Optimizing a Trading System

Discussion in 'Strategy Development' started by traderguy02, Nov 1, 2005.

  1. I would like some advice from some of the experienced traders out there on system development. It is often stated that most good systems can work in a group of markets. If you built a system that seems to operate on 15 minute bars but struggles at 5 min , 1 hour, daily etc....should that be a sign this system cannot handle price action and should be "thrown out"? For those of you that develop various strategies of your own...do you find that many of your systems are [time period specific] and they still hold up ok and are reliable for consistent use?

    Appreciate any advice on the matter...thanks
     
  2. Not nesessarily. When you reduce the timeframe, your cost of brokerage and slippage increases as the system places more trades. At some point, any system not matter how good, will become unprofitable. So just because it doesn't work on 5 minute bars doessn't make it altogether bad.

    However, if you have a profitabale system using 15 minute bars on the S&P, you might expect it to also work on the dax using 15 minute bars. If it doesn't, it could indicate the system is not that robust.

    Runningbear
     
  3. How come all strategies "no matter how good" eventually go bad...systems are based on historical data right?...so why one day does it not work anymore...what are typical changes in the market that throw any one good system
     
  4. I don't think that's necessarily true, but they may need to be adapted to changing markets. For example electronic trading is a huge change to a market. Program trading is another. Increased or decreased volatility is a third....

    I run a system based off the king keltner system from the book "Building Winning Systems with Tradestation" but I've made modifications to account for differences in the two markets I trade it on. OPtimization can be very useful provided you don't optimize *too* much and you also test your optimized system on "out of sample" data (data you didn't run the optimizations on).

    I don't buy any of that stuff that it must run on many other markets to be a valid or robust system. Every market is different and has different characteristics. For example a trend following system will only work well on markets that trend, but even on some strong trending markets it won't work because their "noise" level is too high and your stops get hit too much to be profitable.

    SSB
     
  5. good deal....sounds reasonable
     
  6. mmillar

    mmillar

    Time is like any other system parameter - you need to optimize on it to make sure it is stable. A system that runs well on 60 min bars but not on 55 min or 65 min bars is unstable.

    I wouldn't necessarily expect a system that runs on 60 min bars to work well on 5 min bars.
     
  7. Test the system on different time frames and you can discover the answer. I write programs to backtest trading systems. Some time frames are better than others. Each method is specific to a market. Each stock is different and my methods are optimized for each individual stock. Methods don't really stop working. Study 30 years of price data and you might see losing streaks. Losing streaks don't mean the method stopped working. Losing streaks are part of the process, part of the probability of a win or lose.