The Greatest Geometric Balancers: Renaissance Technologies, Part II

Let’s pick up where we left off earlier in our exploration of the Geometric Balancing techniques hiding inside Renaissance Technologies’ spectacular returns.

For those that don’t remember, we ended part one with Renaissance Technologies–the greatest investment company the world has ever known–having produced their finest returns to date in 1990. Elwyn Berlekamp took over investment decisions a couple years before, implemented ideas from John Kelly and Claude Shannon, returned 77% before fees in 1990 and then decided to head back to academia.

Now we will explore how RenTec refined the system after he left.

Refining the System

Elwyn Berlekamp was gone, but he had set the foundational process for producing amazing returns with Shannon’s and Kelly’s ideas. The firm hired other mathematicians to carry on the work, one being Henry Laufer who worked with Simons at Stony Brook.

Laufer made an early decision that would prove extraordinarily valuable: Medallion would employ a single trading model rather than maintain various models for different investments and market conditions, a style most quantitative firms would embrace. A collection of trading models was simpler and easier to pull off, Laufer acknowledged. But, he argued, a single model could draw on their vast trove of pricing data, detecting Correlation opportunities and other signals across various asset classes. Narrow individual models can suffer from too little data.

Gregory Zuckerman, The Man Who Solved the Market, Chapter 8, page 142

Why do you build a single model? To focus on correlation between investments. Understanding correlation is absolutely necessary to maximizing geometric growth. You simply can’t do this with multiple models, but with a single model it’s theoretically possible. This is a big difference between RenTec and many other firms.

Everyone Has a Single Model

In my opinion, there isn’t a single fund in the world that actually runs multiple models. They all actually run a single model because the fund is only one fund, just like your own personal total wealth portfolio is only one portfolio.

Multi-strategy funds may think they are a combination of multiple strategies, but really they are a single model which chooses to ignore the internal correlation between their individual strategies.

RenTec clearly realized this. Portfolio correlation is vitally important to maximizing geometric growth, which Elwyn Berlekamp embedded at the foundation of their strategy. Correlation determines the position size for every investment in the portfolio. To ignore it is just putting your head in the sand and hoping for the best.

Now this isn’t such a big deal if you have average expectations. Are you aiming for 10% returns? Probably not that big of a deal to ignore internal correlations between investments. But do you want 60% percent returns each year? The only way to strike confidently (with leverage) when the model says to do so is to employ a single model that calculates when to bet big, and how much.

This is a major lesson in how RenTec invests. Don’t fool yourself into thinking you have multiple strategies. You don’t. No one does. You only have one portfolio, therefore a “multi-model approach” is just a single model approach that ignores internal correlation.

The Complexity Compounds

The biggest challenge of calculating a single model comes from the compounded complexity in every additional investment. A 3 asset model is complex enough (3 correlations). Adding the 4th asset is much more complicated than adding the 3rd (6 correlations). And adding the 5th is magnitudes more complicated (10 correlations). This is because you have to deal with an ever growing web of investment correlations, and parameter error consequences. Every new asset’s correlation must be monitored with each current asset in the portfolio, and they grow very quickly.

It was like finding a common solution to hundreds of equations simultaneously.

Gregory Zuckerman, The Man Who Solved the Market, Chapter 10, page 189.

The only way to build a portfolio so complicated is to hire the world’s best programmers and mathematicians. Which is why I believe RenTec has always done just that.

Predicting Volatility and Correlation

In part one, I described how I believed RenTec was using their treasure trove of data to do more than just predict returns. They were also predicting volatility and correlation. Now we see a quote re-confirming this.

The team uncovered predictive effects related to volatility, as well as a series of combination effects, such as the propensity of pairs of investments–such as gold and silver, or heating oil and crude oil– to move in the same direction at certain times in the trading day compared with others.

Gregory Zuckerman, The Man Who Solved the Market, Chapter 8, page 144.

So here we have clear evidence they are predicting volatility. Interesting the quote doesn’t point out that RenTec is predicting correlation, but that’s what “the propensity of pairs of investments to move in the same direction” is. Almost like someone doesn’t wasn’t to admit the importance of correlation within their strategy.

Expanding to Stocks

We saw in part one, RenTec focused on currencies and commodities in the start. But to get big, they had to expand into stocks. Exploring how they did this, and the challenges they overcame, shines further light on their strategy’s inner workings.

Not completing desired trades resulted in more than just poor performance. The factor trading system generated a series of complicated and intertwined trades, each necessary to score profits while also keeping risk at a reasonable level….failing to get just a few moves done threatened to make the entire portfolio sensitive to market shifts, jeopardizing its overall health. And missed trades sometimes cascaded into bigger systemic problems that compromised the accuracy of the entire model.

Gregory Zuckerman, The Man Who Solved the Market, Chapter 10, page 188.

Now some of this is definitely a commentary on long-short trading. Obviously if you only take one leg of a long-short trade, the trade isn’t complete and may not be a good one any longer. But I think it’s more than this.

The part about “intertwined trades”, “cascading into bigger systematic problems”, “threatened the entire portfolio” and “compromising the entire model” says to me this is bigger than just long-short trades. It says to me the model was built around the correlations of all pieces to one another. RenTec is looking at the correlation of everything in the portfolio not just the relationship between a long side and short side in a trade. If one piece is missing, it can throw off the expected volatility of the entire portfolio dramatically changing a great portfolio into an average or a poor one.

This is most critical when using leverage.

Leverage

RenTec averages over 60% return per year. It’s not possible to achieve these kinds of returns without using leverage, and lots of it. Any mistake when applying lots of leverage would be deadly. So challenges arising from applying leverage would be absolutely critical to them.

The trading system scored sizeable gains when it managed tiny amounts of money, but when Simons fed it leverage and the trades got bigger, profits evaporated. Brown and Mercer’s simulations kept saying they should be making money with larger sums but the system’s actual moves were losers…

Gregory Zuckerman, The Man Who Solved the Market, Chapter 10, page 194.

Many may see this quote as about the investments being limited by capacity. I don’t because it says they added leverage, not more funds to invest. This is an acknowledgement that they had a scaling problem with their models. Just like when Elwyn Berlekamp years before said they were sizing their investments wrong, the same thing was happening here, but from the opposite side. The investments stopped working because they levered their portfolio too far.

And this wasn’t a math problem on their part, but a programming mistake.

Now the simulator’s algorithms could finally recommend an ideal portfolio for the Nova system to execute, including how much borrowed money should be employed to expand its stock holdings. The resulting portfolio seemed to generate big profits…

Gregory Zuckerman, The Man Who Solved the Market, Chapter 10, page 194.

See, it was about borrowing the correct amount of money and getting leverage correct. It was a Kelly sizing issue.

Academic Research is Nonsense

Simon’s team kept exploring for edges. They read hundreds of leading academic papers.

After reading several hundred papers, Simons and his colleagues gave up. The tactics sounded tantalizing, but when Medallion’s researchers test the efficacy of the strategies proposed by academics, the trade recommendations usually failed to pan out. Reading so many disappointing papers reinforced a certain cynicism within the firm about the ability to predict financial moves

“Any time you hear financial experts talking about how the market went up because of such and such–remember it’s all nonsense”.

Gregory Zuckerman, The Man Who Solved the Markets, Chapter 11, page 198-199.

An admission that the firm doesn’t predict financial moves? They don’t actually know if their investment is going to go up or down, and they don’t think anyone else does either. To me, this is even more proof that Renaissance’s edge has less to do with getting the returns correct than most think, and more to do with predicting volatility, correlation and sizing correctly which the book already clearly stated they focus on.

Second, I just can’t help to enjoy the poke at academic strategies which aren’t actually useful.

The Essence of Geometric Balancing

Now for a set of quotes which reflect the essence of geometric balancing.

Risk Management

If a strategy wasn’t working, or when market volatility surged, Renaissance’s system tended to automatically reduce positions and risk.

Gregory Zuckerman, The Man Who Solved the Market, Chapter 11, page 213.

This is pure Kelly position sizing. When volatility spikes, reduce exposure.

Programming

[They] treated their challenge as a math problem, just as they had with language recognition at IBM. Their inputs were the fund’s trading costs, its various leverages, risk parameters, and assorted other limitations and requirements. Given all of those factors, they built the system to solve and construct an ideal portfolio, making optimal decisions, all day long, to maximize returns.

Gregory Zuckerman, The Man Who Solved the Market, Chapter 10, page 189.

Trading costs, risk parameters, assorted other limitations and requirements. Solving for the optimal decisions to maximize returns. Notice this isn’t on/off signals. They are solving for a portfolio to maximize returns. Sounding familiar?

Betting Algorithm

Simons was challenging them to solve yet another vexing problem: Given the range of possible trades they had developed and the limited amount of money that Medallion managed, how much should they bet on each trade? And which moves should they pursue and prioritize? Laufer began developing a computer program to identify optimal trades throughout the day, something Simons began calling his betting algorithm. Laufer decide it would be “dynamic,” adapting on its own along the way and relying on real-time analysis to adjust the fund’s mix of holdings given the probabilities of future market moves–an early form of machine learning.

Gregory Zuckerman, The Man Who Solved the Market, Chapter 8, page 144.

Hmmm, let’s disect this.

The problem is to determine how much to bet on each trade with a limited amount of money in a betting algorithm.

That’s Kelly.

Dynamically adapting on it’s own with real time analysis to adjust the fund’s holding given the probabilities of future market moves?

“Probabilities of future market moves” is partially about predictions on volatility. “Dynamically adapting on it’s own with real time data,” sounds a lot like what’s running at the top of the blog.

This paragraph is geometric balancing nearly perfectly explained.

Now if I, a mechanical engineer with good math skills and data pulled from free sources around the web, can develop a strategy that only uses 4 assets to produce returns rivaling some of the better funds in the world, what do you think a team of actual world class mathematicians with the best data set in world, employing every asset at their disposal could do using the same principles?

Do returns over 60% a year, with a Sharpe Ratio of over 2 for 30 years really sound improbable? I’m convinced they’re not.

Passion and Excitement

Driving to Stony Brook with a friend and Medallion investor, Simons could hardly contain his excitement.

“Our System is a living thing; it’s always modifying,” he said. “We really should be able to grow it.”

Gregory Zuckerman, The Man Who Solved the Market, Chapter 8, page 144-145.

People have said they can feel my enthusiasm for this blog. I hope so, because I have a lot of it. Once you’ve seen the possibilities of these ideas open up it’s easy to get excited, and dream of where the system can grow. So I totally get where Simons was coming from with this quote. Once you see the Kelly criteria and Shannon’s demon implemented well within an investment strategy, the possibilities feel limitless.

Summary

This is how I would summarize RenTec successes:

• Simons and Baum were very strong at building models to predict the near future in stocks. Their code breaking backgrounds gave them an edge in this front. However, “very strong” in this area was still very imprecise and the edge they gained was small. It was very easy to squander this edge with poor risk management and position sizing, which is why they were often not successful early.
• Elwyn Berlekamp knew about proper position sizing and betting strategy under uncertainty from Claude Shannon and John Kelly. When he saw the data Simons had gathered, and the methods he used to measure the market, he saw that an edge existed, but was being squandered.
• The 80’s were the time when the most famous economist in the world (Paul Samuelsson) was attacking the Shannon/Kelly methods publicly. Therefore, nobody from academia or proper finance would contemplate implementing them. But Berlekamp wasn’t from those worlds, and knew from their creators why they worked. So he built a system around the Kelly criterion to properly size the small edges Simons’ team identified.
• Berlekamp may not have fully set up the integrated model, but he ingrained the mindset that it was critical and proved its importance. This is why Laufer, Mercer, Brown, Frey and others were so set on building a single model when others didn’t. They knew that to produce eye-popping returns they needed to use ample leverage. You must understand the correlations of the entire portfolio in order to use ample leverage safety.
• So they hired the best mathematicians in the world to continue finding tiny edges in the markets, and to build and maintain an extremely complex program that continuously monitored thousands of investments for returns, standard deviation and inter-correlation, and automatically calculate the preferred portfolio.
• None of this operation works without the Kelly framework provided by Elwyn Berlekamp. You can see its philosophy everywhere. Interesting, these properties shine most clearly in the strategy aspects that confuse people.

While the Kelly and Shannon framework isn’t necessarily the easiest math in the world, the fundamental ideas don’t require a PhD to understand. The every day investor can grasp the ideas Elwyn Berlekamp used to revolutionize the Medallion fund, and employ them in their own portfolio.

Now, I’m not saying you’re going to start returning 60+% a year (you’re not), but you can use the concepts RenTec employs–and I discuss on this blog–to improve your own investing strategies. The basic concepts are within your reach.

Go explore them.