Rolling out a fresh strategy

Discussion in 'Automated Trading' started by lolatency, Apr 28, 2009.

  1. Do you guys throw out 100 shares at a time and put a limit on the number of round trips?

    I want to know what your procedures are. In the corporate world, we'd roll out strategies to new names and watch them run -- any weirdness got killed and the traders manually cleaned up the books. It wasn't the best procedure, but we could afford it.

    Now that I am more or less running my own show, my initial trip-wire is 10-trades on 100 shares at a time and then I go back and paper-analyze what happened, then slowly up the limit on trade counts, then size counts.

    I want to know how your emergency procedures are set up for when disaster strikes. If you have too many strategies, then do you just monitor the overall risk, or do you set up a new account for the strategy with a small amount so that the account blows out on the few dollars you put in?

    ... I'd like to compile a list of methods before I throw this prototype into the wild.
  2. you have no clue what you are doing

    not only is your approach completely wrong, you make it even worse by thinking you know this business

    maybe you worked for some bucket shop like FXCM

    but trading you don't know

    forget the shit you know and start over

    HINT HINT: LESS POSTING ON ET more staring at charts and thinking
  3. anyway...

    yea, onelots is the way to do it. i don't limit trade frequency as i like to get a sample i can look at, but if it's super high frequency i'll limit the number of symbols and then bump that up if it looks ok and test how it handles the basket.

    no such thing as 'too' many strats, but i think you mean more than you can manually monitor. for that, overall risk monitor with loss and exposure limits at the system and strat level. notifications at set thresholds are a smart thing too.

    as far as emergency procedures.. too many to list. my rule of thumb is, if i can imagine a what-if that could blow-out or severely hit my acct, then i ask myself if it's preventable. if some failsafe or combination of failsafes could prevent it, then it goes on my list of todo's. if there's no prevention or the implementation is difficult, then you assess probability of the risk and treat it like any other strategy. auto-trading in this department is always walking the line between failsafe implementation and productivity... risk tempers where the emphasis should lie.
  4. vikana

    vikana Moderator

    When taking a new system live or made a major adjustment, i run it one day in the paper-trading account (at IB). If/when OK i take it live at smallest size and just watch closely.

    After the close i run the system in simulation and do a trade-by-trade comparison to see how close I got. Go back to step one as necessary and repeat.

    Emergency: go "exit only" and if necessary manually exit
  5. bespoke


    always, and i mean always, limit the amount of orders that can be sent out when you're first testing

    in the sub/function where you send orders, always have something like

    if nFills > 5 then exit sub

    i've written lots of programs where if i had not done this i would have sent out 100s of orders because i overlooked something tiny. and it actually happened once... almost ended my trading career and set me back greatly

    also. i start off everything with 100 shares and monitoring 20 or so symbols to make sure it runs properly. then bump up the number of symbols as its easier to do that with less risk than it is to jump from 100 to 200 shares. i usually stay with 100 shares for a week or two so i can collect a good size sample

    but if your time frame isn't minutes then i would just paper trade it. you can still get a good idea of how well your strategy works. not so much with sub minute trades
  6. I'm a fan of the order limit because, consider conditions that persist over time. Even in a simple MA strategy, "buy if over x" can persist for a while and if you made a dummy mistake and didn't forget to check whether you already had a position, you could be sitting there sending ridiculous orders.

    Anyone do the account partitioning, to limit the total risk? i.e., throw 5g into a sub account, let the strategy log in, etc? This might work better in a sub-LLC arrangement than with a retail broker.
  7. vikana

    vikana Moderator

    Good one. Another one i have is to limit the number of shares to a fixed max amount. I use volatility based position sizing, so it's possible to divide by something small if you're not careful.
  8. What measure of volatilty? Historical volatility scoped to some time window?
  9. imo, historical, same for most other data checks that influence position sizing. it's not much fun to throw out a big fat short on a $100 stock your strat thinks is worth $1. don't ask me how i know. naz market watch employees would be out of a job if more firms did this though.
  10. bespoke


    I usually have a text box where I enter the number of shares I want to do or the max $ I'm willing to risk per trade. I always put a max limit on that in the code because it's very easy to add an extra digit by accident. I once took 20K shares when I only meant to do 2K. Never again...

    As for volatility based position sizing, I actually now use only yesterdays volatility as a measure for each individual stock. It's giving me better results than an average of the last 10 days, 1 month, or 3 months. And my measure is really simple... I use the average absolute price change of a stock per minute. That measure is most useful to me cause I trade on short time frames.
    #10     Apr 29, 2009