Sophistication of current market making algos

Discussion in 'Automated Trading' started by NasiWarrior, Nov 4, 2011.

  1. Have we reached the era where algos can track the order book, track where market orders (& marketable limits) traded: how the book has changed in response to said trades, and scale the markets in the hunt of stupidly tight stops ?

    I mean; algos limit orders are always on the move, where as "the dumb money's" orders are just static and so they are totally visible... ...especially if those static orders are already/become badly priced.

    Simply put: Are Current Market Making Algorithms (Globex) just making a market ? Or are they also "flexing" the market to shake retail out, and essentially, swing trading the markets themselves for more than just the spread ?

    I'm thinking in particular of the NYMEX energies complex.
  2. This has been going on for a long time, not just now. Yes they track the book and know what's going on, but they only use that information when they know others are not "cooking" it by placing ghost bids. In most cases the book isn't worth much, but in some cases it can be - when other algos disagree with themselves on how they should "cook" it. Then, a smart one can slice through the said book. They know with millisecond precision how far along the line they are on any given quote, and they're usually first because they'll have quotes ready at every price level long before it gets near there. They can also cancel them within a millisecond - and again, the book doesn't mean much any more.

    Moving limit orders isn't only for the algos. I did it with keyboard commands up/down per tick in real time when i was doing discretionary daytrading. It's something you should do yourself, it's not hard, and it's not an advantage algos should have. Moving them every few seconds is feasable, and also free of charge.

    Market making, it's gone beyond that. They are kind of making it harder to get hit on any limit order because they are way ahead of you in every line. They sometimes place a very large bid, say 10m, but then if the algo calculates the price is slicing through it too quickly, it cancels at 2m and then you're hitting the one below it at a better price - and guess who's first in line to bid you. Yes they know how many people are in front, it's easy to deduce. The exchanges have no problem with that either.

    I also believe there's a way for them to know if they just hit someones stop or limit order, because of the way those different orders are made active at the exchanges via your broker/clearning/ecm, though I'm not quite sure.
  3. sma202


    How do they identify other algos?
  4. For most intents and purposes, they don't need to identify them. But when they do, they look at regular changes made at regular levels. It's like this, when humans act on signals, they're not so precise. Their timing is not quite precise, and if they make 10 orders of N they don't execute them 100ms apart exactly. These precision timings in changes can be detected, but still a lot of it is left to guesswork. Some algos try to introduce randomness to avoid detection, but that's not really useful in the end - predictability is valued more than being undetected. It also depends how much access the analyst (or other algo) has. So it's a combination of frequency and size detection, together with de-noising of the market movements generated by other algos and humans. It's a bit like separating one sound out of a song. On the other hand, others may use wildly different methods. I suspect some have more access to the exchanges to know order routing well in advance, and a bunch of other data that is only available to "authorized" parties, and I can't really speculate on how much or what. It also wouldn't be hard to kind of "tap into" the exchange data lines and see even more info - on the border of legality. I'd be surprised if some of them didn't really do that, because I know it would give them a big edge.