Thursday, 1 December 2011

We Deal in Probabilities

The essence of trading is the probability of success. To know probabilities, I have to back test and collect live or SIM data. From these statistics I know what I could expect. However, as Clint Eastwood said, "if you want a guarantee, buy a toaster" as the markets can unfold in a different sequence to your historical data.

So why is testing any use? Well, what is it exactly that you are testing? What you should be testing is the actual picture you're trading. What is the probability that the picture, when traded, results in a profitable outcome? If you have a large enough sample size over a variety of market volatilities, the results have great value.

What is unknown is the order in which the losing and winning trades will unfold. This is what determines, among other things, what the drawdown of the "system" will be. If you add hard targets and hard stop losses then you add some more unknowns as the hard exits over a large sample will only provide an "average" good stop and target. As volatility changes, as day types change, these "average" exits will sometimes be good and sometimes not. This impacts both profitability and drawdown. Using dynamic exits related to market activity is the way to mitigate the effect of these variables.

Below is a DAX chart during U.S. RTH. Market was quite sloppy today but Flo managed to do quite well. This as a fairly short term algo, entering on a limit order and exiting dynamically at market. First trade was a loser and the others, winners. The yellow line is a 135 CCI which can be used as a filter.


  1. Use 135 for a gauge of Momo in a Range bound market? Larger Sample of data then the 45 periods

  2. Hi EL,

    When discretionary trading, do you prefer using range bars or renko bars?


  3. Anon 17:07, 135 is goodfor very choppy and range bound markets as it gives a better idea of the real trend.
    Anon 18:16, I'm using more renkos for discretionary when the market is very volatile as it gets rid of noise.