A low latency E-Mini strategy

Discussion in 'Automated Trading' started by Rum Runner, Jan 21, 2015.

  1. I have several strategies that take advantage of fast multi-level moves in various E-Minis. But, my mind is always churning and wondering about other possibilities. And since I do not have a setup capable of executing or even testing this low latency idea... I thought I'd throw it out there for discussion purposes.

    Let's say you put in one contract buy/sell orders 1 and 2 ticks away from the best bid/offer. If the BBO is 2022.50 x 2022.75. You have at least two and probably more buy orders at 2022.25, 2022.00, etc. You have sell orders at 2023.00, 2023.25, etc.

    As the BBO changes, after X milliseconds, you refresh your orders to be away from the BBO once again. The idea is that when a big sell or buy order wipes out multiple levels, you get notified of your fills and immediately cancel remaining orders, cover your position, and reverse directions (with more contracts than your feeler positions). Then fill X ticks away.

    The question of course becomes, what is the probability of a continuing move that would force you to come away with a profit? How many levels must be wiped out to make it worth it, and what other factors need to be present in the market?

    I'm sure people out there doing this. I simply don't have the data to look into this further. And even if I could, I couldn't execute at that speed. But it is fun to think about and it may spur other ideas that are possible with higher latency infrastructure.
     
    Last edited: Jan 21, 2015
  2. youll want to ask volpunter about this stuff, but hes a giant ass though
     
  3. garachen

    garachen

    I haven't tried this in ES but maybe with enough cleverness and carefulness it might kind of work. Similar strategies were pretty common in other products until a few years ago.

    To give you an idea of the problem. If you are constantly adjusting your orders you will always be at the end of the queue so when a big order hits and fill notices are sent out there will be a lot of people getting notice ahead of you. Figure 1 mic per person.
     
  4. Yes, but being near the last in the que may have its advantages? If there are any orders left in that level that you just got taken at, perhaps you can grab what is left (if the original large order wasn't going to push it to the next level), you may just push it over yourself. Or, at the very least you covered your position flat.

    I know this has plenty of holes in it. But something may be there.
     
  5. garachen

    garachen

    Your only 'advantage' being last in the queue is that you know the lower bound of the order size of the aggressor slightly more accurately than people at the front of the queue. Again, with enough study I don't doubt that someone could find the ideal place to put these orders on ES to do OK. I spent a lot of effort doing variations of this strategy on lots of products. As I said, it mostly died a few years ago. I never even tried ES because I find almost nothing works on equity indexes.

    I'm impressed that you put out an idea that to my knowledge has never been discussed on this board. An actual trading strategy. If your mind is still busy, I'd suggest considering different ways to measure order flow toxicity. Start with something like VPIN and then consider how to make it more relevant.
     
  6. Hmm. Wonder why it died? I'd think with all the quick price jumps that happen all day long that it would still be viable. Maybe the really fast guys have gotten too fast to make it worthwhile for anyone else.

    Well, if I could actually take advantage of this strategy, I would of kept my mouth shut. :)

    Regarding VPIN, I've briefly looked into it in the past but didn't really find anything conclusively significant about it. As for making it more relevant, well that is always something to think about.
     
  7. Handle123

    Handle123

    I have tested then traded based on high volume "spikes", it is often difficult to see one trader hitting large orders cause unless another trader has large order, it comes across Time and sales as many orders being filled, so you really can't see if it was one trader or small herd of them. I know enough NOT to do in the morning, as this is a "random" method of entry, whereas if you add S/R to it, then it comes close to 90% win rate of making 1-4 ticks, however, just randomly doing it during day I couldn't make it work profitable. I did find trading during 3.25 hours of lunch time to work well enough cause of the lack of volume. Where it worked best for me was non day session, and do have automated method but follows a trend, it has limited signals cause it seeks much more than 1-2 ticks entries and traded over twenty markets. I have also automated method of what the human eye can't see, some orders are "flashed" and removed in nano-seconds and never come up on the Dome.

    Often what happened to me was you get filled and can't get out at profit in the mornings, trends are stronger first 90 minutes.
     
    fortydraws likes this.
  8. garachen

    garachen

    With the raw data feed there's additional information that allows you to piece back together the large orders and distinguish them from the herd of individual trades. I thought it would be useful but then never tried writing a strategy to take advantage of the information because I can't trend follow if my life depended on it. And the 'bag of random heuristics' approach is so tedious to cull through. Certainly a profitable approach, but you never know how far you have to dig before finding something.
     
  9. Edited for stupidity: I need to start reading a little more carefully...sorry for the clutter.
     
    Last edited: Jan 21, 2015
  10. jelite

    jelite

    I like to dig in tick-data... and all I can tell you is that there already is one 'market making' operation that does this all day long on most all liquid US futures I looked at. It is quite unlikely that you will beat them at that game-doesn't however mean that slowed-down variations of that idea couldn't bear some fruit.
     
    #10     Jan 21, 2015