Random Trading

Discussion in 'Automated Trading' started by Loki45, May 19, 2010.

  1. Loki45

    Loki45

    1) I tried several - 75 ticks, 800 ticks, 1 minute, 5 minute. It never made a difference, and I think it shouldn't, because my trading program does not care about how the price data are grouped

    2) I let the program run a minimum of 8 hours

    3) I did not - the entry is strictly random time controlled; in most cases 1 to 60 seconds after the close of the previous trade
     
    #21     May 20, 2010
  2. That is a little more than a day's session for most futures and 1/3 of a session for forex. Hardly enough to claim any significance. You resutls are biased by selection.
     
    #22     May 20, 2010
  3. Loki45

    Loki45

    I planned that already, and I agree it should result in 50/50 minus 1 tick for the spread
     
    #23     May 20, 2010
  4. Loki45

    Loki45

    I am not so sure. 8 hours is about 250 trades minimum, and if all 8 hour runs show negative results, it should be statistcally significant.
     
    #24     May 20, 2010
  5. ventus

    ventus

    check to see how you're using stop and limit orders in your code
     
    #25     May 20, 2010
  6. You need at least 10,000 trades based on my experience in 1 minute timeframes to claim any significance of results. Some others say about 1,000 but I do not agree.
     
    #26     May 20, 2010
  7. Loki45

    Loki45

    Not at all. After a time span determined by the random generator the trade is closed, no matter what the result is. Remember, this trading program is not about making money. The only purpose is to find out what the market's response is to randomly generated entry and exit orders.
     
    #27     May 21, 2010
  8. Note, check the distribution of your random numbers. If they aren't random enough, try using a pseudo-random number generator (PRNG) or true random number generator (TRNG). (This may have minimal impact on your results for small number of trades, but is worth looking into as number trades increases.)

    http://www.random.org/randomness/
     
    #28     May 21, 2010
  9. Loki45

    Loki45

    I use the random class available in .NET languages, it generates pseudo random numbers; but I always change the seed value, so it should be "random" enough for my purposes.
     
    #29     May 21, 2010
  10. You did not answer my question about the sample size. Why do you think 8 hours is enough to prove enything? It may be that in that sample, there is a positive expectancy for random entries.
     
    #30     May 21, 2010