Dynamic portfolios
Dynamic portfolios
It would be interesting to find out how many of us here trade a dynamic portfolio where one adds and eliminates intruments dynamically depending on a systematic or discretionary rule set.
Dynamic Portfolio Selection
I have been exploring the concept of a self-selecting portfolio, where the actual MarketSystem selection is based on various statistical variables.
The general idea is that all 'eligible' MarketSystems form a 'universe' for possible selection. Each day all MarketSystems are ranked from high to low based on the selection statistic, and risk capital is allocated accordingly. When the money runs out all potential trades remaining will be skipped for that day.
The effect is that the higher probability MarketSystems will be traded while the less probable will be skipped. Sort of a Darwinian concept I guess .
I have done some testing and have the framework in place to test any suggestions of performance stats. Results so far indicate that bar-by-bar stats outperform trade-by-trade stats by quite a margin, which is the opposite of what I had expected.
My overall concept is that I can keep on developing systems, toss the eligible ones into the universe and let the software do the actual selection.
Over time I expect to split the selection into two layers - the upper layer will be manual for the sake of minimizing correlation, while the lower layer will be the actual self-selection one.
(BTW, I consider correlation to be primarily between MarketSystems.)
Would anyone care to offer (polite) suggestions on the merit or otherwise of this approach?
The general idea is that all 'eligible' MarketSystems form a 'universe' for possible selection. Each day all MarketSystems are ranked from high to low based on the selection statistic, and risk capital is allocated accordingly. When the money runs out all potential trades remaining will be skipped for that day.
The effect is that the higher probability MarketSystems will be traded while the less probable will be skipped. Sort of a Darwinian concept I guess .
I have done some testing and have the framework in place to test any suggestions of performance stats. Results so far indicate that bar-by-bar stats outperform trade-by-trade stats by quite a margin, which is the opposite of what I had expected.
My overall concept is that I can keep on developing systems, toss the eligible ones into the universe and let the software do the actual selection.
Over time I expect to split the selection into two layers - the upper layer will be manual for the sake of minimizing correlation, while the lower layer will be the actual self-selection one.
(BTW, I consider correlation to be primarily between MarketSystems.)
Would anyone care to offer (polite) suggestions on the merit or otherwise of this approach?
Dynamic Portfolio Selection
Sir G,
Thanks,
Kevin
Can you elaborate on this? Are there principles or themes you've found that could make a difference to somebody looking to construct their own set of dynamic selection rules?My systems include dynamic portfolio logics as systematic rules. These logics aren’t introduced as a last enhancement to the system, but as the primary foundation of the systems.
Thanks,
Kevin