Great job hft. Some questions: 1) How reliable is market depth for algo models, for instance in CME, when so many orders are fastly created and cancelled? 2) Why such a preference for "marketable" limit orders vs market orders to enter or exit a position? 3) How do HFT normalize all timestamps from the different exchanges? Or even inconsistencies within the same exchange between the data feed and matching engine? 4) What is the true challenge in HFT: the quality of the strategy or a seamless execution? 5) Why so many people are raising against HFT? 6) How do HFT trade the news? Again, congratulations for this thread. HFT is not good or bad; it's "just" evolution like in many other industries. The problem with HFT has more to do with unfair trading venues rules that clearly favors the "big" HFT's
None. Local time is local time, and packets from everywhere are captured by a single capture card or server in each colocation. Nothing else is truly synchronized to HFT levels of accuracy. Timestamps between colocations are supposedly synchronized to something like a millisecond by an NTP system of sorts, but that's too much fudgery to label as synchronized in HFT terms.
1) There is some value to depth if analyzed properly, and it depends more on the overall liquidity of the product rather than the speed (i.e. depth on CL is more important than on ZN). 2) This isn't HFT specific in my mind. It's preposterous to me so say 'I want in at ANY price'. Even if it's a heavily liquid market, at least putting some type of limit price in limits your risk and doesn't lose you anything. Would you walk into any retail store and do the same with a piece of merchandise? 3) There's a lot of heavy analysis that goes into just this. It's too detailed to go into here. Let's just say we apply different models, but it certainly isn't perfect. 4) I'd say moreso execution, but both are obviously important. 5) I don't want to get too far into this and open more of a Pandora's box than I already have. In a nutshell I think much of it is fear of the unknown (hence my reason for starting the thread in the first place). There's also valid concerns about possibility of more wild situations like flash crash, which admittedly was at least exacerbated by HFT, if not mainly caused by it. But the solution here should be to regulate it better, not banishment. Like you said, if something can be done to even the playing field between 'big' HFT and regular I'd be all for it. For instance I hate the development of faster inter-colocation networking (i.e. microwave). It kills small HFT startups (of which I'm not by any means) but it stifles innovation and competition that would eventually better the market (more liquidity + tighter spreads), which is bad for the market/world in general. I can't think of how you'd effectively regulate this though, but just throwing it out there as an example. 6) There are direct newsfeed services (Dow Jones, Need To Know News, etc) that will stream the release directly to you as a binary packet. You can parse that and act accordingly.
I don't think I fully follow his logic (and at first glance I think he doesn't go into details about what his alpha really is that triggers the trades). I could be wrong, but it looks like just regular medium frequency trading. Maybe 'market making' because he rests an order on the market to enter, but not in the sense that he's on both sides of the market at all times. In that case, sure I think there is edge in medium frequency automated trading in general. If and when I lose my job that would likely be my next endeavor.
hi HFT, 2Qs: 1: do you have a strategy trades a single contract that has stopped working? how do you determine that it has failed after a long time of winning? can you give example? 2: when you do MM, what is the max holding period till you balance inventory? thanks.
1. Yes, lots of them, PNL ebbs and wanes in all strategies, including the single product scalpers (pick your symbol, ES, EURUSD, ZN, FESX, ...). It's not really a matter of determining 'it has failed', but moreso 'What's the work/reward scenario of analyzing this strategy as opposed to something else in the pipeline'. 2. Until the market closes.
Can you give an example of what you mean when you say the pnl ebs and wanes in single product strategies? Earlier you said that you have virtually no losing days. So does that include occasional losing days from specific strategies but net profitable across all strategies? Or by ebbs and wanes, do you mean that strategies are still profitable everyday but very inefficient as in very low profit per contract?
Ebb/wane is a bad description. "Decay" is a better term for it. So we have edge by defining alphas/etc, but without any proactive analysis its PNL/edge will decay over time. Then you put in some feature or latency enhancement that boosts it back up. If and when PNL goes negative, you have to decide whether to put in more work to put the 'boost' in or simply dump it in favor of working on something that's potentially more profitable.
hft, I'm a software developer, so would like to ask you some related questions. What made your shop build its own execution engine, instead of licensing commercial ones, like Deltix? Is your execution engine CEP-based? How many developers did it take to build your execution engine, excluding trading models? How many full-time infrastructure developers does your shop have at the moment?(just ballpark figures) Thanks.