Super fast trading - all automatic

Discussion in 'Automated Trading' started by trillenium, Jan 17, 2004.

  1. I've started a thread on pings & execution latencies, anybody who wants to contribute is welcome:

    High Pings & Latencies in Day Trading?
     
    #61     Jan 25, 2004
  2. prophet

    prophet

    You sound fearful of new technology, like those who were once afraid of mechanical/programmed trading, or trading over the internet. Technology makes new things possible and reliable all the time. My original point was merely that colocation can supplement or replace the need for a PTP T1 for the purpose of receiving data, analysis and issuing orders. It’s great that you have such high standards for quality and reliability. Lets hope it doesn’t impede development of novel system ideas because you make things out to be too expensive, and not scalable or clean enough for your tastes.
    I still don’t understand why you search for ways to blow stuff out of proportion. The collocation and latency issues are unrelated. So what if MY colocation has a high latency? Only latency jitter affects my systems… and this is well under control. For systems that need lower overall latencies, there are better server farms, some much closer to the exchanges or data providers. My network reliability is excellent. And that’s between IB’s servers and a server farm in San Antonio TX!
    If my talk about colocation offends you so much, then don’t reply. I find it more offensive that you keep bringing up my 10 trade/day or 60ms latency figures, or “big players won’t do this” as if these are reasons why collocation won’t work for the task described. Again you insult our intelligence by (1) arguing that only big players with deep pockets can afford the connectivity to develop ultra-fast systems, (2) not following a logical argument, (3) assuming I’m ignorant about everything, and (4) using heavily condescending language like “make you understand”. All I’m saying is that collocation is a LOW COST tool for distributed acquisition and processing of high frequency data. It’s an entry-level solution for those that can’t afford PTP T1s, or who want to supplement their PTP connections. How many times do I have to repeat myself for you to see the point?
    There you go again mincing separate issues. I was arguing for collocation and talking about the reliability and speed of MS Remote Desktop over unreliable networks. I never said that unreliable networks were adequate for monitoring systems, even though it is tolerable in my situation. You took two things I said and twisted them to use against me. My systems operate at low frequency, 1 to 10 trades/day, 30 minute to 8 hour long positions, trading constant size fully mechanically with no variable sizes. Provided appropriate leverage, stop losses aren’t necessary. In fact they hurt performance because most strong moves against the systems are highly mean reverting. Of course, OTM straddles are always sound insurance for major unpredictable events. Basically, I do not need second-by-second monitoring. Comcast had a few outages last week because they were upgrading some equipment. My 56K dialup backup was more than enough to monitor my systems. Comcast is now reliable again.
    Redundancy mitigates the engineering concerns. The main risks are operator induced, say if I were to over leverage the systems and end up on the wrong side of a 9/11 magnitude move without a hedge.
    On page 8 of this thread. You characterize my idea as using manual executions when I never said that. Maybe you were just being sarcastic.
    I appreciate your concern. My systems monitor data integrity and take the appropriate actions if a data feed is lost, gaps or if the feed acts unusual.
    It depends on account size and leverage. To be as fearful as you appear suggests you are using a great deal of leverage and require ultra reliable executions. You have a right to trade dangerously. However, you can’t legitimately characterize my systems as dangerous when you know nothing about them.

    For clarification, I agree a 1000 trade/day system is potentially dangerous and needs ultra reliable executions, redundant feeds and networks. However, in your last post you were clearly saying MY system setup is dangerous. Lets keep these issues separate.
    But you are condescending all the time. You seem to assume I don’t understand technical and market risks.
     
    #62     Jan 25, 2004
  3. nitro

    nitro

    Pffffffffffffffft. If you only knew... :D

    Now your point is that it can supplement? Well, I may grant you that, but replace? No it can't - not today...

    No problem there...Again, if you only knew :D

    They are? Explain that one to me?

    I just thought you said they weren't related? Make up your mind :eek:

    If you say so...

    I agree with everything you said! Oops, I said all those things. Ugh, back to square one with you.

    BTW, who is "our"? You don't have like 50 handles on ET and have multiple personalities by any chance do you?

    You don't know what you are talking about, but look, you seem really intent on convincing me that you know what you are talking about, when I know for a fact that you don't. So stop trying.

    Uuuuuuhhhm, I am "using things against you" :confused: :eek: See the last line of this post.

    Uh, I don't care?

    LOL. How would there be more "operator" induced if you had the computer systems in house automatically trading as opposed to at some remote location trading? This is getting funny.

    Whoops - I see it. My bad. Doesn't change the issues...

    I doubt it. Again, it is what may happen that you have not thought of of that is scarry...

    I need to know nothing of your systems. If you are running this at a colocated farm using IB as your datafeed, that is enough to know that you are either trading peanuts, ergo your misplaced courage, or the reason for the colocation is misplaced in the first place.

    If you had started that way, this would have been all academic.

    I do not belive they are seperate, why should I keep them seperate? There is nothing that _you_ are doing at a colocation using IB that I could not do at home from a business quality DSL line. The colocation adds nothing to the mix. If you need fast systems, then going to a colocation site that has 60ms of latency is a joke. If you don't need fast executions, then why go to a colocated farm?

    I am nearly 100% sure that you do not. Most likely, you read a couple of books at Borders and now consider yourself an expert on high risk, hard Realtime systems. Go and work as a CCNA at a job for a year, then go and work for a hedge fund for another as a programmer, and then come back and see if your point of view changes. Maybe go to a local nuclear reactor site and see if they want to colocate their systems that monitor the reactor...

    The "condescending" part that you keep accusing me of is something you need to work out with your therapist. Probably father/son issues...

    nitro
     
    #63     Jan 25, 2004
  4. Aaron

    Aaron

    I am considering a colocated farm because I would expect it to have a better internet connection than my office DSL line. Don't you think the internet uptime would be an advantage of a colocated farm versus DSL?

    My DSL goes down on occasion. It's not a problem for manual trading, but if my future automated system is unable to use a dialup backup or phone the broker, it would be a problem.
     
    #64     Jan 25, 2004
  5. nitro

    nitro

    Aaron,

    Don't do it. Contact me offline if you want my input...

    nitro
     
    #65     Jan 25, 2004
  6. Aaron

    Aaron

    Thanks, Nitro. Your PM mailbox is full and you don't have emailing enabled. Here's what I wrote to you, please PM or email me...

    Hi Nitro,

    Yes, I'd like to get your input. Thanks for offering.

    In automating my trading, a big risk that I'd like to minimize is that my internet connection (Earthlink DSL) goes down. I am considering a server farm to reduce the internet connection worries. What would you recommend?
     
    #66     Jan 25, 2004
  7. prophet

    prophet

    Well, first you are critical about a technological tool …
    …then you admit it is useful, albeit only as a supplemental technology. Which is it? Useful or not?
    Why does my 66ms latency generalize to mean that ALL colocation is inadequate? It’s easy to find colos with 30ms. Why is 30ms round trip delay detrimental to ultra fast systems? You never answer these question with any technical arguments.
    They aren’t related. Somehow you still refuse to explain why they are.
    Other readers on ET.
    I don’t want to convince you of anything. I’m just defending the things I’ve said, against your deliberate distortions.
    It is irrelevant where the systems operate. The concern is the “operator” using safe leverage, hedging, and using simple coding safeguards to avoid mechanical overtrading or over leveraging. Losses are therefore controlled even if the systems, broker or exchange crash, of if there is another terrorist attack.
    I’ve detailed most of my safeguards. If you’re so much smarter in these regards as you claim to be, then tell me specifically what I’ve missed.
    Again you claim to know everything but won’t give specific reasons why this won’t work. If you do reply to this challenge, please don’t ignore the following safeguards and aspects regarding my systems, including:

    6 non correlated, similar performing systems
    0.5 to 8 hour intraday position lengths
    low slippage
    multiple server site redundancy
    multiple feed redundancy
    data integrity monitoring
    overtrading limits
    fixed position size
    full hedging with OTM straddles
    composite historical max drawdown set at 6% of capital.
    There is a huge difference. You obviously don’t understand quality of service for TCP/IP networks. Average latency says very little about connection reliability, especially for server farms with redundant routes. Try comparing packet statistics between a collocated server on 100 Mbit/sec ethernet to DSL/cable.

    I don’t need fast executions. I use a collocated server to obtain high quality data. As I’ve said, I have compared it to data collected on lower latency networks and found no major differences.
    Such blanket statements suggest you have a paranoia that you are wrong and someone else is right. Why don’t you try arguing your points in an academic way?
    Just come to an academic level for once.
     
    #67     Jan 25, 2004
  8. nitro

    nitro

    Oy vey,

    Round and around we go. You are still trying to get my approval. I don't think you know what you are talking about, no matter how much you cry and I don't have time to educate you, certainly not for free.

    But keep advertising yourself, maybe you'll find some sucker on ET.

    nitro
     
    #68     Jan 25, 2004
  9. abogdan

    abogdan

    Hi guys:
    Attached is the block diagram of the Robotic system that we are running in our office. It might answer some of your questions. Each robot executes around 2,500 trades a day.
    Cheers,
     
    #69     Jan 25, 2004
  10. prophet

    prophet

    Nitro,

    I don’t want your approval in the slightest. I’m just defending my statements against your distortions and unprofessional remarks... nothing more.

    If you want, we can drop this.
     
    #70     Jan 25, 2004