Would you Trade a System that failed a backtest ?

Discussion in 'Strategy Development' started by Murray Ruggiero, Dec 6, 2005.

Would you trade a system which lost money during a 10 year backtest ?

  1. Yes

    12 vote(s)
  2. No

    44 vote(s)
  1. Murray Ruggiero

    Murray Ruggiero ET Sponsor

    I made the statement that , I would never trade a system which did not make money during a backtest. This is because a backtest is a best case look at the system results. I made these statements in another thread. It seems like some people don't agree with me. What does the group think ?
  2. In backtesting a specific period you may encounter varying market conditions. The market may trend up or down, chop, or the myriad of other scenarios that can be encountered in differing time frames.
    So although a particular system may fail in an overall test in may be well suited to the present market conditions. Thus that system would be turned "on". This could only be identified through diligent backtesting.
    In my opinion correctly recognizing the "where are we now?" quickly is the key to superior performance.
  3. of course not
    which idiot on here said they would hahahah

    who cares about different market conditions

    define the market conditions if it relates to market conditions which is a requirement in all profitable systems.
  4. I've backtested my system on metastock before and out of the 25 stocks I tested it with, it failed probably 18 of them. So, I guess my system is not meant for automation but I use my system day in and day out and I've been profitable.

    My position trading and daytrading are based on this system. A lot of it are also dependent on money management. I truly believe that a trader can randomly trade a stock and be profitable regardless of the system if he has great money management skill.
  5. sure, if it failed bad enough.

    fading or reversing bad systems can work nicely at times.

    good luck

  6. DrChaos


    "A lot of it are also dependent on money management. I truly believe that a trader can randomly trade a stock and be profitable regardless of the system if he has great money management skill."

    This is untrue, and mathematically provable to be so, if you mean 'profitable' to be 'more profitable than luck on ensemble averages'. You can always be profitable because you were lucky, but then your past performance has no relation to what will happen in the future.

    Money management changes the relative sizes and perhaps orderings of individual profits & losses. That may be able to help Sharpe ratios (reward divided by portfolio fluctuation size), and if you use that to lever up (an already positive expectation strategy), then, indirectly, money management can help profitability.

    But if there's no positive expectation edge in the underlying system, it's useless. Think about it. Can "money management" result in a long-term positive profit on a perfect & fair roulette table?

    One interesting phenomenon (which may be entirely theoretical). You could combine a significant positive expectation system, but one with large volatility in returns, with a negative expectation system, and get a better Sharpe ratio than the first one alone. This happens if the returns from the second system are sufficiently and predictably anti-correlated with the the first.

    I don't know if this ever happens in real life, and I suspect that the anti-correlation could easily go away a whole bunch faster than the negative expectation! And if you used the higher Sharpe ratio for more leverage....oops....that's the LTCM error.
  7. Most of the stuff I do, I am told, is no good because it did not backtest and show profits.

    In ET a while back several of the back testing evaluations had data on the degree of the failures. The distribution exceeded an order of magnitude.

    100 percent of the backtesting did not include volume which I use as a leading indicator of price.

    So far, my conclusions regarding backtesting are that the people doing it do not know how the market works and second they are have no skill level in doing what isrequired to construct back tests.

    For what I do I use more than one fractal to trade. Often, I believe, the backtestors do not deploy that aspect for their testing.

    for the equities stuff that I do a person used C language to make it mechanical. When the program was operated using cash, it made 11.1 % every 6.6 days on average. I do not program but I can read long lists of program stuff. We tweeked it for 4 1/2 hours by putting notes on the print out. There was a 20% improvement in performance. This may be the equivalent of forward testing using software and an operator of the software who used cash to make the record. He did not do backtesting; he simply programmed for a series of emails (four layers removed from me who started the daily waves (4 sets per day in the first 1 1/2 hours of the market operations.

    Back testing seems to be a lot of bullshit so far. No one has written much about it so far as I understand it.
  8. Cheese


    Whether a failed system or any system, the efficacy of backtesting available raw data is the real issue.

    If you want to establish predictability for each trading day in advance of the market Open (eg YM) then backtesting will not give you tradeable or reliable or sustainable results of any worth.

    However I would not deter anyone from pursuing such quest. There is nothing better than finding out for yourself.
  9. from Jack Hershey (AKA Grob109) Jun 2 2000, 3:00 am
    Newsgroups: misc.invest.technical

    Date: 2000/06/02
    Subject: 30 minute warmup bar trading.

    Fundamental Money Making Concepts.

    I use simple mechanical systems to get people to understand the basic
    concept of making money steadily and with little or no risk.

    When you trade daily for 6 1/2 hours a key thing to consider is not doing
    too much to make some money.

    By choosing a futures index of any sort on any exchange in the world, you
    have put yourself, for 6 1/2 hours a day in a place that is truly dull and
    unexciting. Being there is fairly safe and not too demanding so you can
    relax and repeat a few tasks over and over to make some money.

    I work first with 30 minute bars to frankly eliminate any sense of urgency.
    I use the prior days last bar to get the ball rolling, or I suggest you wait
    until the second begins to eliminate the end effects of the market.

    Here is a progression of four mechanical methods to illustrate making money
    primarily and secondarily to illustrate that losses are neatly reduced more
    and more as a little sophistication enters the picture. I also introduce
    how in a trend you can switch to the most favorable side of the channel to
    exit. Because this is very simple and mechanical there is no need to
    clutter it with a stop system as yet mostly because it an index tied to the
    performance of and aggregation of stocks. We can tuck stops in easily
    though as a commitment to our ordinary discipline.

    The four items in the progression are:

    1. break out of prior bar.
    2. slope pairs of bars.
    3. overlapped pairs slopes
    4. retracement.

    Here is the progression:

    1. set up a 30 bar display for a futures index.
    2. enter on the breakout beyond (above or below) the prior days last bar
    3. hold until the current bar breaks out of the other end (from your long or
    short entry) of the prior bar.
    4. hold on inside bars.
    5. hold on successive bar break outs in the same trend.
    6. on breakout of 3., reverse so you can take on new trend trade.
    7. repeat 3. through 6. for remaining bars of the day.
    8. settle at end of day.

    Here's something which Grob109 offered and I backtested. Its quite a loser. Its easy enough for anyone to backtest. The bullshit is not the backtesting so far..............
  10. Murray -- Are we talking about trading with our own money or with your money? I don't care what I trade if it's your money :D :D :D
    #10     Dec 7, 2005