Famed Medallion Fund “Stretches . . . Explanation to the Limit,” Professor Claims

Here is a nice article from Institutional investor (+ the academic paper that is reference in the article) about the famed Medallion Fund managed by Renaissance Technologies.

Famed Medallion Fund “Stretches . . .Explanation to the Limit,” Professor Claims

In a paper, UCLA professor Bradford Cornell raises questions about how Renaissance Technologies’ flagship fund could
produce 66 percent gross returns since inception. Others — including some who advised him on the paper — are less skeptical.

By Amy Whyte

January 26, 2020

When finance professor Bradford Cornell first saw the annual investment returns of Renaissance Technologies’ Medallion fund, he was “dumbfounded.”

“It was like the sun rising in the west,” says Cornell, a professor emeritus at the University of California Los Angeles. “I’ve been a finance professor all my career, I’ve read thousands of papers on investment performance, and I’ve never seen anything like it.”

The secretive fund’s performance figures — revealed publicly in Gregory Zuckerman’s 2019 book on Renaissance Technologies founder James Simons, The Man Who Solved the Market — paint a portrait of a wildly successful hedge fund that has not once, in 31 years, delivered a negative gross annual return. Struck by Medallion’s “extraordinary” performance, Cornell set about trying to understand what could possibly explain the fund’s reported annual returns, which averaged 66 percent before fees during the period from 1988 to 2018.

The finance professor’s conclusions were published in a brief paper dated December 2, entitled “Medallion Fund: The Ultimate Counterexample?”

Cornell writes in the paper’s abstract: “The performance of Renaissance Technologies’ Medallion fund provides the ultimate counterexample to the hypothesis of market efficiency. To date, there is no adequate rational market explanation for this performance.”

The paper proceeds to analyze the Medallion fund’s performance from a variety of angles: comparing it to the overall stock market, computing the sheer amount of wealth that such returns could have hypothetically created, and running a regression to determine if the results were driven by risk factors. Cornell also looked for answers in Zuckerman’s book and in the results of two other Renaissance Technologies funds — neither of which has delivered returns anywhere close to those of Medallion, he claims. Last, he sought input from asset management executives and prominent academics, including Research Affiliates founder Rob Arnott and Dartmouth College’s Ken French, who were thanked in the paper for providing “helpful comments,” alongside a half-dozen others. (When contacted for this article, Arnott declined to comment on the record about the Medallion fund.)

Ultimately, Cornell wrote that he could come up with no “convincing” explanation for the Medallion fund’s outsize returns, noting that even if Medallion was simply better at trading than any other fund, “the returns are so large, it stretches that explanation to the limit.”

Cornell’s incredulity is not an unusual reaction to the Medallion fund, which has long baffled industry insiders and observers. The secret to Renaissance Technologies’ performance has been debated in news articles and academic circles and on online message boards — and its success no doubt inspired the launch of other quantitative hedge funds that have tried and failed to replicate Medallion’s returns. (Renaissance Technologies declined to comment for this article.)

According to The Man Who Solved the Market, Medallion’s strategy involves holding thousands of short-term positions, both long and short, at any given time. The fund makes high-frequency trades, but has also held positions for up to one or two weeks, per Zuckerman’s description. Robert Mercer, the former co–chief executive of Renaissance Technologies, allegedly told a friend that Medallion was right 50.75 percent of the time when it came to its millions of trades — adding that “you can make billions that way.”

In simple terms, the Medallion fund reportedly makes money in much the same way that a casino does. The house doesn’t always win — but enough small wins over time can add up to large profits.

“In Medallion’s situation they’re probably not taking larger bets — they’re taking small bets that are all about the same in terms of profitability,” explains Campbell Harvey, a finance professor at Duke University’s Fuqua School of Business. Harvey, one of the professors thanked in the acknowledgements of Cornell’s paper, explains that being right just over half the time could theoretically result in “a lot of money.”

He adds, “If you’re doing potentially hundreds of thousands or millions of trades, even a small amount of profitability per trade turns out to be a big amount.”

By Cornell’s estimation, Medallion’s apparent trading skill would have turned a $100 investment in the fund at the start of 1988 into $398,723,873 by the end of 2018. “In 31 years, Medallion would have turned a $100 investment into a $400 million fortune,” Cornell writes.

By comparison, $100 invested in the stock market at the beginning of 1988 — using the Center for Research in Security Prices’ value-weighted index — would have grown to $1,910 over the same time period, with dividends reinvested. Even if an investor had the ability to perfectly predict stock market returns on a monthly basis — and had invested in Treasury bills during times of stock market underperformance — Cornell asserts that the investor’s $100 starting investment would have grown to only $331,288 over that time frame.

Such a comparison makes Medallion’s returns appear improbably high — but it’s the wrong one to make, according to a hedge fund consultant who reviewed Cornell’s paper.

“What Medallion does (and other high-frequency market maker and trading businesses) is a technology business that requires some, but not a lot of capital,” the consultant said by email. “One where the more you spend on quants and computers and data, the more of an edge you have over rivals.”

The consultant suggests that Medallion’s returns would be better judged in comparison to “pure firms that do this, like Hudson River Trading, and not conventional investment funds.” Hudson River Trading, which describes itself as “first and foremost a math and technology company,” conducts algorithmic trades that as of 2014 accounted for 5 percent of all U.S. stock trading, according to a Wall Street Journal article on the firm. (A spokesperson for Hudson River Trading declined to comment.)

Still, even if other firms have found some success using high-frequency trading strategies, Cornell and others interviewed for this story could not identify any quantitative funds that had achieved returns that were as good, and for as long, as Renaissance Technologies has — and certainly not at the scale of the Medallion fund, which manages $10 billion.

“It appears that they’re just better than the rest of us,” suggests French, a business school professor best known for his work with Eugene Fama on asset pricing and investment factors.

In an emailed statement following the publication of this article, Zuckerman commented that “Medallion’s returns are less perplexing to me, though no less remarkable.”

“An overlooked key is ample and cheap leverage,” he added. “You get that by having consistent returns and a crazy-high Sharpe ratio. That, in turn, comes from the firm’s numerous advantages and advances — superior talent, better data, a unique management approach, a focus on mid-frequency trading, a willingness to cap the fund, and more.”

For his part, Harvey believes that Medallion’s outperformance comes down to three factors. For one, the Duke professor suggests, Renaissance Technologies must have built an infrastructure to keep execution costs very low. “I don’t know for sure, but I surmise that the execution ability of this fund must be stellar,” Harvey says.

Cornell makes a similar observation, noting in his paper that “the reported gross returns are after trading costs, [making] Medallion’s performance even more extraordinary.”

Another likely driver of Medallion’s success, according to Harvey, is a high rate of employee retention at Renaissance, allowing the firm to keep its proprietary algorithms and strategies secret.

“Often with hedge funds there’s a lot of turnover, and when there’s turnover ideas get communicated,” Harvey explains. “I think the good ideas that Medallion developed years ago have had very long legs because people stay at the firm. And they have a good reason to stay, because it’s so profitable.” (Renaissance Technologies has also been willing to employ legal remedies to protect its intellectual property.)

Still, though Renaissance appears to have successfully outpaced its competitors for more than three decades, Harvey believes that the Medallion fund’s trade secrets will be found out eventually.

“With the confluence of systematic investment, machine learning, big data, and low-cost execution, it is just a matter of time before people figure out the algorithms that RenTech is using,” he says.

More firms trading in the same way as Medallion would arbitrage away the potential for excess returns — which brings Harvey to the third driver behind Renaissance Technologies’ extended run of outperformance: “They have been extremely disciplined in not taking on too much money.”

The Medallion fund has been closed to external capital since 1993, and analysis of the flagship fund’s annual returns suggests that significant distributions are made each year to keep the fund about the same size. For example, despite the fact that Medallion reported annual net returns above 29 percent every year between 2010 and 2018, the fund’s assets under management stayed at about $10 billion throughout that period.

“They are closed, and whatever profit they make, they pay out,” Harvey notes. “With a lot of funds that have a good idea, they take on extra money, and as that money is applied to the same strategies, the profitability goes down.”

The theory that Medallion’s strategy is capacity-constrained could also explain why Renaissance Technologies’ other hedge funds — Renaissance Institutional Equities Fund and Renaissance Institutional Diversified Alpha — don’t follow the same strategy as Medallion, as Zuckerman reports. The two funds, which unlike Medallion have allowed outside investments, have delivered returns that are “relatively mundane and in no way comparable to Medallion,” according to Cornell’s paper. (For example, the Financial Times reported that RIEF and Diversified Alpha were up 8.5 percent and 3.2 percent, respectively, in 2018; Medallion was up 76 percent.)

“The other two funds are doing well, but it’s nothing out of the ordinary,” the UCLA professor clarifies by phone. “They’re like going out into my backyard and seeing a coyote and a raccoon.”

As to the Medallion fund?

“It’s like I saw a T. rex in my backyard,” Cornell says. “I just don’t get it.”


Medallion Fund- The Ultimate Counterexample.pdf (42.5 KB)

Really interesting article. Thank you for sharing it

The key to Medallion funds success boils down to the following:

  • Outperforming well guarded secrets
  • Limited market size < $10B

Why doesn’t Warren Buffet beat the market anymore? He is the market. It does show as I’ve stated before that we small investors have an edge due to our investment capital being an incredibly small portion of the market. Even if he hasn’t stated so, Jim Simons knows that if his fund gets too large he will loose performance.

Jeff

I would also throw in a LOT of leverage … which is actually part of Buffett’s success too.

How much leverage? 17X, to be precise. See https://www.institutionalinvestor.com/article/b18bk9r2xk3sl2/the-morning-brief-the-secret-to-medallions-returns-leverage (from 2014).
So let’s say you settle on an extremely low volatility trading strategy. Let’s say you make 4% per year on that strategy. Leverage that seventeen times and you get 68% returns. Keep your capital low, keep making 4% a year, and you’re Medallion.

I read the Simon’s biography. My take on how they outperformed is that they had: 1) incredible trading algos, (coded by 100+ PHDs), 6x leverage, and an OBSESSION WITH DATA COLLECTION.

In my opinion, this third factor is where we (the Port123 community) can significantly improve our quant performance. If all of you are like me, you have tested most every permutation on the Port123 database in an attempt to achieve excess returns. I think this is one of the reasons the DMs (and traditional quant funds like AQR–despite their ability to use advanced quant methods) are suffering: most everyone is using similar “tortured” datasets.

The conversion to Factset should be a big plus in this regard. Two huge pluses, off the top of my head, are international stocks and better (from what I have heard) insider trading data.

But, in my opinion, Port123 should not stop there–the Medallion fund soaked up as much original data as they could find. Geov and others have had good suggestions about implementing economic data, etc. I would recommend Port123 license other data sources. One example is Sentimentrader.com. They have an incredibly extensive database of sentiment indicators–way too long for me to list. I doubt a lot of our competition is using these “second level” type of indicators. I’m sure there are other ideas for additional unique datasets.

After using various types of second level indicators as an overlay, my out of sample performance has significantly improved.

To sum up, in my opinion, our greatest chance for outperformance is not with implementing ideas like machine learning, at least not at first (even recent research has pointed out it has not been the holy grail). The answer in my opinion is incorporating (unique) DATA, DATA, DATA.

Doug

Doug,

I think you are 100% correct about the link between DATA and the success of the Medallion fund. Renaissance Technologies taps data in its petabyte-scale data warehouse to assess statistical probabilities for the direction of securities prices in many different markets. Renaissance Technologies, along with a few other firms such as Two Sigma and DE Shaw, has been synthesizing terabytes of data daily and extracting information signals from petabytes of data for more than two decades now, well before big data and data analytics that we are now talking about today.

AQR has been sticking with “factor investing” which are now commonly employed by smart beta ETFs. The “alpha decay” of these well known factors is probably the main reason why AQR’s performance has been doing so bad for the past 2-3 years (except for its risk parity funds). If we ignore the issue of the “alpha decay” and continue to only employs factors which has worked well in the past but may no longer works today without tapping more datasets, it is likely that we will underperform the market and follow the footsteps of AQR.

Regards
James

What are sources of external stock data that we could purchase? Has anyone tested any of these sources and found them to be useful?

deleted

Doug, agree 100%, Data (“alternative Data”) is the way to go.

Factors still work, but “only” very good (which is fine so far by me, bc. still do not have a big price impact based on the size of my positions) with low vol and size (the smaller the better). That river is too small for the big funds, so my guess is that this will stay that way in the next several years (lets see!). International stocks might help, bc. the factors
might not be as exploited as in the US, pretty sure, that some german small and mid caps are very underfollowed, even much
more then in the US.

But I agree 100%, if we want to hold on to an edge, Data is key to adapt to ever changing markets, adapting to change is the name of the game.

Question: how do you overlay (e.g. sentimentrader.com or other “alternative Data” Sources with P123?

Thank you and best regards.

Andreas

Andreas and Jeff,

I think this also makes your points.

Buffett is not the best comparison to the Medallion Fund. Rather Soro’s fund is a better comparison.

As a private fund he has more latitude in how he invests (like the Medallion Fund). Perhaps this serves as and example: “In June 2018, the (Soros) firm was reported to own 15% of Justify, the horse that won the 2018 Preakness Stakes, Kentucky Derby and Belmont Stakes, through the international breeding and racing operation of SF Bloodstock and SF Racing Group.”

But racehorses aside, they both invest in commodities and currencies for example. Things that, I think, Buffett pays no attention to.

Not to mention the advantages both funds have with regard to leverage, hedging and the use of short-term trading options when advantageous.

Jeff, correctly notes that even Buffett is not doing as well as he used to with stocks. We are small but we are playing in the same space as the Medallion Fund (with stocks anyway) unless we are investing in micro-caps, perhaps.

I do not think we are completely immune to some of the problems Buffett is having. More and more we will have to get it exactly right to make a profit, IMHO.

One only need look at the Designer Models for objective evidence on this last point.

-Jim

Guys, we use old technology here. I know people at these places and have been to their offices. I think the team here finally got the message after seeing the dismal results of the DMs (at least my hope is they’ve seen the light).

You will not be able to compete against these firms with what’s offered here. For example, they all optimize in real time and use long and short securities. There is no optimization offered here at P123 and long and short books are built separately.

Well, we can do quite well with the tools of P123.

For example, here is a 500 position port without market timing, beats the S&P500 every year, with AR=20%. Stocks are selected from a universe of 750 stocks.

If one adds some market timing return will be similar to Medallion. But I can’t show you the picture with market timing because of restrictions imposed.

AR=28% from Jan-2009 to now.



Georg,

Could you make one of these highly successful backtests and cherry-picked examples available in your designer models, please.

Thank you.

P123, I thought you had a policy on this.

-Jim

Jim, do you really want to buy 500 stocks? I don’t think so.
I just wanted to demonstrate to people complaining that P123 does not provide sufficient tools, that there is plenty opportunity by using what is available.

Georg/Jim,

I don’t think you can really compare the backtesting results which has been optimized to Medallion’s 66% annualized historical return in the past 30 years which are real returns.

It is easy to assume that we can beat the returns of a lot of funds by playing around with Portfolio123. Many quant firms have tried to replicate Medallion’s return and their stat arb stock trading strategies (which generate the majority of its return) with leverage (as Yuval has pointed out) in the past decade and have proven unsuccessful.

Regards
James

Georg,

I am just saying the best way to really see how good people’s ideas are is by looking at their Designer Models.

We can all show a successful backtest.

If P123 has the tools to be successful then you should help people to understand why the Designer Models are not doing well (including yours).

I have asked you before and I will ask again. You can start by discussing your ideas on AIC (Akaike information criterion) which you have recommended twice in the forum. Or any newer ideas you may have with regard to overfitting.

One could begin to wonder whether you want people to understand the problems of overfitted backtests. Whether you might have a financial interest in people believing in overfitted backtest.

Especially, considering that Marco understands the problem. He was very articulate in why this is a problem. I thought he was clear on this.

Marco was right about this, I think. I welcome any corrections regarding new policies or a change in policies by P123.

-Jim

Jrinne,

instead of attacking Georg YOU could surely design DMs who will pass your own stated criteria with flying colors.
I am looking forward to see them. THAT would be a constructive contribution.

Werner

If I did I would sell my Designer Model and not a series of overfitted backtests.

Don’t you think that would make sense?

Georg has asserted that Marco and Marc can ‘suggest’ that he stop doing that. As if it were even a suggestion.

What did I miss?

I do not always agree with Marc on statistics but his training puts him on the right side of the issue much or even most of the time.

Right on this issue again, IMHO.

I always agree with Marc that I do not like bad statistics-like believing in overfitted backtests.

Marc was even more clear than Marco, I thought. He has been clear on this since the development of Designer models. Again I welcome any correction in this.

Georg has posted about some solutions to the problems of overfitting in the past. Maybe he just changed his mind and thinks more backtests are the solution now.

Maybe, but I cannot explain it.

For sure, opposite ends of the spectrum from Korr123 and James to Marc seem to agree on this. I think they have a point.

-Jim

Jim, the agreement with Marc is not to show any backtests with market timing. I am not aware that all postings of simulations is prohibited.
That would not make any sense.

As to the 500 stock Port, the question that should be asked is: where does the universe come from?
As I said before there are lots of good tools on P123, and we do not need AI to design good models.