Renaissance Technologies

Interesting article on Renaissance Technologies: http://finance.yahoo.com/news/inside-moneymaking-machine-no-other-050104225.html

Some insight into their philosophy:

“By studying cloud cover data, they found a correlation between sunny days and rising markets from New York to Tokyo. “It turns out that when it’s cloudy in Paris, the French market is less likely to go up than when it’s sunny in Paris,” he said.”

I saw that too. I even had a post with a link but couldn’t figure out what point I wanted to make.

I guess if I did try to make a point it would be that they have 300 employees with 90 Ph.D.s. Anyone at P123 claiming that they have all of the answers should be a little humbled. Probably to the point that they should not say they know what doesn’t work even if they have found something that does work.

Jim

I get the feeling the weather indicator was given to throw people off the scent. 300 employees and 90 PhDs may sound impressive but if you think about it there is probably comparable if not greater brainpower using this site.

[quote]
I get the feeling the weather indicator was given to throw people off the scent. 300 employees and 90 PhDs may sound impressive but if you think about it there is probably comparable if not greater brainpower using this site.
[/quote]Sterling, maybe, but they have far more data to work with. I watched an interview with Simons and he said that they collect data by the gigabyte like it’s going out of style.

All:

   My memory is that they collect roughly a terabyte/day of data.  That may sound like a lot but most of it is probably tick data. Simons, and the people around him are smart, no doubt about it, but that doesn't mean I can't make significant money from the markets.

Bill

Bill,

Just want to say that I’m absolutely convince that you can make money in the market. As others have said: P123 rocks!

One of the things I like about the article is that it is documented public proof that it can be done. No one has to believe an anecdotal story from one of our users. It absolutely can be done. And we have some advantages at P123—being able to invest in smaller-cap stocks and fewer problems with moving the market.

Perhaps, it is amazing that the Medallion Fund can do it with that much money. They must be limited to larger-cap stocks. Presumably they invest in Paris (if they are keeping track of the weather there). Maybe this is to diversify but this may also be to manage their liquidity problems.

One could speculate forever about which scales up more quickly: the benefit of additional pieces of less and less significant data or the problems with liquidity. But I’m guessing they released the data about the weather in Paris because it isn’t useful it a ranking system (or multivariate regression).

Your post comparing the users of P123 to the employees of the Medallion Fund gave me a nightmare. Last night I woke up in cold-sweat having dreamed that all of us at P123 had to pool our money and resources and agree on our models and investments. What a nightmare! Assuming that the Medallion Fund works this way—that alone is an accomplishment. But it is probably better to be at P123 with complete control of our investments.

Jim

One difference is they’re likely working in a more collective, collaborative and open knowledge sharing environment working for the overall performance of the fund. So on that note, if anyone would like to be more like Renaissance and set your most productive Ports public, feel free to do so :wink:

One of the reasons I linked the article: it reminded me of when, some years ago, I worked for a small quant advisor. I remember vividly the day I spouted off something about the importance of cash flow measures versus reported earnings, and the absolute glare of disdain the CIO gave me - “You think you’re the first person who thought of that?” he said. A little later in the meeting, he said, “If you find a correlation between the price of potatoes and the stock market, you’d better pay close attention to the price of potatoes.” I thought he was joking, but he was absolutely serious. To him, and therefore the philosophy of the firm, all data we collected was just data. We threw math at it and traded what stuck. Incidentally, I was the least smart person there. To this day, I still don’t know why they hired me.

The Ph.D’s they listed are astrophysicists and language specialists, not economists. They are used to looking at data and then drawing conclusions. This isn’t to say that this is the only way to do things, it’s just the philosophy of their firm.

My guess would be that they make the majority of their money off high-frequency trading. It’s the only way their draw down could be so low, if was the article says is true.

I’ll consider believing you on this. But first I will have to know if the quant advisor who hired you is still in business. And I suppose it would be helpful knowing how the price of potatoes is doing.

It takes a true astrophysicist to know the level of significance necessary to use price indicators like this as a signal. Fortunately for us, most astrophysics are well……astrophysicists.

At least you had the sense to sign up with P123—instead of say ProfitsPi (almost purely technical and were not PIT) where I used to be.

But I do wish D.E. Shaw would go back to Columbia University. Actually, D.E. Shaw and Co. is now managed by a 6 member executive committee which once included Lawrence Summers. But you get my point—it really can be done by some.

Jim

I worked there around 17 years ago. The firm isn’t in business anymore, though the guys I keep in touch with still are. The CIO has long since retired. I should also say we did use Vestek to run fundamental based stock selection strategies. I use P123 and Tradestation.

Renaissance is pretty secretive about their systems, but there are some clues:

  1. They have a patent filed for an HFT system.
    2a. They have a bunch of people who specialize in writing algorithms that translate plain English into data; sort of like IBM’s Watson.
    2b. I once read an article that detailed how a market maker in options was clobbered when someone bought tons of options on a biotech, microseconds after the news came out that they had gotten FDA approval for a drug. If anyone is doing that sort of thing then it’s probably Renaissance.
  2. There is actually a simple, highly profitable, fundamentals-based system that I use that specializes in micro-caps with a relatively low turnover rate. Often when I check out the list of stockholders on record I find Renaissance on the list.

Don’t worry about competition from Renaissance, how many of us are making 30%+ with a system that can handle institutional money?

That is 30%+ after fees of 40%+ as far as I can remember.
They have made something like 70% before fees for years with their medallion fund, which has about 5 billion under management and is mainly held by Reneissance employees.

The performance of their Renaissance Institutional Equities Fund RIEF are doing ok and they did not manage to get their targeted 100 billion.
At the moment it has 29 billion AUM.

Their Renaissance Institutional Futures Fund has been closed.
http://www.reuters.com/article/us-hedgefunds-renaissance-idUSKCN0S72SG20151013

Bringing up this thread to note that there is a new book about Simons/RT called The Man Who Solved Markets by WSJ writer Gregory Zuckerman. He interviewed over 30 current and former RenTech employees. I’m about 1/3rd the way through at the moment, and it’s a good read. I can’t say it gives away much quantitative “secret sauce”, but it’s good storytelling with an interesting and eclectic group of characters. I’m around the 1995 point in time in the book, when they were really just running highly sophisticated technical analysis pattern finding/trend following on non-equity assets with much better datasets than everyone else (the early RT researchers were consumed with hoarding granular ticker intraday data even back in the early 1980s). I’m getting to the part where Mercer and co. start desigining their models for equities. These were Gods among men when it came to mathematics, but still have the same foibles all of us have to fight in regards to things like trusting your model, not overriding or abandoning things based on some headline or newsstory you just heard and avoiding “go with your gut” trading. I’ll give some highlights when I finish it up.

They are algorthmic, (like us) not necessarily HFT

They are algorithmic. Not HFT, but definitely high turnover. I haven’t got to the part in the book where they start developing their equity models yet, but I heard the author elsewhere talking about the book and he said they typically hold a position for 2 days or so. I don’t think much of what they do is driven by fundamental factors, in fact they almost go out of their way to hire people who don’t have backgrounds in business. Everyone they interview has to pass a coding test and are asked to solve a probability problem. I think it’s more price prediction based on Markov modeling. That’s good … I don’t want to be playing the same game as them. Simons has a quote in the book saying “I don’t know why planets orbit the sun, but I can tell where they’re going.” At least in the earlier days that they cover in the book, they didn’t really care much about why, if a pattern had a passable p-factor it got tossed into the model.

I met Bob Mercer on an interview at IBM Watson way back in the early 80’s- he was a member of their speech recognition research group. They were using HMM’s (Hidden Markov Models) for phonetic modeling, and their “stack decoder” for language modeling. So it is probable that he applied his knowledge gained there to trading algorithms. (No, I didn’t take a job there, I ended up at Bell Laboratories doing the same stuff though, speech recognition R&D using HMM’s, Neural Networks, Context Free Grammars, etc.)

sglinski, that’s really cool. Yeah, Simons first business partner was Lenny Baum, who co-invented the Baum-Welch algorithm that hidden markov modeling was built on. That was part of the pitch Simons used to poach both Mercer and Peter Brown from the IBM speech recognition team (along with $$$). When Mercer and Brown got to Ren Tech they immediately recognized that they were going to have to build up their computational muscle, and recommended that they hire their former IBM teammate computer scientist David Magerman. So those markov modeling ties are pretty obvioius.

Cary,

Thank you for making me aware of the book.

Kernel regression is mentioned multiple times in the book. This is trivially implemented in R (with whatever data you may posses). I have not tried to use kernel regression after switching to Python but I am sure some type of kernel regression can be done in Python.

For anyone interested in exploring Simons’ methods in R, I would recommend using LOESS (local regression). It worked well for what I was using it for. There are other Kernel Regression methods that can be found in R. One can also find programs that use “Splines” which will accomplish the same thing.

According to the book:

“The firm began incorporating higher dimensional kernel regression approaches, which seemed to work best for trending models, or those predicting how long certain investments would keep moving in a trend.”

Most of the programs in R and Python are “higher dimensional” and this is used here to add drama and make it seem difficult, as I am sure you know.

I also find this interesting:

“…as long as they had p-values, or probability values, under 0.01—meaning they appeared statistically significant, with a low probability of being statistical mirages—they were added to the system.”

Yep. The dreaded statistics. Sadly, it has been well established in the forum that this only works for RT. The laws of science and mathematics are just different for multi-billion dollar firms. Either that or we, at P123, just are not smart enough to use this—it takes a genius like Simons to get a p-value. In any case, you should forget about the idea of using this in the small- and micro-cap space where RT might not be playing, we are told.

The above is demonstrably false with what Simons has done being adequate proof. Unless, of course, any edge we could have is already lost because of multiple institutions already using similar methods.

Note that–as established above in this thread–Simon’s and others at RT had no “Domain Knowledge” in Finance. However, using established Financial principles may have potential, Cary suggests.

Cary is spot-on with this and this has great potential, I think. Cary notes that the book gives no evidence that, RT at least, would be competing with regard to Financial Fundamental factors as predictors. I believe this is an idea we are already exploring at P123 with our sims, rank performance optimization and spreadsheets.

I strongly agree with Marc (and Cary I think) that “domain knowledge” in Finance will benefit—especially here on the P123 platform. But there is no reason we cannot use domain knowledge in Finance and learn a lesson (or two) from Simons at the same time, I think.

-Jim

A staff member of RT (Laufer) was asked where their profits were coming from. His answer:

“Its a lot of dentist.” I take this to mean retail investors.

Midway though the book: “Medallion trades about eight thousand stocks.”

Speaks to how much of an edge we might have, I think.

Oh, and at one point their Sharpe Ratio was 6 (later in the book 7.5) over a period of several years.

-Jim

Jim,

Maybe you can affirm my reading of it. RT is known for their legendary non disclosure and non compete agreements, so they weren’t get too much into the details of what their fund does. It was more an oral history of Simons and the firm, and through that you could attempt to deduce what they do. My read was that the Medallion Fund has very little to do with business or fundamentals. It’s almost entirely driven on price action based on pattern correlations. The model is entirely self perpetuating through machine learning. They noted that when Mercer and Brown took over the equity model from Robert Frey the code lines exploded from a few thousands to half a million. This was in the late 90s, God knows what it is now. Medallian is also very sensitive to size. They place a mind bogglig amount of orders a day, but most of it is the same position broken up many times to minimize slippage … as they’re absolutely maniacal about limiting slippage. It seems like slippage was one of the reasons why Frey couldn’t get his models to perform as well in real world trading as he could in his simulations. This research was genrally established by astromers and theoretical physicists and etc. etc.

Simons then set up larger fund to handle more money, the Institutional Equities Fund, that can handle up to 100 billion and take on large clients like pension funds but also doesn’t have near the returns that Medallion does. It beats the it’s benchmark regularly now, but it struggled out of the gate. It seems that model does use things like fundamentals and factors. In fact, there are job postings up on RTs website right now for a Computer Programmer, and aside from technical qualifications, it says accounting experience is a plus. But maybe this is their own internal accounting software.