Jim Simons: what a retail trader can actually copy from Renaissance

Contents

1. Can a retail trader copy Jim Simons and Renaissance?
2. Who was Jim Simons?
3. What is the Medallion Fund and how much did it return?
4. What strategy did Jim Simons use?
5. Why can't outsiders invest in Medallion?
6. What does many small edges mean in practice?
7. Which parts of Renaissance are out of reach for a retail trader?
8. How does a retail workflow mirror the method at small scale?
9. Which books and sources are worth reading?
10. Frequently asked questions
11. References

Can a retail trader copy Jim Simons and Renaissance?

No, a retail trader cannot copy Jim Simons’s Medallion Fund: its signals, speed and capacity stay inside Renaissance. What you can copy is the firm’s refusal to trade any idea that has not survived a test with costs included. That habit costs nothing but time and a written record, and it scales down to one computer.

Most pages about Jim Simons and Renaissance Technologies stop at the biography and the fund results. Those are the parts you cannot use. The useful part is the order in which the firm did things: data first, hypothesis second, test third, trade only what passed. A retail futures trader with one computer and bar data can keep that order. One comparison helps: Renaissance built a factory for testing ideas. You cannot build the factory, but you can run the same assembly line on a smaller machine, with fewer markets and slower data, as long as every station stays in place.

Four habits transfer at any account size:

1. Test everything before trading it. A rule you believe in is still a hypothesis until it has been run over years of bar data. If it fails, it is dropped without debate, the way a failed experiment is dropped in a laboratory.
2. Charge costs from the very first run. Commission, exchange fees and slippage go into the first test, not the last. A test without costs ranks fast-trading rules at the top because they look smoothest, and those are the first to fail when costs are added. Trading costs in backtests shows how to set them and how to compare them with the average trade.
3. Drop the narrative. A story about why a market should move is not evidence. Renaissance traded patterns it could not always explain, as long as they held up in tests after costs. The reverse also applies: a rule with a convincing story and a failing test is still a failing rule.
4. Count your trials. If you test ten rules and pick the best, the best result is partly luck. If you test ten thousand and pick the best, the best result is mostly luck unless you check it on data the selection did not see. Overfitting is what happens when a rule fits the random noise of one data set and therefore fails on new data, and the more rules you try, the more likely the winner is an overfit. The deflated Sharpe ratio is one way to correct a result for the number of trials behind it.

The fourth habit is the one most retail traders skip, and it is the one the mathematics is clearest about: Bailey, Borwein, López de Prado and Zhu (2014) show with simple arithmetic that the more variants you try on the same data, the better the best one looks by chance alone. The remedy is not to test less. It is to remember how much you tested.

Who was Jim Simons?

Jim Simons (詹姆斯·西蒙斯) was a mathematician and code breaker who came to trading late, built Renaissance Technologies from scientists rather than finance professionals, and let tested patterns decide what the firm traded. The order of his career explains the firm’s culture: the laboratory habit came first, the markets came second.

Before trading, Simons worked as a code breaker for a defence research institute and then chaired the mathematics department at Stony Brook University, where he built a department by recruiting strong researchers. His work with Shiing-Shen Chern, known as Chern–Simons theory, is still used in physics. He left academia in 1978 to trade, and his firm took the Renaissance Technologies name in 1982. Medallion, the fund that made the firm famous, started trading in 1988. Simons died in May 2024.

The hiring policy is the detail that matters for a retail trader. Zuckerman (2019) describes a firm staffed by mathematicians, physicists, statisticians and computer scientists, many of whom had never worked on a trading desk. People trained in a laboratory treat an idea as a hypothesis: you state it, you test it on data, and if the test fails, the idea is finished no matter who proposed it or how good it sounded. That habit, not any formula, is what the firm carried from science into trading.

The headlines that fill the search results for his name are about money. Estimates of his wealth describe his share of the firm’s results over several decades. They say nothing about the method, and a trader learns nothing from them. The biography is worth knowing for one reason: it shows that the method came from outside finance, which is why it looks so different from what a trading desk usually does.

What is the Medallion Fund and how much did it return?

The Medallion Fund is Renaissance Technologies’ flagship fund, trading since 1988 and later closed to outside investors, and the only return figures worth quoting are the ones with a named source. Two sources report them: Zuckerman’s biography and a working paper by Bradford Cornell that treats the fund as a counterexample to market theory.

The sourced figures are these:

• Zuckerman (2019) reports an average annual return of about 66 percent before fees from 1988 through 2018, and about 39 percent a year after fees.
• Cornell (2019) reports that 100 dollars invested at the start of 1988 would have grown to about 398.7 million dollars by the end of 2018, a compound return of about 63 percent a year before fees, and that the fund did not have a losing year in those 31 years.

The two gross figures differ because they measure differently. An arithmetic average adds the yearly returns and divides by the number of years; a compound rate is the single yearly growth rate that would turn the starting amount into the ending amount. The compound rate is always lower when returns vary from year to year. Both numbers describe the same history.

The gap between the gross and the net figure is the fee. Zuckerman reports that Medallion’s fees eventually reached 5 percent of assets and 44 percent of profits, far above what most funds charge. An investor in the fund kept the net figure, not the gross one. When you read a Medallion return, check which of the two it is.

Cornell’s point is that these numbers should not exist under the usual theory of markets. The efficient-market view says that any pattern that pays reliably attracts capital until it stops paying, and that returns above the market are payment for bearing risk. Cornell finds that Medallion’s exposure to the market and to the standard risk factors was negative, so its returns cannot be read as a reward for risk, and that no accepted theory explains the record. That is why he calls it the ultimate counterexample.

A counterexample is not a target. The figures describe a fund that was closed to outsiders, run by hundreds of researchers, trading at a size the firm deliberately capped, over three decades. None of those conditions holds for a retail account, so the number has no role in planning one. Use it as what it is: evidence that systematic testing can find things the theory says are not there. Then turn to the method, which is the part that transfers.

What strategy did Jim Simons use?

There was no single Jim Simons strategy; Medallion combined a very large number of short-term statistical patterns, each traded only after it had held up in testing with costs, and none of them was overridden by hand. The signals themselves have never been published. What is known, mostly from Zuckerman (2019), is the method of finding and keeping them.

The people who built the models came from mathematics, physics, statistics and computer science. Two of the later leaders, Peter Brown and Robert Mercer, came from IBM’s speech-recognition group, where the daily work was finding weak statistical regularities in noisy data and combining many of them into a usable prediction. That background shaped the firm’s approach: price data is noisy, a single pattern explains only a little, and the value comes from testing many candidate patterns and keeping the few that survive.

Two rules of the firm matter for a retail trader. First, a trade needed a pattern that had held up in testing. A convincing story about why it should work counted for nothing on its own, and a pattern with no story was traded as long as the test passed. Second, the firm let the models decide. Zuckerman describes moments when a position looked uncomfortable and the pressure to intervene was strong; the rule was to leave the model alone, because a model that is overridden whenever it feels wrong is no longer the model that was tested.

That order of work, data first and belief last, is the plain answer to what is quantitative trading: the rules are written down in full, the test is run before the trade, and the result of the test decides. A discretionary trader starts from a view and looks for confirmation. A quantitative trader starts from a candidate rule and looks for a reason to reject it.

The figure below shows the two paths that decide it. A good story that fails the test is not traded. An odd pattern that passes the test is traded. The story changes nothing; the test decides.

Why can't outsiders invest in Medallion?

Medallion closed to outside money because short-term edges have limited capacity: each pattern pays only a small amount per trade, and pushing more capital through it moves the price and erases the edge. Zuckerman (2019) reports that the fund stopped taking new outside investors in 1993 and returned the last outside money by 2005, and that the firm kept the fund near a fixed size by paying out profits rather than letting them compound inside it.

Capacity is the amount of money a strategy can trade before its own orders change the prices it depends on. A long-term stock portfolio has large capacity, because buying over weeks barely moves the market. A pattern that lasts hours has small capacity, because the whole position must go in and out quickly, and a large order placed quickly pays for its own urgency in a worse fill. The edge of a short-term pattern is a few ticks per contract; a few ticks of price impact is enough to remove it.

This is why a fund with Medallion’s record did not grow by taking in every dollar offered to it. More capital would have lowered the return on all of it. The firm chose to keep the size where the edge survived, and the only way to do that was to keep outsiders out.

The same logic cuts the other way for a retail trader, and it is good news. A copy of the fund is not available, so stop looking for one. But the method applies best exactly where a few contracts leave no footprint: a strategy traded with one or two contracts on a liquid index future fills close to the quoted price, so a test that assumes one tick of slippage per side is a fair description of what the account will get. Capacity is the reason the fund closed, and the reason the method works at small size.

What does many small edges mean in practice?

Many small edges means holding several weakly related strategies so that their losing periods rarely line up, and it only works when each strategy has an edge of its own. The first half of that sentence is arithmetic that always holds. The second half is the part a measured run below makes plain.

An edge is a repeatable price pattern that, after costs, has paid more than it cost across many past trades. Renaissance did not rely on one large edge; it combined a very large number of small ones. A retail trader can combine a handful, and the everyday picture is a shop that sells both umbrellas and sunglasses. Each product alone has bad weeks, but the bad weeks for one are often good weeks for the other, so the shop’s total sales swing less than either product does.

In trading, the measure of "swing together" is correlation. Correlation is a number between −1 and +1 that says how closely two series of returns move together; a value near 0 means their good and bad periods are largely independent, and a value near +1 means they rise and fall in the same months. Two strategies with low correlation smooth each other. Two trend-following strategies on ES and NQ usually lose in the same weeks, so their correlation is high and combining them smooths little, whatever their names say.

The practical check takes four steps and a spreadsheet:

1. Export the monthly net result of each surviving strategy, after costs, for the same months.
2. Put each strategy in its own column, one row per month.
3. Compute the correlation of every pair of columns (most spreadsheets have a CORREL function).
4. Keep the group whose pairwise correlations are lowest, and count any pair above your cutoff as one strategy when you size positions.

The check is the measured correlation of their returns, not the indicator names or the market names. Two rules on different markets can still lose in the same months. Portfolio backtesting walks through the combination and the drawdown check in detail.

Now the limit. The run below comes from the hosted backtest of AlphaPrime (Valynox): one simple breakout rule with fixed settings, no optimization, applied to all 17 markets of the hosted backtest with one contract each, from September 2018 to October 2026, with each market’s own commission and slippage charged.

Read the three numbers in order. The average monthly correlation across the 17 markets was 0.08, close to zero, so the drawdowns did not line up: added one by one they came to $1,110,925, but the drawdown of all 17 traded together was $432,206, less than half the sum. That is the diversification arithmetic working exactly as the umbrella shop predicts. Then look at the result: only 5 of the 17 markets finished above zero, and the 17 together lost $202,057. Low correlation spread the loss over time and kept the drawdown shallower; it did not turn the loss into a gain, because most of the individual bets had no edge after costs to begin with.

What this run can show is the arithmetic: how much a low correlation reduces a joint drawdown, and that the reduction is separate from whether the strategies pay. What it cannot show is whether a different rule, a different setting or a shorter list of markets would have had an edge; it tests one rule with one setting, long only, and it says nothing about the future. It is one measured example of a general point, not a verdict on breakouts.

Which parts of Renaissance are out of reach for a retail trader?

Tick-level data across thousands of instruments, microsecond execution, capital that supports hundreds of positions and a team of researchers do not scale down, so the retail substitute is bar-based strategies on a few liquid index futures with next-bar execution and one tick of slippage per side. The table separates what you can copy from what you can only replace with a smaller substitute.

Element of the Renaissance approach	Transfers to retail	Retail substitute
Hypothesis testing	Yes	Backtest every rule before trading it
Costs in every test	Yes	Fees and slippage set in the search, not added later
Many small edges	Yes	3 to 10 weakly correlated strategies, each tested alone first
No narrative	Yes	Keep only rules the test supports
Tick data across thousands of instruments	No	Bar data on a few liquid futures
Microsecond execution	No	Next-bar execution with one tick of slippage per side
Capital for hundreds of positions	No	A few contracts across a few markets
Team of researchers	No	Strategy generator plus validation on one computer

The choice of market follows from the gaps. Liquid index futures such as ES, NQ or TAIEX futures have tight spreads and deep order books, so the assumption of a fill on the next bar at one tick of slippage is close to what a retail order gets. Bar data for these markets is available at reasonable cost, and a strategy that decides once per bar does not compete with anyone on speed. Thin markets and sub-second ideas are where the missing resources hurt most, so they are the markets and time frames to leave out.

The gap in the right-hand column is the same one that separates quantitative trading firms from a retail desk in every other respect: they win on data depth, latency, capital and head count, not on a different idea of what a test is. The idea of a test is identical at both sizes. A retail trader who keeps the left-hand column intact is doing the same kind of work, on fewer markets and with slower data.

The team is the hardest gap to see, because it is not a line on a chart. Hundreds of researchers mean a steady stream of hypotheses, each one written down, each one judged by the same standard. A single trader cannot generate and judge that many by hand. What replaces the team is a generator that produces candidates inside a fixed rule space and a validation step that applies one written standard to every one of them, so that the volume and the consistency come from the software rather than from head count.

How does a retail workflow mirror the method at small scale?

The same loop runs on one computer in a fixed order, from the rule space to the exported strategy, with validation before anything reaches the databank and revalidation closing the loop when new bars arrive. The order is the method; changing it is how the method is lost.

1. Fix the rule space and parameter ranges. Decide which indicators, entry and exit conditions and parameter ranges are allowed before you see any result. Changing the ranges after a bad result is fitting by hand, and it breaks the trial count.
2. Generate many candidates. Every combination inside the rule space is a hypothesis. Because the candidates come from a fixed space, the count of what was tried is known from the start rather than reconstructed later; the strategy generation page covers the scope you set.
3. Backtest each candidate with fees, slippage and next-bar execution. Next-bar execution means a signal computed at the close of one bar is filled on the following bar, which is the earliest moment a real order could have been placed. Commission and slippage belong inside every evaluation, so the ranking you read is already net of costs; the backtesting page lists the settings this depends on.
4. Validate with walk-forward windows and thresholds. Walk-forward analysis is a test that tunes a strategy on one window of data, trades the chosen settings unchanged on the window that follows, then moves forward and repeats. Set the thresholds once, apply the same thresholds to every candidate, and stop the later checks when a strategy fails one; strategy validation covers those thresholds.
5. Store the survivors in a databank. A strategy databank is a searchable store of tested strategies with their results, so that survivors from different searches can be reviewed side by side later. Only strategies that passed step 4 go in.
6. Combine the least correlated ones. Compare the survivors’ monthly results and keep the group whose losing months overlap least, using the four-step correlation check from the section above.
7. Export to the trading platform. Export the strategy’s source code to the platform where you place orders and leave the rules as they are, because an edited rule is no longer the rule that passed. The platform export page keeps the current list of supported platforms.

Keep one written record per search, because the trial count and the fixed settings are the evidence that steps 1 and 2 were done in order. The record holds:

• the market, bar size, session and date range of the data used;
• the rule space and parameter ranges exactly as fixed in step 1;
• the number of candidates evaluated;
• the commission per side and the slippage in ticks assumed in step 3;
• the walk-forward window lengths and the thresholds used in step 4;
• the number of survivors and the date of the next revalidation.

For the size of step 2, the published benchmark gives one measured point: AlphaPrime evaluated about 1.2 billion candidate strategies in about 18 hours on one RTX 3090, with commission and slippage in every evaluation (measured 2026-10-05). Real speed depends on the graphics card or processor, the data length and the research settings. The number is useful for one purpose here: it shows that a retail search can be large enough that counting the trials is not optional.

When new data arrives, the loop returns to step 4: the stored strategies are revalidated on the bars that did not exist when they were found, and the ones that fail are retired. That only works if each stored strategy stays next to its results and settings, which is what a strategy databank is for. This is the retail version of the rule against overriding the model: the model is not changed because a position feels wrong; it is retired because the written standard, applied to new data, says so.

Which books and sources are worth reading?

Three sources cover the history, the numbers and the arithmetic, and two pages on this site turn them into a procedure; read them in this order and stop when your search records have changed.

1. Gregory Zuckerman, The Man Who Solved the Market (2019). The most detailed account of Simons, the early models, the hiring from IBM and the people who built Medallion. Read it for the culture and for the rule against overriding the models when a position looks wrong. It contains no signals or formulas, and anyone who tells you it does has not read it.
2. Bradford Cornell, Medallion Fund: The Ultimate Counterexample? (2019). A short working paper with the sourced return figures quoted above and the argument that they cannot be explained as a reward for risk. Read it for the numbers and for a clear statement of why the record defies the usual theory of markets.
3. Bailey, Borwein, López de Prado and Zhu, Pseudo-Mathematics and Financial Charlatanism (2014). Simple arithmetic showing how many strategy variants you need to try before one looks good by chance alone. It is the source of "count your trials", and it is the paper that changes how you keep records.
4. Ernest Chan’s lessons on testing. For a working trader’s view of the same discipline at retail scale, the site’s summary of Ernest Chan’s quantitative trading covers data snooping, costs and the gap between a backtest and live trading in practical terms.
5. A step-by-step process. When you are ready to run the loop from the previous section, the guide to build a quantitative trading process walks through the same steps with the settings to fix at each one.

Frequently asked questions

What strategy did Jim Simons use?

There was no single strategy. Renaissance combined a very large number of short-term statistical patterns, each tested on historical data and traded only after it held up with costs included, and the models were not overridden by hand. The specific signals have never been published; what is known is the method of finding and testing them.

What was Jim Simons’ net worth?

Published estimates vary by source and year, and they describe his share of the firm’s results over several decades rather than anything about how the firm traded. Simons died in May 2024. For a trader, the figure carries no information; the research method does.

Can I invest in the Medallion Fund?

No. Medallion stopped taking new outside investors in 1993 and had returned the last outside money by 2005, according to Zuckerman (2019); it is open only to people connected with the firm. Renaissance has run other funds for outside investors, and their results are not Medallion’s.

Can a retail trader copy Renaissance Technologies?

A retail trader can copy the method but not the resources. Hypothesis testing, costs in every test, many small edges and ignoring narratives all work on one computer with bar data. Tick data across thousands of instruments, microsecond execution, large capital and a research team do not scale down, so the method is applied to a few liquid futures markets instead.

Did Renaissance use technical analysis?

Not in the chart-pattern sense. Renaissance looked for statistical regularities in price and other data and kept them only if they survived testing and costs. A retail trader can treat classic indicators the same way: as hypotheses to be tested, not as rules to be believed.

References

• Zuckerman, G. (2019). The Man Who Solved the Market: How Jim Simons Launched the Quant Revolution. Portfolio/Penguin.
• Cornell, B. (2019). Medallion Fund: The Ultimate Counterexample? Working paper, SSRN 3504766.
• Bailey, D. H., Borwein, J., López de Prado, M., & Zhu, Q. J. (2014). Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance. Notices of the American Mathematical Society, 61(5), 458–471.