Since sub 1ms latency trading was a STANDARD PRODUCT... With Reuters 4-5 years ago (except in Montana)... One might assume all that 70% of volume Algo trading... Is done by a few hundred groups operating at 1ms latency... And with tech budgets that START in the low 7 figures. And it's STILL a Zero Sum Game less massive costs... So 90% of these ultra-sophisticated players are losing money. Why on earth would one CHOSE to go HTH... HTH with these guys via a retail broker using retail tools? Why not just go sit in the Big Game in Vegas? Even if you are a living legend like TJ Winston. There are many ways to make money trading... They may not work as well as they did 5-10 years ago... But winning the Speed Game or the Terabyte Analysis Game... Is not an option for anyone here.
Good post. I have often wondered this. Apparently, Clearinghouse trades automatic latency sensitive strategies as a retail. So, I am very curious to hear Clearinghouse opinion/rebuttal to the above post. This discussion is getting educational and interesting!
My opinion on this is really simple: regardless of time frame, you can't ignore market microstructure. The more you ignore it, the more it costs you. It does not matter if your time frame is 5min or 2 hours; microstructure helps you get the entry right. Every tick or quote of the day is telling you something about the market's current opinion, so if you're just looking at some 1min chart or some larger mathematical relationship, you may be missing key insights on whether you're entering or exiting at the right point. Now, if you have an edge that makes money, more power to you. I just care about every last tick, every little rebate, and making sure I'm always trading with the market instead of against it. So, it's in my interest to assess what the market is doing as quickly as possible, HFT or not, so I can make fine-grained revisions to my trades when I need to. If I'm legging into a spread, I want to minimize the slippage and the damage. I want real world results to mirror tests as closely as possible, and I don't want to hand-wave and dismiss everything as something belonging to some kind of generic "error term." If you read 90% of the posts where traders are wondering why they are getting slammed or placing stops in the wrong spot, you'll notice these guys don't "zoom in" to figure out why they picked the wrong spot. Sometimes there are obvious clues in the market as to where the right entry is; it just takes some patience and time to learn what the market is telling you, but it takes lower latency to act on that information properly -- and I think this is a fundamental truth that's applicable to all traders of all asset classes, prop or retail, and at all time frames.
Appreciate your response. It is clear that you find it useful for your trading and you have mentioned the reasons for it. However, it reads like a lot of hand-waving to me (no offense intended). Can you at least give one example of how having a low latency feed that saves you say 100/200 ms actually helps you in achieving better execution over time? I hope even one example will make the discussion much more concrete. Feel free to PM if you don't want to write about it here. Thanks.
That's a simple answer. The average retail trader is close to one full second behind the market and most retail shops (like the etrade & scottrades of the world) don't even supply a full book, meaning they only subscribe to a partial market data feed so your data is both late and not complete. I mean one second from the time the exchange records the record to when it hits your desktop. It isn't just about a low latency feed. Half of what I do is clean up networks so that they are efficient enough to deliver feeds on time. For example, a burst of market activity can overcome the buffer (if any) on your network card which would then push the data into the machine's cache... if you are paying thousands each month using a cheap router or network card can ruin your data feed. Most individual traders and even in the small hedge funds that I see all have stuff like iTunes and screen savers or desktop wallpaper installed as well as a slew of other stuff almost like downloading too many aps on your new iPhone and slowing it to a crawl. I know clearinghouse specifically was asking about low-latency feeds so I won't go off on a tangent too much but my point is that the quality of the data as well as the infrastructure is equally important as the latency of the feed. Getting back on topic and addressing how a low latency feed can help over time - the reason why I mentioned infrastructure is because most people willing to pay $2,500/month or more are going to do things properly: dedicated hardware, proper infrastructure, legit internet connection, decent router, etc. If you are trading on your $1,000 laptop why would you ever pay $2,500++ a month on a data feed to drive laptop data? As clearinghouse said, he drills down to the market micro structure. If you have poor infrastructure then during fast moving markets or bursts of activity your data will be delayed and you will drop packets whether it be held up or dropped locally by you or held up & dropped by your service provider at their syndication servers. Over time this is when entries, exits and stops are usually hit. As the data comes out it is essentially delayed or not complete (dropped packets) so as you enter, exit or approach a stop you are doing so with incomplete market data. Trading over time based on inaccurate or incomplete market data can hurt a strategy. Most of the time when you hear traders say things like "why didn't I get filled" or "why was my stop hit (or not hit)" during those times the market is moving quickly and you are caught on your heels - you are never trading proactively, you are trying to figure out what just happened and trying to minimize the damage. You have two choices, either change your strategy so that your entries, exits and stops aren't impacted by faster moving markets or spend up for the complete picture of what is happening. It can be as small as realizing you need to be more or less aggressive on your pricing or even something as little as changing size. Many traders find (speaking from personal experience) that a certain order size is more or less likely to be filled. If you have the tick by tick you could read the tape and see that every 100 share order got filled but your 200 share order didn't or that the book keeps getting swept up/down three cents not four... so your bid/offer just matched the new NBBO but didn't get filled. It's as simple as whether or not you want to be trading based on complete information or only a partial picture. The good news about doing nerdy IT stuff is I can type lengthy replies like this at 2am while I wait for servers to rebuild... bad news is that something broke in the first place.
I'm not sure what brokers people are using here, but if we assume IB I'm not sure what the point of optimizing feed latency so heavily is if you're going to send orders though TWS? Conversely, if you have co-located DMA, you're probably not worried about the price of feeds.
I don' t trade through TWS for anything that needs precision. IB will give you bad execution latency, but I'd argue you still need to do your best because their commission structure is so extreme (relative to prop) that you're opening a position somewhere around a full tick to a half tick against you on account of just commissions. And this is off-topic, but this is why I feel like IB should let customers trade against each other and internalize flow for a far reduced fee. I should be able to post passives for execution against a removing customer, or they should allow us to meet each other at the middle. Why let the cash flow go to the exchanges? Let us post order to an internal IB exchange and charge us .0010/share or less, not .0050+pass-through-rebates or .01+pass-through-rebate, or let us cross each other at the midpoint. What's it to them on their backend? It's just shuffling shares from one account to another.
It takes 16ms just to update the screen the dude is staring at. And what just updated is a full frame (or more, if you're using a "tv" instead of a "monitor") behind what's coming in - meaning you're 32ms behind before it's even registered in your brain that something on the screen has changed. You simply can't do "low latency" if you're a mouse jockey.