Contemporary reporting said Renaissance stopped taking new outside money in 1993 and bought out the last outside investors in 2005. US Department of Labor records are firmer evidence of employee participation: a 2012 Federal Register exemption covered certain employee and spouse retirement-account investments in Medallion vehicles.

That record does not mean every Renaissance employee receives the same access or economics. Our Renaissance compensation guide explains what the firm actually discloses.

Fees changed over time. Berlekamp records 5% and 20% in 1990. By the 2000s, published accounts described fees of 5% of assets and 44% of profits. Because investors and management were increasingly drawn from the same employee group, the economic incidence was unusual: a fee could leave an investor's fund interest while benefiting the management company in which insiders also had stakes.

The tax history deserves similar care. A 2014 Senate report criticised basket-option structures used with banks for tax and leverage treatment. In 2021, Reuters reported a settlement with the IRS that could require about $7 billion in tax, interest and penalties. This was a dispute over historic structures, not proof of how the trading signals worked.

What the history supports

Medallion is strong evidence that many small, measured edges can produce a remarkable aggregate result. It is not evidence that systematic trading always beats discretionary investing. One exceptional survivor cannot answer that question, and the failed models are less visible.

It also shows that model governance is part of model quality. The pivotal act in 1989 was not predicting the next price. It was admitting that a complicated system was losing money and stopping it.

What were Medallion's returns?

Zuckerman reported average annual returns of 66.1% gross and 39.1% net for 1988 to 2018. Those should not be extended to later years or automatically described as a compounded investor return.

Did Medallion ever lose money?

Yes. Published histories report a 4.1% loss in 1989 and a much larger peak-to-trough drawdown during that year.

Can the public invest in Medallion?

No. The fund has been closed to new outside capital since the 1990s, and remaining external investors were bought out by 2005.

Who rebuilt Medallion after 1989?

Simons, Henry Laufer and Elwyn Berlekamp led the six-month study period, with Sandor Straus also central to the revised system. Berlekamp ran the operation during the 1990 turnaround before selling his stake back to Simons.

Who leads Renaissance now?

Peter Brown succeeded Simons as co-chief executive in 2010, became sole chief executive after Mercer's 2017 departure and took over as board chair in 2021.

The great number in Medallion's history is not 66%. It is six months: the time its builders were willing to stop trading and find out why they were wrong.