Forums  > Trading  > Monetizing order flow in futures  
     
Page 1 of 1
Display using:  

EspressoLover


Total Posts: 225
Joined: Jan 2015
 
Posted: 2017-05-25 16:51
The topic is whether captive order flow has any value in a space like futures. I.e. a market you can't internalize and have to trade on a single centralized exchange. My knee jerk reaction: of course not! Payment for order flow only works, because you can preferentially cross incoming flow in your dark pool before routing to exchanges. The value comes from the priority your liquidity gets against uninformed flow. Even in a space like options, without dark trading but with multiple exchanges, you can kind of hack this by routing to the venue you're quoting at.

But futures, this doesn't work. Everyone's in the same order book. Even if you're sourcing the order flow, your liquidity doesn't enjoy any special privilege. There's no leg up. The fact that there's no major payment for order flow operations (AFAIK) in futures space seems to confirm this world view.

However, I don't know if this is exactly true. I'm not sure if it's legal or practical, but I do think you can monetize captive uninformed flow, even if you have to trade at a centralized venue. Let's say some instrument is quoting 10.50x10.53, and a retail market sell order comes in. What's to stop the broker from submitting a bid at 10.51 immediately before? It still captures $0.005 in spread against the mid.

At first glance this seems like it should be prohibited as front-running. I don't know the letter of the law, however it's not against the spirit of the law. Front-running involves looking at a client order and trading ahead of it in the same direction. That's bad because it worsens the quality of the liquidity the client accesses (worse fill prices, smaller size, and larger market impact). This has the exact opposite impact. The client is getting better execution. In the example she sells at the higher price of 10.51 instead of 10.50. The logic isn't any different from internalization, except the cross occurs inside the venue instead of outside.

Even if it's theoretically possible, there's the question of whether it actually would generate any real money. Exchange fees still apply. You have to improve by at least a tick (unlike the minuscule improvements done by equity internalizers). Also most major futures trade at tick-tight spreads, where that kind of price improvement isn't possible. Still, even if you can only join, there's some value in knowing the queue is about to get smaller due to uninformed flow. Probably not a huge amount given retail order sizes relative to queue sizes, so maybe it's really not worth it.

But in instruments with non-FIFO matching, there may be bigger advantages. Particularly pro-rata or size priority. For example you can flash a very large quote for the tiny interval you know the client order will be arriving. By dominating the quote size, you'll match against almost the entire incoming order. If you queue up [Quote]->[Client Order]->[Cancel] on the NIC, another ATS won't have time to process and lift your quote.

Anyone ever hear of something like this before?

dVega/dRho


Total Posts: 5
Joined: Oct 2015
 
Posted: 2017-05-25 20:58
Captive flow means signed flow. Whether you try and meet the execution of that flow via the exchange or not there is potentially value in the flow as you are no longer observing anonymous orders reported by the exchange. If the probability of the flow from a given order channel (customer) being informed is non zero then you have an advantage compared to everyone else. At the limit if all the flow going into the exchange was via your channel then you alone would have the best insight to the relationship between the orders of a given participant and the subsequent price action. To everyone else all orders other than their own are anonymous. There has to be value in that.

Now, what you do with that? You made some good suggestions based on no insight into particular customers. In all likelihood it would be hard to profile individual customers but you may be able to assign them to a informed / uninformed group.

HitmanH


Total Posts: 424
Joined: Apr 2005
 
Posted: 2017-05-25 23:52
Very interesting post EL - Thank you for this - though provoking.

First issue - and I'm not sure if this is really persuasive - is that a lot of exchanges (particularly those which are less liquid) permit OTC trading of exchange traded products - although (depends on where) have to report to venue for trade reporting and clearing within a certain window - so (should that flow be large enough) - you can act, and the market acts delayed post the print to clearing (this happens in equities re: some dark trading too in EMEA).

The example you give is interesting - and niche. One can't always get between the bid and the ask - so is your example only interesting in those situations (as you suggest the the following para).

Typically - I'd imagine more retail flow isn't indicative of something else - and not in size - so the main thing I can think of is if you can categorise dumb and smart flow, and not hit the order book - or - as I said more the less liquid contracts.

(not heard anything specific like that)

ronin


Total Posts: 206
Joined: May 2006
 
Posted: 2017-05-26 10:47
@el,

I wonder if your example really works.

Think about 4 scenarios:
(i) you open a level at 10.51, you get filled by your internal flow, market is back to 10.50
(ii) you open a level at 10.51, you get filled by some random flow, market is back to 10.50
(iii) you open a level at 10.51, you don't get filled, you cancel, market is back to 10.50.
(iv) you open a level at 10.51, a queue forms at 10.51, you have priority at 10.51.

I don't really see the difference between (i) and (ii), and they are both losing you one cent. Only (iv) makes you money, but you are expressly excluding (iv).

Pro rata I see it, but the exchange might not like it. Do it a few times, and they will create a rule especially for you.

If you are happy to improve the market, why would it matter whether that benefit goes to your captive flow or to random flow? It only works as marketing ("guaranteed to beat the market by 1 cent") - if you charge more than 1 cent in comissions, it is worth it for capturing extra flow. But if you don't get paid for client flow (or even worse, if you are paying for it), I don't see it.

There is some value in using the captive order to probe the market without risking own money. I.e. if the queue replenishes after the client fill, I gain information which would otherwise cost me one spread.

Of course everybody else gets the same benefit, so it doesn't really give me a lot of advantage.


"People say nothing's impossible, but I do nothing every day" --Winnie The Pooh

EspressoLover


Total Posts: 225
Joined: Jan 2015
 
Posted: 2017-05-26 20:31
@dVega

Yeah, I neglected to talk about informed flow. I agree there's value to segmenting informed and uninformed flow. Trade with the former and against the latter. The way I look at it though, in the long run captive flow is pretty much just uninformed flow. Traders who are sophisticated enough to to be informed, are also usually sophisticated enough to do TCA. Given enough time follow-on trading will show up as high market impact and crappy execution. Then they'll drop you.

Is that true everywhere and always? Of course not. But from the marketing side, keeping informed flow captive is a very hard proposition. Somewhat anecdotally confirming this, the sizable majority of payment-for-order-flow business comes from retail traders, who are pretty much the archetype of dumb flow.

@HitmanH

Good point, about OTC trades. I'm not really familiar with the less liquid futures, or how OTC works with exchange products. But my general understanding is that there's a whole bunch of rules and restrictions, that the facilities mostly exist for large block trades, and the ability to do OTC entirely exists at the pleasure of the exchange. I'd imagine if you started trying to divert major natural volume off-venue, the exchange would pretty much shut you down. Again, this is just a vague understanding.

In FIFO matching, it seems like the sure shot is getting between spread. However even in tight spreads, you do still get the option of being the first to replenish the queue. E.g. if bid is 150 contracts, and retail order hits 10, you're going to know about this before anyone else gets it on the data feed. Essentially you have the first option to refill queue by adding 10 back to bid. Or maybe add 2, expecting others will refill ~8 behind you. Legally the advantage here is you don't even have to trade in front of the customer order. But I'm pretty skeptical that this would be valuable enough to cover the cost of paying the brokerage.

Definitely agree with you regarding retail flow characteristics.

@ronin

I see where you're coming from, but disagree. (i) and (ii) are different due to differing information content of the matching trade. Assume that the captive flow is characteristic of retail order flow: small, uninformed, uncorrelated orders. That means the 1) market impact of incoming orders is near-zero, 2) the direction of orders in the captive flow does not predict subsequent market movement, and 3) each incoming order is 50/50 long or short regardless of the prior orders.

I can cross an incoming order at 10.51, and the bid moves back to 10.50. But all I have to do is wait for an offsetting sell order. At that point I can jump in front of the 10.53 ask by quoting at 10.52, closing out the position and netting $0.01. In expectation, I will only have to route a single incoming sell order to market before getting a buy that I can cross to close. 1 small sell order will have minimal price impact, and is very unlikely to move the market.

Yes, the non-captive flow could move the market against my position. But uninformed flow is by definition unpredictive of the behavior of other market participants. So the market's as likely to move with me as against. It adds risk, but in expectation it doesn't lower profits.

Price improving against general market flow isn't the same story. My counter-party could be anyone. It could be someone working a huge position, a system with a directional alpha, or a massive market order that wipes out the next level as well. All problems basically precluded in the other scenario. Either way conditional on being filled there's now some expectation that the market moves against the open position. That means this scenario shouldn't have lower expected PnL as the captive flow scenario. Overall one should be able to quote tighter with pure uninformed flow, and still make money.

moriarty


Total Posts: 3
Joined: May 2017
 
Posted: 2017-05-27 04:49
I know Eric from Nanex isn't the most loved guy in automated trading circles, but he's spotted these patterns in non-FIFO markets for quite a while. A TWAP algo starts working in futures, and the order book grows around the time when the next TWAP slice executes:



TWAP anticipation in ZT future

Not to get all "weird American Beauty guy waxing poetic about the plastic bag blowing around", but when I first saw these charts, I thought it was one of the most beautiful things I've ever seen. You have a fund somewhere using a crappy execution algorithm, and a prop trader they've never met providing liquidity. Both their bots end up communicating over a covert channel, with no explicit coordination, yet they act in both their economic interests. The fund's algo signals that it's a dumb liquidity trader, and the prop liquidity provider learns that they're likely able to hedge the trade profitably, and steps up to provide size to win the trade. Maybe some human programmed their prop algo to look at a Fourier transform of trade timings. Maybe the prop algo just picked it up through some machine learning technique. It's possible that neither party knows how or why their system works, but the fund keeps TWAPing their volume because it gets good fills and the prop trader keeps some wonky features in their pricing model because it helps their PnL.

But now your post has me wondering, is this really what's happening? If you find a dumb TWAP algo ex-post in tick data, when did traders start beefing up the opposite side? If they are merely detecting the TWAP which starts at time T and ends at T+x, they should start beefing up the opposite side a few cycles after T, since it takes time to detect the TWAP, and continue beefing it up after T+x since they don't know when it ends. If there are players acting on their own captive order flow, the beef ups should start at time T and fade out dramatically right after T+x.

moriarty


Total Posts: 3
Joined: May 2017
 
Posted: 2017-05-27 05:20
@EspressoLover Captive flow may also be informed longer-term without ultra short-term alpha. A lot of institutional equities flow fits this mold and trades in bank dark pools. Many prop firms line up to trade this stuff since they're turning over trades so rapidly that they don't care if it moves against over minutes or hours.

I would imagine a big bank can do a lot of alternative OTC trades even if trading futures contracts off-exchange is prohibited. Write a bespoke forward contract and hedge off in exchange traded futures. Trade blocks in a highly correlated ETF and hedge with futures. Most banks have a big balance sheet of cheap leverage to put on basis positions.

You're definitely spot-on about pricing uninformed flow tighter. The all-to-all exchange spread is pricing in the risk of trading with very sharp cptys. If you know your customer, giving PI is easy.

Patrik
Founding Member

Total Posts: 1333
Joined: Mar 2004
 
Posted: 2017-05-27 22:16
Re "OTC" trades in futures markets:
In many commodities it's very common to agree trades bilaterally and give it up to the clearing house - this is not usually captive flow though. Broker market of professionals trades pretty much exclusively on this basis, and most banks making markets for clients end up executing and clearing this way post 2008 (i.e. airline hedging, refinery hedging, sovereign hedging programs, etc). I'm aware of some more captive flow executing like that, but it's relatively minor - most of it would be competed, at least over time.

At times there's some trading of look-alike bullet swaps and similar in commods (as moriarty alluded to), but that's historically been more to game position limits (i.e. someone borrows position from bank, which can get hedge exemptions) - and post 2008 there's not a lot of appetite for bilateral credit risk.

Capital Structure Demolition LLC Radiation

ronin


Total Posts: 206
Joined: May 2006
 
Posted: 2017-05-30 10:30
@el,

Fair enough. You want to internalise your crosses in all but name. But the fact that you have to go to market for every single order will shave a lot of your per trade profits. It's not at all clear that you can make enough per trade to make it work.

It also sounds like you are making a lot of assumptions about the quantity and nature of your uninformed flow. Have you quantified and stress-tested those assumptions? E.g., what happens if you can only exit 80% of your positions to uninformed flow at the same mid? What happens if the information content of uninformed flow goes up by 10%? How much do you need to make per trade to compensate?

The common wisdom in retail broking is "internalise everything, charge a lot" - the exact opposite of what you want to do.



"People say nothing's impossible, but I do nothing every day" --Winnie The Pooh

EspressoLover


Total Posts: 225
Joined: Jan 2015
 
Posted: 2017-05-30 18:50
@ronin

Totally agree with your caveats. Very important points to consider. The fact that no one seems to pays for order flow (AFAIK), does seem to imply that these issues break the model.
Previous Thread :: Next Thread 
Page 1 of 1