OmniVest User Forums
OmniVest Pro Tools Beta
What to do about PB (and PW for that matter)
2 replies, 1001 viewings
|
OmniVest User Forums
OmniVest Pro Tools Beta
What to do about PB (and PW for that matter)
|
Last Activity 10/7/2016 12:00 PM 2 replies, 1001 viewings |
|
|
|
| Printer friendly version |
kmcintyre Elite    Posts: 890 Joined: 10/11/2012 Location: Portland, OR  User Profile |
The BEST result I saw was CAR of 49.6% with a MaxDD of 39.5% (CALMAR = 1.3). 1) I can't stand up to a 39.6% MaxDD. Way too rocky a road for me. 2) I have no confidence that the CAR EF that generated the best results will reproduce its top ranking given any small variation in the simulation date range. (More testing required.) 3) I know from ISS tests that OV has the potential for MUCH higher returns and MUCH higher CALMARs. (If only one could accurately predict which strategies to use in advance...) 4) I still have this feeling that PB (and PW) are automated dart throwing machines. I've collected a bit of data and it appears there are wild swings in performance that defy my idea of consistent performance. (More data analysis required...) 5) Perhaps the best thing PW/PB provides is walk-forward testing. But I need to have truly predictive and reliable EFs to make all this strategy and portfolio switching worthwhile. It's been months and I'm no closer to having anything I'm willing to trade. So is this as good as it gets? Keith  Attached file : PB DC Trade Pools tabular.png (101KB - 261 downloads) Attached file : PB DC Trade Pools graph.png (100KB - 264 downloads) | ||
| gbarber Veteran Posts: 282 Joined: 12/30/2012 Location: Pearland, TX User Profile |
I would say try a different strategy in picking the strategies you are using and work with strat lab first to create good custom strategies, then PW to create good portfolios. I have just begun experimenting with PB using some portfolios I have developed with PW. So far I haven't been able to get an improvement over simply using the portfolios individually but far from finished with experiments. My strategy was to create a large pile of custom strategies for particular market types. Rather than go through the mental gymnastics of trying to figure out strategy type would go well with a certain market type, I just tested a very wide variety strategies (all with dynamic lists) in strategy lab. From each run I tried every variety with all the standard set of conditions and some more custom conditions, I picked the few (sometimes just one, other times 4 or 5, and occasionally 0) that had good stats. By good stats I mean CAR of > 20 with MDD < 10. Sometimes I would pick one with a CAR down to 12 if the MDD was < 4. Other times I would pick one with a MDD in the teens if the CAR was above 50. I also looked at % wins. I filtered out those with a % wins less than 60. As you can see, I was going for those with Calmar greater than 2. I found many with Calmar of 5 or more up to 10. The result of this search gave me a very large pile of strategies for the different market types I was investigating. I have not yet completed that search. Still have 2 or 3 more market types to go through. Then I used the strategy page to recalculate that very large pile over the last 3 years to see how each strategy behaved in the current market. I pulled the resulting list out to an Excel spreadsheet. Then I sorted that to find the ones with the best combination of CAR, Calmar, and % wins for the current 3 years. I fed that list to PW with settings of 200, 1, 15, and 1 and got the attached set of results for the multitude of EFs I am using. This set was pretty reasonable. I haven't started live trading with it for several reasons. I am having trouble getting the account i set up with the best portfolio to compute trades. Still waiting on N to help with that. Also, I still have a lot of experimenting still to do to make an even better portfolio set. So it seems to me that these tools will give good results. Of course the proof is in the live trade results and that is still to come. Attached file : sample portfolio set from PW.docx (19KB - 306 downloads) | ||
| kmcintyre Elite Posts: 890 Joined: 10/11/2012 Location: Portland, OR User Profile |
I appreciate your post, but I went down the path you are on half a year ago. (I posted many spreadsheets documenting exactly where I found all the good bananas to save others time and effort. Search "monkey business"...) I've ran extensive tests using the best of the 750+ custom strategies I built. I tried using simulation periods covering 7+ years, and shorter periods that matched the market segments used when building the custom strategies. And two years, and one year, and 3 months... I've automated the ISS process and ran tests to determine what the potential returns were using the custom strategies and market segments previously mentioned. My conclusion was (and is) that EFs aren't sufficiently predictive and/or the ECA process robust enough, to achieve high return/low drawdown performance on a consistent basis. Yes, I have found combinations of strategies and EFs that perform well given specific account settings. But those "winners" fall apart with small changes in the simulation period. And they are the result of dart throwing on my part. PW didn't find the magic beans hidden in the bag of muggle beans... I'll continue to collect data and gain knowledge. And it may well be that I'm just too stupid or lazy or fearful to achieve success with OV. But one thing seems clear to me. PW doesn't automate the task of building great portfolios. Rather, IMO, it has added more complexity and mystery and variation. Is this as good as it gets? Keith |
|
|
|
|
Administrator
Forum Moderator |
Registered User
Unregistered User |
 E-Mail this thread to a friend(un)/Freeze thread | |
Toggle e-mail notification |
