Forums  > General  > Aggressive HFTs in ES  
     
Page 1 of 2Goto to page: [1], 2 Next
Display using:  

gaj


Total Posts: 37
Joined: Apr 2018
 
Posted: 2019-01-12 19:01
This just shows my ignorance to HFT world outside my niche. But I'm a little surprised by the result of this report: https://www.cftc.gov/sites/default/files/idc/groups/public/@economicanalysis/documents/file/oce_riskandreturn0414.pdf

It says that aggressive HFTs make the most money in ES. They make money against all other groups of participants (including mixed and passive HFTs). They do >15% in ES -- I think it means 30% ADV since they count both sides.

I had always assumed that ES is almost always the first mover in the equity world. So HFTs would look at ES as a signal to hit other products. I can imagine some HFTs do aggressive trades in ES for hedging or reducing inventory, but the article says 84% of their trades are aggressive.

What kind of signals could they be looking at? My best guess is the underlying stocks, with some kind of index arb play. This would be surprising if true because I've always thought that price discovery happens in ES first.

Strange


Total Posts: 1517
Joined: Jun 2004
 
Posted: 2019-01-12 19:13
> What kind of signals could they be looking at? My best guess is the underlying stocks, with some kind of index arb play.

I heard the same thing from several HF people I know. My guess is primarily various order book signals - quote/trade pressures, depth changes etc.

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

anonq


Total Posts: 32
Joined: Aug 2018
 
Posted: 2019-01-12 21:17
one simple example of a signal that would originate in cash equities would be someone coming in and buying a ton of every S&P 500 constituent, if you have the right features to capture that then you have a momentum signal in the ES futures which would warrant taking liquidity, if one wasn't listening to cash equities market data then that would be a signal that one wouldn't see until it's too late, ie the people that saw that signal have already started transacting in the futures

but works both ways, if somone is loading up on ES contracts then that'll not only create a signal in the ES contract itself but also create a signal in the cash equities

those signals would be extracted from order book data

now that's the most obvious but lots of variants of in effect a longer term market participant transacting a lot and causing ripples throughout the market..

gaj


Total Posts: 37
Joined: Apr 2018
 
Posted: 2019-01-13 08:30
Interesting. I had always assumed that large aggressive trades in ES are initiated by long term participants.

>one simple example of a signal that would originate in cash equities would be someone coming in and buying a ton of every S&P 500 constituent

How speed sensitive is this kind of signals? It sounds unlikely that someone would sweep 500 stocks in a couple millis. So you're probably looking at some kind of momentum in O(100ms) timeframe. Sounds like you don't need to be super fast to do this trade, since it's unlikely that someone else pulls the trigger at exactly the same time.

anonq


Total Posts: 32
Joined: Aug 2018
 
Posted: 2019-01-13 19:01
> How speed sensitive is this kind of signals? It sounds unlikely that someone would sweep 500 stocks in a couple millis. So you're probably looking at some kind of momentum in O(100ms) timeframe. Sounds like you don't need to be super fast to do this trade, since it's unlikely that someone else pulls the trigger at exactly the same time.

very speed sensitive, it's a very competitive space and for a majority of these signals if you're not first your last even if the alpha isn't fully realized for 30 min, the majority of the alpha is realized very quickly. I know HRT does this trade and of other smaller groups that are no where near their size that have 100's of colocated servers across every exchange listening to every book feed effectively data mining for signals

and it certainly isn't necessary for a sweep of 500 stocks within millis to create a signal although it does happen, there is lots of information not only in executions but also orders submitted to the market and thus why order book feeds are so valuable




gaj


Total Posts: 37
Joined: Apr 2018
 
Posted: 2019-01-14 17:45
Very informative. This opens a whole new world for me.

ronin


Total Posts: 414
Joined: May 2006
 
Posted: 2019-01-14 18:30
> do >15% in ES
> ...
> 84% of their trades are aggressive

That sounds like single stock market making. They make markets in single stocks, where they make most spread per unit of volatility. They hedge the net by crossing the spread in ES, where they can put the same volatility back for lower spread.

In aggregate, ES ends up accounting for xx% of turnover and it's the single biggest symbol you trade. 15% sounds ballpark ok.

But it's been a while since I did anything even remotely hft, so it all comes with a big grain of salt.

"There is a SIX am?" -- Arthur

EspressoLover


Total Posts: 362
Joined: Jan 2015
 
Posted: 2019-01-14 18:35
Used to play this game in ES (though was never very successful). Like others have mentioned you can still derive a tremendous amount of signal from the evolution of the order book and flow just within a single instrument. In addition, ES does predominately lead, but it’s far short of 100%.

On the CME alone, you can derive a fair bit of signal from the relative basis with NQ, YM, ZN. Even 6E or CL depending on the current macro regime. The thing is ES has a 1 bp spread, plus taker fees are relatively small. So you don’t need a very large magnitude signal to cross the spread. (Edit: Removed dumb example)

You can dive deeper into interactors to isolate periods where price discovery is occurring on other instruments. Very simple example: cash tends to relatively lead the index around the opening auction when participants are digesting overnight events in single-names. More sophisticated is to look at the comparative order book evolution between the pairs.

Effectively what happens a lot of the time, is that there are “trigger events”, where it becomes obvious that the remaining liquidity at the touch is easy money. The typical example would be a bid showing 1000+ contracts then getting smashed down by a giant sell to <10. Grabbing the remaining liquidity is pretty close to a sure shot in terms of profitability.

ES is a pretty thick book. Even if you’re using more sophisticated and subtle signals, they mostly contribute by titrating the threshold to trade trigger events. In the above example you may normally lift the bid if size falls to 25 or under, but go up to 100 contracts if NQ is cheap and order flow’s trending short. Latency becomes not just an issue in terms of getting fills, but also adverse selection. If you’re slow, your fills mostly happen when others don’t want to trade on an event.

Good questions outrank easy answers. -Paul Samuelson

gaj


Total Posts: 37
Joined: Apr 2018
 
Posted: 2019-01-14 19:54
@ronin: That's definitely a possibility but it's counterintuitive. The expected value of a hedge trade is usually negative -- you pay the spread in return for lower volatility.

@EL: Your examples are very enlightening. Thanks.


> Compare to something like SPY with a 4 bp spread, and you can still trade ES a lot even if SPY lags 80% of the time.

Isn't SPY 0.4 bp wide?


> Effectively what happens a lot of the time, is that there are “trigger events”, where it becomes obvious that the remaining liquidity at the touch is easy money.

This is what got me confused about anonq's post. Could an event in a single stock be a trigger event for ES? That sounds very unlikely to me. What's more likely is that a series of events accumulate until the signal is strong enough to fire in ES. But then different people will have different thresholds, so they won't get triggered at exactly the same event. What am I missing?

anonq


Total Posts: 32
Joined: Aug 2018
 
Posted: 2019-01-14 20:39
> This is what got me confused about anonq's post. Could an event in a single stock be a trigger event for ES? That sounds very unlikely to me. What's more likely is that a series of events accumulate until the signal is strong enough to fire in ES. But then different people will have different thresholds, so they won't get triggered at exactly the same event. What am I missing?

unlikely a single stock would be the trigger but rather a series of events across a wide swath of stocks (thus my simple example of someone transacting across all the constituents).. undoubtedly different features/independent variables with different training methodologies and thresholds and risk constraints but there is enough overlap where there is still competition for hitting a quote and thus where speed matters... but to your point if you've found an event that no one else has then latency won't be an issue.. but i'd be surprised, lots of people playing in this space and so lots of overlap in signals

EspressoLover


Total Posts: 362
Joined: Jan 2015
 
Posted: 2019-01-14 21:06
@gaj

Shit, you're right. Bad example...

> Could an event in a single stock be a trigger event for ES? That sounds very unlikely to me. What's more likely is that a series of events accumulate until the signal is strong enough to fire in ES.

Here's kind of my mental model. You have some sort of net alpha, which is the sum of two sub-signals: X and Y. X is a smoothly diffusing stochastic process, whereas Y is discontinuous jump process. Most of the time when net alpha crosses a threshold, it's the direct result of a jump in Y. Not always (sometimes you diffuse through the threshold) but usually. Most participants have basically the same Y signal, but somewhat orthogonal X signals. If Alice is the only one to trade because her private X is stronger than Bob's, it's probably still in response to a public Y jump.

Never really used the cash equity data that much, so take this with a grain of salt. But, like you said, price evolution in the single name equity space is probably closer to a smooth stochastic signal. To the extent that price discovery in single name is independent, then even discontinuous ticks in stocks are going to look pretty smooth aggregated over 500 names.

This type of trading would rarely trigger an ES move by itself. It might adjust traders' threshold of what size jump they respond to. But even if traders are looking at different signals, you're almost always racing someone. Sometimes the size of the jump is so large that the cash basis signal is irrelevant. Everyone wants in. Or maybe not, and a jump only triggers a minority of the HFTs. But even if only 20% of the HFTs jump in, that's still a hell of a race to try to win. Almost always, you gotta fight with latency.

But in terms of what @anonq said, there can be action in the cash space that isn't smooth/independent. Sometimes price discovery across single names can wind up being pretty coordinated. You can get a cascade, where some of the names in the index tick down, then more start ticking down, then quickly almost all of them move. This is the result of equity HFTs sweeping the market once price discovery telegraphs that the entire market's moving down. A coordinated cash move like that could definitely trigger an ES move.

Good questions outrank easy answers. -Paul Samuelson

prikolno


Total Posts: 37
Joined: Jul 2018
 
Posted: 2019-01-14 23:06
The time span of the study overlaps with when 1 person was responsible for about 40% of the trade you're talking about. They would regularly have positions and multiple orders of 10^3+ each. You can move the market favorably in your direction at that participation rate. This trade later dried up. Curiously, said firm was also the 2nd largest in Eurex by ADV at the time so they could've pursued some of the ideas anonq had in mind, but it wasn't necessary.

anonq


Total Posts: 32
Joined: Aug 2018
 
Posted: 2019-01-14 23:31
some hard numbers I've seen are from a group that are generating 90% of the revenue from cash equities with the rest coming from the equity futures market where any signals from equities are combined with signals from the futures market and that pnl is much lower sharpe but uncorrelated.. definitely not trying to make it seem like this is some great trade but rather per one of the the original questions that there are signals that can come from elsewhere that can warrant taking liquidity in ES

Strange


Total Posts: 1517
Joined: Jun 2004
 
Posted: 2019-01-15 05:54
On a slightly different topic, since we are talking about ES order book. I had this wacky idea that balance of the resting orders (away from the touch) are a potential indicator of the gap risk in the market. The logic was that most of the people resting orders are smart HF guys who are weighing queue priority against the possible risk of getting taken out on a large gap.

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

ronin


Total Posts: 414
Joined: May 2006
 
Posted: 2019-01-15 13:30
> @ronin: That's definitely a possibility but it's counterintuitive. The expected value of a hedge trade is usually negative -- you pay the spread in return for lower volatility.


@gaj, sorry - I didn't actually read the article, I just commented on your highlights. I don't off-hand know of any HFT strategy that makes Sharpe 4 just by crossing the spread in ES.

ES is the cheapest hedge, and you cross the spread when hit rate is more important than price. So I assume this is some hedging signature, but I have no idea.

"There is a SIX am?" -- Arthur

Strange


Total Posts: 1517
Joined: Jun 2004
 
Posted: 2019-01-17 05:59
> I don't off-hand know of any HFT strategy that makes Sharpe 4 just by crossing the spread in ES.
But if you do recall it, do you mind posting the details here? Preferably with the source code

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

ronin


Total Posts: 414
Joined: May 2006
 
Posted: 2019-01-17 13:54
> But if you do recall it, do you mind posting the details here? Preferably with the source code

Oh man - you think that might get me an interview with Strange Asset Management? I'm on it.

That's actually a good name, Strange Asset Management. Should have thought of it earlier...


"There is a SIX am?" -- Arthur

tbretagn


Total Posts: 263
Joined: Oct 2004
 
Posted: 2019-01-17 13:56
nowadays you need to add a "Point" to the name (point72, exodus point etc).
But "Strange Point AM" sounds cool

Et meme si ce n'est pas vrai, il faut croire en l'histoire ancienne

Strange


Total Posts: 1517
Joined: Jun 2004
 
Posted: 2019-01-17 16:49
Sure, any imaginary alpha can get you an interview with Strange AM!


I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

ronin


Total Posts: 414
Joined: May 2006
 
Posted: 2019-01-17 20:46
> imaginary alpha

Meh. At least it would be orthogonal.

"Where do you work?"
"Strange hedge fund."
"Which one? They are all strange."
"Strange with capital S."

The story just writes itself...

"There is a SIX am?" -- Arthur

Strange


Total Posts: 1517
Joined: Jun 2004
 
Posted: 2019-01-17 22:57
Eventually, we can spin-off the Weird and Downright Crazy groups!

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

ronin


Total Posts: 414
Joined: May 2006
 
Posted: 2019-01-18 15:23
Adventures of Strange Point Asset Management
Day 27

"Boss, we have investors on the phone. They are complaining that our alpha is imaginary."

"Just tell them to rotate their laptops by pi/2 and try again."

"There is a SIX am?" -- Arthur

Strange


Total Posts: 1517
Joined: Jun 2004
 
Posted: 2019-01-19 18:04
Strange capital would not have that issue - zero alpha is both real and imaginary.

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'

ronin


Total Posts: 414
Joined: May 2006
 
Posted: 2019-01-20 22:54
Aggressive.

Right now, 90% of the street would kill their grandmother for zero alpha.

"There is a SIX am?" -- Arthur

Strange


Total Posts: 1517
Joined: Jun 2004
 
Posted: 2019-01-21 03:53
flat is the new up, baby!

Back to the topic, now that I have looked at the paper and have some doubts.

First, here is a quote "the average Aggressive HFTs earns an annualized alpha of 90.67%, while the average Passive firm earns 23.22%". Doesn't that simply imply that so negatively selected passive providers make lower return on capital, but probably have better sharpe ratios due to higher number of trades. A liquidity provider can capture the spread and get negatively selected once in a while. An informed liquidity taker can make a lot when he's able to catch a large move (e.g. on the back of events, news etc), but he's only able to do that relatively infrequently so they trade size.

Also, he actually tries to back out profits and that's where the numbers confuse me: "trading revenues of over $18.6, $7.0, and $3.2 million for Aggressive, Mixed, and PassiveHFTs". Maybe I am missing something, but can 3.2 million even support a non-amateur MM operation in something like ES?

I don't interest myself in 'why?'. I think more often in terms of 'when?'...sometimes 'where?'. And always how much?'
Previous Thread :: Next Thread 
Page 1 of 2Goto to page: [1], 2 Next