I developed a successful system ... maybe

Discussion in 'Automated Trading' started by bridenour, Jul 30, 2009.

  1. Hallo folks


    So I have been developing automated systems for a while...most were either not really successful or there were a few that were successful but only based on current market conditions, which gave them a fairly short lifespan.

    Recently, over the last 12 months or so, I have developed an automated system that I am pretty certain works and works very well. The reason for this post is that, realizing how difficult this actually is to accomplish, i'm looking for a gut check or validation that what I am saying is true.

    The situation is more or less as follows...this particular automated system is only partially automated - that is to say it gives me the signals at end of day and I place the orders before market opens. Most of my systems are fully automated but this one in particular is more of a swing trade system, with most trades lasting a few days to max a few weeks.

    I have backtested this system over a 10 year period, and it basically has the following characteristics:

    CAR: 160%
    Max DrawDown: 35%


    Now, I realize that isn't very impressive, but what I think is more relevant is that no single year since 1999 has produced less than a 35% return, and most have been between 100% - 200%. These stats include running the system with significant leverage (2:1) -- i can achieve a much smaller drawdown without leverage.

    Most years, even with leverage, the MDD is only about 15%.

    Anyway, i have been trading the system for a couple of months, and I have been able to validate that I can enter and exit the trades without much slippage, given that most entries are on market open orders, and most exits are stop limit orders.


    With all of the above being said -- what am I overlooking? Is there some element that might make this system dangerous or unprofitable that I have missed? Any feedback or challenges are appreciated....

    br
     
  2. You have to account for increased/(decrease) volatility in your backtesting, position sizing, stop losses and profit targets.
     
  3. If you actually achieve those returns they are spectacular. Like better than any other fund manager ever.

    I would get data for markets you have tested on and run it on that. Like if your current portfolio and the data you tested on is US from 1999 to 2009, get date from 1989 and forward. Also get data for the past 20 years on foreign markets. If your system performs anywhere near on that data, please PM me your parameters.

    :)

    5yr
     
  4. bridenour

    I think a 35% drawdown would be very hard to handle. Knowing nothing else about your system, it is hard to offer anything constructive. Personally, I would be willing to trade a smaller total return for a milder drawdown.
     
  5. Specifically, which year was your Max DD? If the MaxDD occurred in a poor market environment it's acceptable, but if it was during 2003 to 2007 I would have to say it's not acceptable. Happening during 95 to 97 and 99 to 2000 would also be unacceptable.

    Do you know how correlated your system is to the market? What kind of securities was it trading? What was the timeframe involved?

    On the overlooked question 35% is a bit excessive if that was unleveraged returns. Leveraged returns at those levels are fine.

    A Sharpe above 2 on a backtest like this is the golden rule. If you also have a PF of 2 or greater you're good. Most likely though, if you were to cut your CAR in half and double the drawdown which is a good rule of thumb to live by for a system like this, I'm sure you'd find you are using a position size that is too large.

    On the Sharpe I noted your 10 year testing period, so a sharpe above 1 with a pf above 1.5 is more in line with an acceptable risk reward trade off for that time period.
     
  6. Mandlebrotset -

    Everything i do considers volatility. My entries even exclude certain signals if they are too volatile (based on ATR as a % of pps). Anyway, in particular, position sizing and exit management is solely volatility based.


    Syrtrader --

    Via backtesting this system performs well over any 6 month period between 1999 and 2009, specifically on the W1000. I think the principal would apply eslewhere, but would need tweaking.


    Truckload -

    I have considered your concern about the 35% drawdown, and agree it might be hard to stomach. However, it has happened only twice since 1999. I am starting the system with a 100k stake and my general feeling is that i will probably reduce leverage and bet sizing as the stake stake becomes larger.

    bwolinsky-

    first, i dont think the system is very well correlated with the general market, but im no expert. I will say that in 2008 it would have (apparently) returned around 130%...the two MDD periods happened in 2003 and 2009. The 2009 MDD happened on the March 6th market reversal. What I have noticed about the MDD's is that they tend to be from a point where the equity level has quickly accelerated to extreme highs (meaning it was a DD from a sudden ridiculous equity appreciation level).





    Over the weekend I will go through the ten year period and record the supposed returns using a few different position risk levels and leverage levels. I feel pretty confident about this system but I have been investing long enough to know things can be unexpectedly treacherous, so I am truly interested in feedback from those who might point out flaws or challenges I might have overlooked.
     
  7. You're on the right track. But again it goes back to leverage. PTQQS has a 160+% APR and DD of 28% closed, but that's at the equivalent of 4:1 leverage. The system has more than halved the APR going forward, but DD has been within its backtested norm.

    Might I suggest adding a linearregressionline predicted value to your signal? It really keeps you from fighting a trend.