Algorithmic FX

Discussion in 'Journals' started by The_Gekko, Mar 30, 2011.

  1. Preface...

    Over the years I've looked at forums, here and there, finding less and less interesting things that people seem to talk about. Nothing personal to ET of course... There are several good journals here, some good news stories posted, and I've seen some great discussions being had. I think personally for me that the experience and knowledge I've gained over the years has just left me finding more and more things that I already know, or seeing things that simply don't apply to my methodology. As I recall most of the good journals here actually showed you how to do something new you didn't know how to before, or gave you the ability to reproduce the results the author shared.... So, case in hand, I'm hoping that I'm able to at least give something useful away in the course of me documenting and outing some of my work. Whether it be how to's, new ideas, or remembering things you forgot. Although, keep in mind, this is my journal so I'm going to be doing things the way I see fit. I'm not necessarily trying to teach you anything. I'm merely keeping account of, and exposing some of what I'm doing not necessarily to or for you...

    A couple other caveats....

    I trade spot FX mostly. I trade other markets and may decide to discuss them, if I find myself posting that long, but FX is my niche.

    I'm sure I'll have a tendency to drift off, at times, from pure subject. If nothing else it may help provide for a more interesting/entertaining read. I've been a trader, on and off, most of my adult life, and I have all the battle scars of the business. I know success...I've known pain....but I always come for the high.
    I also plan on sharing some things via youtube videos. In some cases it may be because it's something that physically would be good to show, rather than explain. In others, it may be that it might take me too long to type about it. To be perfectly honest, it sometimes can take me a good hour or more to write a well thought out written message/email and I don't have the time to waist in my day. So, please don't always expect me to have well written sentences with punctuation and such. I'm not a writer, it just takes me too long... Which leads to another thing. If I get a private message and don't respond quickly please don't take it personally/offensively. I do wish to communicate with others as it's been people I've met along the way that have helped me garner more success at what I'm doing. So just give me a moment... I have a lot going on ;-)

    Additionally, but foremost, I'd like it well understood, from the begining, that I have no affiliation with the companies whose products I'm choosing to use. I've found them helpful to me in shortening the gap between being a full blown financial institution and a one man show. I'm speaking pariticularly of Microsoft, Matlab, and Amazon products. There are several free alternatives, other places and means to achieve some of the same capabilities. So I'm not suggesting their use. It just works well for me... Although, I understand Microsoft has some very good programs for small businesses as well as trials. So most interested should be able to follow along. Matlab on the other hand is a little less adverse, but most of us know it's quite easy to attain by alternate means. I only suggest that when you can afford it buy it. Their support team is uncanny. If anyone in between the posts would like to follow up or point to R's version of code examples, over Matlab, feel free to do so. I just don't know about R well enough to keep differing copies, but I know it can be a good alterantive .

    Finally, understand that although I'm sharing and discussing I don't have any alterior motives behind this other than for people to know. This is not my way of advertisement I'm just documenting proof.
  2. Words To Trade By.......

    "Where you want to be is always in control, never wishing, always trading, and always, first and foremost, protecting your ass."-- Paul Tudor Jones
  3. Introduction...

    I have a little more time to devote to this today... To anyone reading this I'll tell you now it may start off slow, but I plan on doing this daily very soon. I'm not going to jump off immediately into the trading part because I plan on discussing my setup and the algorithms a bit first. No more than what a CTA would describe his/her trading program, but more than what they would typically describe on how they got there. What I'll be leading to is something that I already know works, but I'll be approaching it as if I were trying to find out if it does....

    Now here's the scary part...Some of the terminology may keep some from immediately understanding what I'm talking about. If you decide to read on I'll tell you the subject matters that will come now so you may get a chance to Google/Wikipedia them to gain a little knowledge first.

    1. High Performance Computing
    2. Pairs Trading
    3. Error Correction Models
    4. Support Vector Machines (online variant)
    5. Wavelets
    6. Unscented Particle Filters (online variant)
    7. Fuzzy Logic (just concepts... won't be using it much if at all in this journal)
    8. Neural Networks (just concepts...I don't use them in the manner you might assume)
    9. Diffusion (understand concepts)
    10. Homotopy(understand concepts)
    11. Law of large numbers

    I think that's most of it... At least some of the more important things. Not necessarily in that order though.... One other essential thought behind my use of these scientific methods is that I don't believe you can/should predict market behavior. I use them instead to control outcomes. Although any trading system you build is thought to be predictive in nature.... I don't like the idea that I create things that are predictive, but instead I believe they are reactive. So if you're looking for a thread that's going try to teach you what will happen tomorrow... look elsewhere. Although the mind has the ability to perceive things in a nonlinear way i.e.(look into the future). You generally don't realize just how stupid you are until you try to predict markets rather than react to them when trading real capital. Being smart is a great attribute I hold dear but knowing when to play stupid always gets you further in everything you do. So I'll leave that there...


    I've come to think, before I really give this a go, that the reader knows a few more things and should agree/understand them, because you're not authorized to read any further if not.

    I'm 30 years old. Not exactly old by any means, but certainly not a youngin. I've been doing this nearly 10 years, and gave away my 20's to this. I've even worked for a major financial institution that burnt to the ground because of it's CEO. While others were partying and doing gosh knows. I was doing this. So if you believe you can't learn anything from someone who isn't 40, 50, or older then don't read any further.

    I've grown up in an age of computers, Facebook, Twitter, Youtube, RockNRoll, HipHop, etc... So it's instilled in my personality, and I love all walks of life as well. Kevin Rudolf and Lil Wayne have been on the playlist all morning. There's only one race in my eyes. The human race. So if you believe you need to be some pompous stuff shirt douche bag to write a good journal. Look elsewhere because I'm not one. Although I am extremely sharp minded and I have to put an emphasis on that statement because I don't always meet a lot of people like myself except over the net.

    I've never traded hundreds of millions of dollars before. Mainly because I choose not too. Yes I do this for the money but I don't have some grand dreams of being a huge fund manager or being a billionaire for that matter. I also have other ways to make money so this isn't my end game. I just like the freedoms this provides me, especially the way I do it. I have traded tens of them and managed a couple hundred accounts though. I typically take runs at the market. Like 2 years in 1 year off etc... but I'm going for a 5 year on period right now considering some of the technology advancements and market conditions of late. Mind you this isn't meant to sound like a braggart I'm just explaining myself.

    I value suggestions...I don't believe I know everything. Opinions on the other hand are like assholes as they say. So keep them to yourselves. I don't get mad easily, but I'd prefer not be bothered with flamers because when I do get mad I might be following you around every post just to let you know who the bigger asshole is.

    Lastly... part of the reason I'm doing this, as I said before, is for documentation purposes, but it's also for therapeutic reasons. Sometimes you find yourself doing this from inside your head too often. Which from my experience isn't a good thing. So consider this journal as me doing this from outside of my head as well.

    So if I haven't offended you yet you can feel free to keep on reading. I promise I'm getting to the good stuff.

    Now Playing
    Kevin Rudolph
    In The City
  4. In2Deep


    This idea intrigues me. Looking forward to seeing your results. I'm working on an ATS myself, which uses back-testing on various time frames to determine entries, so I guess that would be predictive. I take it you don't do back-testing.

  5. No I actually do back/forward test, monte carlo analysis... the whole works. What I mean by saying that I don't like to think of my systems as predicting anything is that I don't let a neural network or some other method dictate trading direction, ahead of it actually beginning to move in that direction, or where a stop or limit should be. Instead everything is event based and nothing is set in stone as to how many pips I'm bagging or anything like that. Rules are defined more like after you're in a profit exit at break even point, or trail stop such and such percent else exit on impulse event. It's just all about control...never trying to predict per say.

    The biggest focuses are put on how much I could potentially loose...Like say for instance if I make 100 trades all with a 5 pip stop loss the max I could potentially loose in a perfect world would be 500 pips plus commission. I make rules that would decrease the chance of me loosing my max possible loss based on these events I'm able to control. It may not be exactly clear what I'm saying right now, but I'll get into more detail down the line.
  6. njrookie


    Sounds very interesting. Especially interested in how to apply ML to quantitative forecasting. I played with SVM, ECM, PCA, VAR, Bayesian DLM etc, myself, and find them better at predicting inter-relationships than the overall trend as a whole.

    Please keep it coming...


  7. It doesn't matter whether or not you are trying to sell anything; people here will still give you shit. You might see it as "giving back," (I thought the same thing) but you will soon realize how ungrateful some people here are.

    If you plan to attract $$$ the best documentation of proof is going to be audited record(s); not a thread on ET.

    Good luck!
  8. Ahhhhh yes... Very important concept/thought you have there that I share as well. I use SVM, for example, to decide what the best state space is in determining if you're in a up market or down market. Then use another to decide if you're in a trending market or sideways. Yet again not letting it decide the actual trading rules, or letting it use wild kernel methods in giving an answer. Overall I'm just trying to improve linear methods by adding some nonlinear techniques to them. Although some may think the use of a lot of what I'll be discussing may seem complicated... I tend to think I use variations of different methods that are simplistic actually.
  9. Yeah I'm expecting a little shit. ET is very critical. The long ramblings on the first page were meant to avoid any onlookers who felt like offering opinions. The documenting part is for myself, not to attract $$$. If you think about what I'm doing you might notice it sort of appears to start off like it's a book. The auditing part is accounted for. I may even let some look at this thing in real time... but again NOT TO ATTRACT INVESTORS. THIS IS NOT AN ADVERTISEMENT. I'm going to put a great effort into letting others know I know what I'm talking about. Then... one day... someone will ask a question... and there will be no response... At least from me... I'll probably be gone from ET, and at least from this thread, but hopefully my journal will live on.

    Good Luck to you too my friend, you have a long road ahead. I recall your posts were looking pretty interesting till the flames got in the way.
  10. njrookie


    A good model in my mind should be adaptive to market environment (automatically detecting regimes of trending vs. range markets; quickly re-tuning model parameters/trading size due to different market volatility; change of magnitude/directions b/w different securities/instruments). It should also more or less applicable to different time frame.

    I greatly appreciate your efforts and PLEASE do not let others deterring your efforts. I am very eager to read and learn from your experience. No need to show records. It is much easier to make up a record than to make up a good model. A general discussion about the market and approach can stimulate ideas to develop one's own model for backtesting and implementation.

    My experience is that linear model is more stable and capture 90% of variation if you know what variables to look at. End of day a good model seems to translate some dimensions a good discretionary trader is looking at into something quantifiable and allow models to test/estimate the intuitive linkage more precisely. For example, a trous Haar wavelet multiple scale analysis is MACD in nature.

    Anyway please keep it coming...

    #10     Apr 3, 2011