[b0b10101]

> XTX -- the name refers to a mathematical formula used in its trading algorithms

Wow so they do linear regressions?? Incredible.

But in all seriousness, for how long are we going to dress up statistics as some new cool and exciting methods? Are VC's and bosses really this easily fooled?

[abbadadda]

Love this comment! Still, XTX is a cool name, even if the author of the article didn't really get it: https://math.stackexchange.com/questions/2624986/the-meaning... . The nice thing is no VCs are really necessary to fund trading firms like this. Usually the owners have 100% of the capital, split between one or more owners. Not positive of XTX's capital structure, but it is unusual to have VCs involved. GETCO was an exception, and they were pressed to eventually go public (as KCG): https://finance.yahoo.com/quote/KCG/

[zelly]

What they actually do: colocated servers next to the exchange collecting $0.0000001 per trade with nanosecond latency.

Create mystique of doing spooky AI to get more capital which they use to hedge even more trades/second, making more money, which they then attribute to AI which gets them even more capital.

[vecter]

If you knew anything about HFT, you'd know that speed got arbitraged out more than a decade ago. Everyone is fast now. In order to make money, you need true alpha and predictive ability. It's sad to see you disparage their impressive achievements when what they've done is incredibly difficult to accomplish, especially as such a latecomer. I know because I tried it myself ten years ago when it was even less competitive and couldn't make it work.

[PopeDotNinja]

The real money is probably made by you renting out the server racks close to the exchanges!

[kasey_junk]

Lots of exchanges now capture that themselves.

[rrauenza]

When you say “speed got arbitraged out” — what does that mean?

[smabie]

It means that it’s no longer profitable for the vast majority of firms that used to do latency arb. Most of those firms have either gone out of business or managed to discover less obvious signals

[continuations]

How does XTX differ from a firm like Renaissance Tech?

[qeternity]

Impossible to really say without splitting RenTech into its different funds (you're probably referring to Medallion) and knowing what the underlying strats are. But at a high level the difference between MMs and quant funds are AUM which means different time horizons. An MM might have a time horizon measured in mics to seconds, and might be focused on micro market analysis like the balance of order book over the next 1,000 microseconds. RenTech (likely) can't deploy all of their capital in these types of strategies so they focus on strats with better liquidity and almost certainly longer time horizons (minutes to hours and days, possibly weeks for more CTA style strats). And at the opposite end of the spectrum, Berkshire can deploy tens of billions in a single trade, but that means an investment horizon of years to decades.

[nicoburns]

> you need true alpha and predictive ability

Surely all high-frequency signals are "follow what the market does, but slightly faster", and thus provide no actual value to society. It might be technically impressive, but it's useless work for personal gain. I reserve the right to be disparaging.

[throwawaymath]

That's part of it, but HFT is also a pretty saturated field. It's typically not enough to be fast and physically near the exchange these days; that's a necessary but insufficient condition. A lot of nontrivial math and CS is still poured into HFT at places like Two Sigma, Jump and Hudson River Trading.

[mjfl]

I like that it's become a common understanding that AI research is most effective in the marketing department.

[anonu]

So much negativity in this thread.

[Abishek_Muthian]

In India, the exchange was bribed to have the servers within the exchange for algo trading by some high profile institutional traders!

[1]https://wikipedia.org/wiki/NSE_co-location_scam

[zorked]

Is this fraud in India? I think most stock markets sell this as a service.

[Abishek_Muthian]

At that point SEBI (SEC equivalent) didn't permit exchanges to do it and were in the process of forming co-location guidelines, but this exchange went ahead and gave preferential treatment to few traders.

[smabie]

You can’t make less than a cent per trade in most markets. The bid-ask spread can never be less than a cent. I guess you could make less than cent with commissions, but you certainly can’t make $0.0000001.

[notavalleyman]

This is a result of the common Maker/Taker pricing model. [1]

Those who "make" liquidity by publicly quoting ask/bids are rebated fractions of a cent when their orders are filled, and those who "take" liquidity by exercising the Maker's position are charged.

This is separate to the spreads. The book Flash Boys has a very good explanation of the model.

[1] https://www.investopedia.com/articles/active-trading/042414/...

[cheez]

* Number of trades = 100000

* Number of successful trades = 1

* Number of unsuccessful trades = 999,999

* Profit on successful trade = 1000000

* Loss on each unsuccessful trade = -1

* Net profit: $1

* Net profit/trade: $0.00001

[smabie]

Well if we’re averaging, sure. But that’s not what the parent is talking about. Moreover, what you’re describing isn’t a HFT strategy, it’s an investment strategy.

[cheez]

No, that is how you think about high frequency trades. Lots of trades in a small amount of time, with a tiny positive expectancy.

[smabie]

https://www.reuters.com/article/2014/11/13/us-markets-virtu-...

So Virtu profits on 51-52% of its trades. The type of math you are describing does not make any sense. Unless Virtu is significantly different from every other HFT firm, your idea of high asymmetry of profit and losses is not true.

[skunkworker]

Time to make your own microwave network

[deepnotderp]

Some (not all!) of the techniques at places like RenTec are fairly sophisticated. Hence their returns ;)

[hogFeast]

Nick Patterson has said they never used anything more complicated than linear regression whilst he was there. The trick was how to use it, not the tool itself...I think people who read a lot of textbooks get obsessed with finding the next complex "secret silver bullet" technique. Good analytical work with basic techniques is often more worthwhile.

[superpermutat0r]

Linear regression can be the reduction form of many sophisticated methods. For example, HMM or CRF like inference can be easily done with linear regression.

Hierarchical models with several layers of symbolic inference can be done with linear regression.

Reinforcement learning too.

You can reduce almost anything to a binary classifier and implement it in practice to work well. The reduction is tricky but performance is not.

[hogFeast]

As I explained, he said it was none of this.

[superpermutat0r]

I do not see where you explained it. Doing cost sensitive linear regression (which is pretty trivial to implement) allows you to approximate Markov models, any hierarchical model, and all sorts of other stuff (like minimizing different cost functions, quantile regression etc.) All achievable with the same linear regression algorithm and additional data modifications.

Maybe sometimes you need to change the weight update rule, but that's it.

[hogFeast]

It is funny to watch you struggle with this (maybe this is why you don't see). He specifically said that wasn't what they were doing (and said it was often simple linear regression with one or two variables), they used no complicated techniques (he even pointed out that people assume that must be true...but it isn't), and that what they did was a combination of good analytical work/hiring. Btw, this is also quite obvious from reading Zuckerman's book.

I understand why almost no-one gets it. And that is why firms like RenTech are able to print money. I actually do work like this in a similar field, and everyone assumes you have some kind of secret algo that is the product of some unpublished, very complex work (these days, usually deep/reinforcement learning). But the reality is doing the simple stuff well and using that to build a deep understanding of the data. I suppose that is less fun for researchers who want to publish flashy stuff about deep learning and write lots of equations on whiteboards...but I prefer the money.

[anonu]

Exactly. In my experience in prop trading, simple things win. Convoluted strategies probably don't...

Also, the winners usually have a structural edge that others don't. In the past this used to be low latency infra but now everyone has it. Usually edge means unique flow or market access that isn't common.

[hogFeast]

Just my 2c but I think RenTech's edge is hiring.

I think people assume that because you hire all these brilliant people that their domain knowledge is the reason you are hiring them (which usually plays into most people's understanding that what you need is the secret knowledge, and once you have your piece of paper then life is over, you have won).

An alternate hypothesis is that RenTech looks for people who have good domain knowledge AND have proven analytical ability translating that knowledge into practical results. Example: Nick Patterson is clearly a genius, he clearly has immense knowledge about stats/probability but he also seems like a very practical guy who has built a reputation on getting things done. It seems like RenTech hire smart people but also people who get things done.

I worked in fundamental equity research. Latency or tech is definitely an advantage but hiring/training processes can be very powerful too (it is often impossible for some firms to replicate at all). I have no idea if it is true here but given the focus RenTech place on hiring, it seems logical that is where their edge is.

[BubRoss]

Like what?

[larnmar]

Nobody really knows, except for a few people who aren’t telling. There’s some clues in the mathematical backgrounds of the kind of people they hire though. Fifteen years ago they were snapping up experts in stochastic calculus. Nowadays, I’m not sure whom they’re hiring.

[throwawaymath]

They're not exactly forthcoming, but it's not the hyperbolic mystery the mythos would have you believe. You can piece together what matters to them - in the abstract - by searching MathSciNet for papers published by their researchers prior to joining. For example, it's pretty easy to see they hired an information theorist away from Princeton in 2017.

But obviously that's not enough to reproduce everything they do, or else they wouldn't still be legendary. As it happens a lot of what makes them successful is not the sophistication of their trading algorithms and research, but also the sophistication of the execution and reliability.

As an aside, from friends there Renaissance hires researchers in three primary ways:

1. They have a small network of professors who they solicit for promising new PhDs willing to leave academia each year.

2. They watch professors and postdocs in specific specializations, and reach out to those whose research meaningfully interacts with a thesis they're interested in internally.

3. They send small groups to conferences to poach people working elsewhere in industry (particularly tech) whose work is applicable to their own.

They also do hire people who directly apply of course, but most hires are reactive. They especially like to hire people whose work or research looks like it might begin to encroach on their own, or is just notable and impressive. The math is certainly important to them, but that's just one dimension of it.

[deepnotderp]

Would you say that their edge in algorithms comes mostly from signal processing or from ML? I'm guessing it's mostly signal processing and signal extraction?

[chatmasta]

I would be surprised if rentech's profits are based purely on algorithmic advantages. More likely, IMO, is they've curated some really good sources of alternative data and combined it with otherwise standard mathematical and technical trading techniques. They’ve probably also got a well-engineered pipeline for identifying that data, systematically evaluating it, and bringing it to production.

This is of course pure speculation. But I doubt all that alpha would come from mathematical differences alone. The market isn’t magic... it’s just a question of having access to the right information and the ability to capitalize on it.

[deepnotderp]

Two Sigma has an entire team devoted to alternative data collection, with much worse results. Alternative data isn't as useful as you think- short term truth is only one variable in market prices.

[chatmasta]

I’m not saying they have access to a single magical data source. Rather, they’re very good at identifying when a source of data is valuable to them, and milking it until it runs dry.

[msla]

> Some (not all!) of the techniques at places like RenTec are fairly sophisticated. Hence their returns ;)

Non sequitur. A chimp throwing darts can beat professional brokers.

https://www.forbes.com/sites/rickferri/2012/12/20/any-monkey...

Maybe their sophistication is a really good PRNG? Or just one that smells of bananas?

Or maybe they've lucked into a model which happens to be doing well now, but won't after the market shifts a bit. Like how the random pickers above did well because random picks include a lot of small companies, and small companies did well over the period being considered in the study.

[deepnotderp]

Go calculate the p-value of making 66% before fees for 30+ years.

[roenxi]

Which still leaves a non-sequiter in saying the sophisticated methods led to the high returns. It is more likely to be some form of 'cheating' where they are actually making the money off something that isn't normal trading. Or even corruption in the worst case.

Beating the market by very large margins is a suspicious activity. If raw mathematical insight is enough to get a 66% return then there are a lot of mathematicians out thereto figure out what is happening. If that is the secret sauce it would be extraordinary in a history littered with best performers who were just flat out frauds.