Apart from the great pay you get to invest in RenTech funds. I might be wrong but I think at this point theyre just working to manage their own money and not taking any money from investors. Above all else - working with smart people is its own reward.

> Above all else - working with smart people is its own reward

Of course, there's also the reward of building, creating, and contributing to something that extends beyond yourself. It's hard to believe that a career at a secretive hedge fund that invests its own funds would provide much along that axis of career satisfaction.

Some individuals are motivated by their interaction with smart colleagues, but others by the process of the work itself, others by financial incentive, others by prestige, others by societal impact, etc. There is no one-size-fits-all metric that guarantees satisfaction.

I think all employees can invest in the Medallion fund, which has a 50%/year return.

Would anyone speculate about what mathematics and methods they use to achieve a 50%/yr return? I wonder if there's a way to reverse engineer their model by e.g. looking up which research papers their employees published prior to joining RenTech.

If anyone knew they wouldn't certainly wouldn't share it. No one seemingly knows because no other fund has a comparable return, which is surprising because typically ideas diffuse as employees go work for other firms.

I suspect they use the ability to continue to invest in Medallion as an incentive to prevent people from going to work elsewhere.

It should be possible to figure out their methods. Besides, it'll be a lot of fun.

Is there a data dump somewhere of Medallion's trades over time? Since the stock market is public, that means we could treat their investment engine as a black box, and try to reverse engineer their engine by looking at the decisions it makes over time. We should pay careful attention to how it performed leading up to the 2008 crash, since that reveals how their engine performs in adverse conditions, which might indicate what underlying metrics their engine is using.

I don't know why I can't reply to the below comment.

The volume traded on any exchange is absolutely massive, especially on the products where funds might be putting in millions of dollars. There's also an enormous number of prop firms, banks, institutions, hedge funds, mutual funds, and whoever else constantly buying an selling. It would be near impossible to be able to say "Firm X traded at these specific times."

Furthermore, any worthwhile firm moving that much money knows to break up its trades. If they immediately dump $10 million into the market, its going to cause a large market impact, and they're going to leak information they probably don't want to give away (since that's their edge). (See the whole Flash Boys nonsense.)

Lastly, if you could identify a set of trades that were owned by a single firm, it would be a massive benefit to anybody opposing them (ie: everybody else).

My own guess of their method is that it's built upon a philosophy of very short lifetimes for any given strategy. I'd guess they've optimized their ability to track the life cycle of a strategy, and perhaps generalized to different classes of approaches for generating new strategies. They probably have rules of thumb they follow and a culture and that promotes the group's ability to find new signals and temporary inefficiencies.

Hence they'd be just about as non-stationary as the market itself. If you were dead set on doing an almost impossible finance project, then instead of reverse engineering you'd probably be better served by figuring out how to do life cycle management of signals yourself.

Any firm's trades are absolutely not public information.

Given a dump of all stock market trades over a month, I wonder if it's possible to algorithmically group those trades into cohorts to identify the source of the trades. In other words, Medallion fund presumably manages a large amount of money; therefore, when they move that money into different investments (on a daily basis?) then we should see large trades (or many small trades that add up to a large amount).

I wonder what the daily trade volume of the NYSE is? It seems like if someone is moving around a billion dollars via thousands of trades, it should be possible to detect and categorize those seemingly-disparate trades as "these trades originated from the same owner".

If it's possible to trace the flow of money in that way, then it may be possible to generate a list of probable Medallion trades: search for a cohort that achieved >40% return on billions in volume.

Another way of figuring out Medallion's methods is to examine its interaction with academia. Presumably they search for candidates from fields related to their trade engine. If we can generate a list of research papers which resulted in its authors being hired by Medallion, then those research papers may reveal some of the techniques used by their trade engine.

I used to work for a financial software startup. One of our products was all about analyzing the NYSE TAQ (Trade-And-Quote) data to find such patterns. Another of our products was an execution engine that would take a large block order and automatically slice it up into individual trade executions so that nobody could a.) identify the source of the large order and b.) trade against it. The software would automatically compensate for the amount of liquidity in the market at a given time and execute trades only when there was sufficient order depth to avoid moving the market.

Basically, people have thought of these strategies before. And they've thought of countermeasures to those strategies. And they probably have countermeasures to the countermeasures - for all I know, maybe RennTech's secret is that they've figured out how to statistically identify everyone in the market and trade ahead of them.

The financial industry is a massive arms race which draws some of the smartest people in the world. It's very unlikely that you (or any individual investor) can compete unless you've worked at one of the big hedge funds.

The reverse engineering is near impossible. My guess is that their execution algorithms will be as complex as their strategies themselves, so finding out their trades and figuring out the underlying strategies (which constantly evolve with the market) is really hard. Moreover, if one had so much quant ability, they'd be better off making some money for themselves and beating RenTec at their own game.

Manifold learning and generalized SVM is one component.

Ty. Now pls tell me what are the features for the SVM so I can replicate, thanx.

I will speculate. They mine asset returns for predictable statistical patterns without making prior assumptions about what these patterns are. Once they find the patterns and confirm that overfitting is not responsible for generating them, they trade on these patterns.

If it really had that high a return, wouldn't they basically own the economy by now?

They distribute back all profits and keep the total capitalization of the fund constant. Yes, it really has returned that amount for 20+ years.

Doesn't this sound like Madoff and Picower?