|
kmcintyre
 Elite
       Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
Previously I posted about some experiments I performed capturing performance stats on the 21 default EFs against a universe of stock RTM strategies. The takeaway for me was that PW generated wildly varying results which (to me) defied logic. Even over a simulation range of 7,75 years... (see my other posts if interested.)
I had this idea that I might be able reduce the performance randomness by running multiple parallel dynamic portfolios. To that end I assemble 16 DPs. I used 4 EFs which exhibited the lowest Coefficient of Variation from my 7.75 year runs. I selected 4 different rebalance dates. (3, 11, 17,and 23 cuz they were nice prime numbers...)
I built an account with the 16 DPs enabled, which resulted in 564 strategies being used over the 7.75 year simulation range. I then tested the account over the same 28 starting dates used in my previous experiments.
The result was a much more predictable account that didn't care very much what day in January, 2007 I pulled the lever. The CV was reduced to 0.0551, I'm attaching a spreadsheet for anyone who cares.
But before everyone gets all excited and starts building accounts of parallel dynamic portfolios (which will probably drive OV into the ground)...
I then tried the same kind of experiment using the 17 strategies native to ARM4 Margin. I saw even more consistency using just those 17 strategies. CV was down to 0.0288. The return, as measured by EE, was less, but the % invested was also much lower. And this solution only required 144 strategies to be used over the 7.75 year simulation.
I'm including a spreadsheet with those results as well.
But before you run off and try to implement parallel dynamic portfolios (which might ruin OV for everyone)...
I took plain old ARM4 Margin and ran 28 simulations over the same date ranges. I collected performance data so I could compare it to all the machinations of PDP. Using a static portfolio (17 strategies) generated much better returns with a fairly comparable CV (0.0422) without using all the computational load and wiz bang algorithms of PDP.
I believe PDP has merit. IF I could find EFs that remotely approximated the performance I see with AISS (automated iterative strategy selection), PW might be a profit reaping machine.
But as it is, I've not been able to find any dynamic portfolio (or group of parallel dynamic portfolios) that can beat ARN4 Margin, a simple static portfolio of 17 strategies which don't even use dynamic lists.
Why am I sharing all this? I'm hoping either Nirvana or one of you Elite Traders will benefit from my research and come up with a product that will consistently make me money. (The markets had another banner year. I lost my ass[ets], again.)
Happy New Years
Keith
Attached file : PDP - 7 year - Stock RTMs.xlsx (12KB - 260 downloads)
Attached file : PDP - 7 year - ARM4 Margin RTMs.xlsx (14KB - 289 downloads)
|
|
Mark Holstius
 Elite
    Posts: 744
Joined: 10/11/2012
Location: Sleepy Hollow, IL
User Profile |
Thanks for your work Keith - very interesting idea (PDP) that might dovetail with something I'm doing too.
Would you be so kind as to post the equity curve graphs for both your PDP account and ARM4 from 1/1/2007 thru 1/1/2015?
Thanks,
Mark
|
|
kmcintyre
 Elite
       Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
Mark,
Here is the export and a couple screen grabs for the PDP portfolio that used the stock RTM strategies.
Keith
Attached file : PDP.xls (178KB - 264 downloads)
Attached file : PDP portfolios.png (110KB - 269 downloads)
Attached file : PDP graph.png (111KB - 269 downloads)
|
|
kmcintyre
 Elite
       Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
Mark,
Here is the export and screen grabs for the PDP ARM account (that used the 17 ARM4 Margin strategies only).
Keith
Attached file : PDP ARM.xls (176KB - 264 downloads)
Attached file : PDP ARM portfolios.png (114KB - 271 downloads)
Attached file : PDP ARM graph.png (119KB - 271 downloads)
|