Geoff Veteran   Posts: 180 Joined: 12/4/2012 Location: Byron Bay NSW Australia  User Profile | Yes true Gary, I think the other issue that Keith raises with this example and also in another thread, is the effect on performance outcome by running the same simulation with start and end dates that have been shifted by just a few days. IMHO this phenomenon is to be expected because from my analysis of portfolio moving performance data, the day-to-day performance (CAR) can be very volatile meaning that if you start a simulation on a date when the port is trending down, the result will be very different to the result you get if you start the sim when the port is trending up. The compounding effect (CAR) can then produce a large difference in sim results. The other factor (IMHO) that can possibly skew results are 'freak' trades which are low probability (maybe 1:300 trades) but with very high %P/L, usually between 6 to 13 STD Deviations above the MEAN, meaning they are way above the average or expected individual trade performance results for that Port. Now, my point is that if one of these 'freak' trades happens to occur on or near your sim start date then the final CAR for this sim analysis run will be heavily skewed, compared to an identical simulation that has a start date that misses this trade. (I have an Excel spreadsheet that you can dump the PW 'Historical Data' into which will analyse the data for volatility and 'freak' trades, if anyone is interested let me know.)(Also, I have seen PW results effected by as much as ~10% by these trades) Lastly, it was Jim Dean who alluded to this phenomenon when he said in another thread (I hope that I get this right Jim) that possibly the only way to get a 'more' reliable sim result is to run multiple PW sims with similar settings (maybe moving the dates around) and then average the results to evaluate its true performance and avoid 'saving' a DP that has a performance result that may not be valid. [Edited by Geoff on 9/10/2014 7:39 PM] | |
