You are reading it wrong.
A low latency E-Mini strategy
- Thread starter Rum Runner
- Start date
You are reading it wrong.
The thread has an interesting premise - that perhaps informed trading and therefore future order flow can be deduced from microstructure events alone. And therefore there may be an advantage in coat tailing a large order for X ticks.
Assuming there is liquidity to do your business at prices which represent value after the big order has filled. You can make some inferences about how aggressively the big order has been priced - but that won't always correlate with future order flow.
Again, assuming there is consistent predictable liquidity to do this. One would think that without appropriate filters your fills will be offside immediately a very large percentage of the time. And the filters you need are likely not to be found in analysis of microstructure events however detailed. Too close to the data and missing the big picture. The higher level analysis required doesn't lend itself to automation.
Certainly there are low latency market crossing strategies which work. But I'd speculate that anything which is fully automated and crosses the market is extremely capacity constrained, probably needs fairly regular tweaking of parameters, and will not work for very long on a particular product. Given the fairly high barriers to entry (notably technical expertise and the need to generate sufficient volume to negotiate low fees) I wonder why so many are still looking at what appears to be a very crowded space (low latency).
Perhaps because its obvious and proven. Rather than an abstract to be explored with no apparent structure at first.
I'd hint that there may be a better advantage (not as speed/fee sensitive and more scalable) in looking at the opposite side- can toxic/informed flow be predicted, and is there advantage to positioning yourself ahead of this expected flow or with it - and for more than a few ticks. What is the primary restriction of large informed traders? And therefore what information do they sometimes reveal.
The winner takes all phenomenon makes sense of course. Which is I suppose why many or most concentrate on market making and arbitrage. All very logical, mathematical, very "left brain". But it has its limitations.
So you're playing the difference between when you get your fill notification and when the information is disseminated by the exchange to other participants. You have a small and quantifiable information advantage, but what does this tell you (if anything) about future order flow?
Assuming there is liquidity to do your business at prices which represent value after the big order has filled. You can make some inferences about how aggressively the big order has been priced - but that won't always correlate with future order flow.
Again, assuming there is consistent predictable liquidity to do this. One would think that without appropriate filters your fills will be offside immediately a very large percentage of the time. And the filters you need are likely not to be found in analysis of microstructure events however detailed. Too close to the data and missing the big picture. The higher level analysis required doesn't lend itself to automation.
Certainly there are low latency market crossing strategies which work. But I'd speculate that anything which is fully automated and crosses the market is extremely capacity constrained, probably needs fairly regular tweaking of parameters, and will not work for very long on a particular product. Given the fairly high barriers to entry (notably technical expertise and the need to generate sufficient volume to negotiate low fees) I wonder why so many are still looking at what appears to be a very crowded space (low latency).
Perhaps because its obvious and proven. Rather than an abstract to be explored with no apparent structure at first.
I'd hint that there may be a better advantage (not as speed/fee sensitive and more scalable) in looking at the opposite side- can toxic/informed flow be predicted, and is there advantage to positioning yourself ahead of this expected flow or with it - and for more than a few ticks. What is the primary restriction of large informed traders? And therefore what information do they sometimes reveal.
The winner takes all phenomenon makes sense of course. Which is I suppose why many or most concentrate on market making and arbitrage. All very logical, mathematical, very "left brain". But it has its limitations.
Exactly. My point, badly put in post above, is that there are ways to know the outcome ahead of the momentum ignition, rather than reacting to it. Therefore you are already positioned. This would give an edge an order of magnitude greater than trying to scalp a tick or two off the momentum. Not really in my interest to give more detail yet it always surprises me that this isn't thought of by others, or at least discussed in public.
ops: I reread the OP and I am reading it wrong. Now I see that the OP wants to react to the fills and follow through with the sweeping order. I suggest that you forget reacting to fills and place stops (or stop limits ) around the bbo. However, I dont think this will workGreat to see a thread with a hint of "order book" mentioned. I hope not to diverge off topic but what Garachen talked about being last in the queue is a very valid microstructure issue.
One pontential strategy I've thought about but not tested is watching the level 2 to see if I catch traders with positions placing orders ahead to make sure they are at the top of the queue and almost guarantee themselves an exit. I have heard mention of traders placing exit orders before entry to guarantee themselves to be top of the queue. Obviously this behaviour is visible and might be of worthwhile information.
Anyone have any thoughts on this?
One pontential strategy I've thought about but not tested is watching the level 2 to see if I catch traders with positions placing orders ahead to make sure they are at the top of the queue and almost guarantee themselves an exit. I have heard mention of traders placing exit orders before entry to guarantee themselves to be top of the queue. Obviously this behaviour is visible and might be of worthwhile information.
Anyone have any thoughts on this?
Didn't CME change this rule and now informs everybody of the fill at the same time? Could anyone chime in on this?
Absolutely agree, although not sure what the answer to your quiz question would be. Clearly, depending on how much time they have to execute may decide whether or not your would see any 'traces' of them. If time is of essence and thus liquidity is limited, they might just hit the market with a large market order-easily detectable. Most such occurrences are 'locally faded' although the fade move size is variable and therefore not easy to take advantage of-but often annoying enough to preclude any simple automation to go 'with size'.
a) nobody knows, this is a pure game of probabilities. But there have been many academic papers that researched large order impact on order books and information derived from that. Most such papers have shown that certainly there is no edge in fading such larger orders. The opposite in isolation has been proven to be equally futile. The order book and its information content is only one single variable in a complex equation.
It's not a CME rule. It's just a technological reality that some things have to be serialized and some things do not. When a big order hits the matching engine does certain things in a certain order. And the market data comes out with a different latency profile. There's some variation on different products due to their experimentation which you ought to be aware of if you are heavily into microstructure. But generally fills and especially top of book fills will be transmitted ahead of the multicast market data feed. But this stuff could easily change. CME just published a new protocol which has quite a bit more information in it than the old feed. It all could be very different a year or two from now.
The new mdp3 protocol was launched last year in parallel to the old fix fast feed, all I can say is it definitely treats market events differently. The latency can be both faster or slower than the old feed at different times.
The fill report you get back can be faster than the multicast feed but this is not always the case and more so during big events. This also depends on the product traded.
The fill report you get back can be faster than the multicast feed but this is not always the case and more so during big events. This also depends on the product traded.
I am quite skeptical when someone says something can't be done. It mostly means I/we can't do it but others might be able to. It's definitely not easy but academic researchers are not motivated enough to tackle this from a practical point of view. Yes, fading all large orders with some (fixed/dynamic) stop and some exit is not viable, as it's not to go with it. But that doesn't mean one can't add some structural logic around it that would make it work.