July 19, 2011
After doing the Myst’s XDev simulation of a few days ago a few questions popped up. Would the stop-loss distribution be the same for another dataset? Does this modified script have enough general properties to be extendable to another data set? Would the performance metric average about the same?
These questions can only be answered by doing another simulation on a different dataset. Still, the need to compare to previous simulations using other scripts, the 2nd dataset was chosen.
The following graph is taken on the same basis as in the first tested modified version of the Myst’s XDev script:
(click to enlarge)
The graph has the same message as in the first test. Stop losses are taken relatively early on average. Again, the number of bars held for losing positions decreases exponentially over this group, having an R-square of 0.97, an indication of a pretty close fit. Just as in the first test, the small group of stocks having been held with losses for the longest time have a high probability of still being in the portfolio and might simply be unrealized losses.
Again, the unsorted version of the above graph does not show as well the loss concentration in just a few of the stocks or the concentration of very small losses at the other end of the spectrum:
(click to enlarge)
The average number of bars held was 541 for profitable trades, with a minimum of 258 and a maximum of 769 out of the possible 1500 bars. I find that these numbers are similar to the previous test.
The total number of trades for this data set is a little less, averaging some 2,300 per stock over the portfolio's life with a hit rate of 84%. As in the first test, the sum of all stop losses and unrealized losses amounted to about 3% of the total profits generated by this system; again, in line with my previous test.
Overall, the performance metrics were also interesting, as can be seen below:
(click to enlarge)
The system traded over 99,000 profitable trades with an average profit of over 5,700 per trade, while some 19,000 losing trades averaged a loss of about 547 each, a 10.5:1 profit factor. Considering that the script hasn’t been trained on this particular data set, having seen the data only once and only during this test, the performance results are outstanding. Even if, in my opinion, the method misbehaves at times, I still like the numbers and the way it operated over these two different datasets. To me, it is just another proof of concept that my trading methodology has real merit, and I also presume great value.
(click to enlarge)
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
( click to enlarge)
Created on ... July 19, 2011, © Guy R. Fleury. All rights reserved.