The blending of algorithms of various period lengths is central to my method. However I do not apply the technique to asset management -- I apply it to individual security selection.

Why so specific opposed to re balancing ?

You have to understand how each strategy is generating alpha.

you have to control the risk.. if you can do that by combining a number of algorithms then you are good.. if not.. then it is pointless..

but yes.. it is possible at the level of an individual security..

Trying to manange risk at the level of the indicidual security will make the portfolo return closer to the market return and possibly remove some of the benefits that algorythmic trading and computerisation can provide. You’ve got access to tools and techniques non-algo strategies don’t have, extend and innovate accordingly.

As a practical matter, how else can you approach risk except at the individual security level? One can say by owning several securities, but in the end you buy them one at a time and each stands on it's own merit, yes? What can an algorithm do to change this? Thanks, David

i am a proponent of trading various algo's on one market rather than trade one algo on various markets. i focus on a few markets it's easy to focus and succeed.

We've written one algo that manages multiple strategies independently (8 so far) and ranks them based on various factors including probability of success and market sentiment. It's an interesting project that is gaining traction. You can see our updated results at RagingFX.com. Comments welcome.

Thank you guys for the insight, ive been looking into this for some time. @mark brown i have been using this sort of blend dependant on market regime. If we can define what direction the market is going in and how strong the trend is for this direction, and also understand how each algo or strategy behaves in each specific regime then i can move assets according to this.

I think too many algos might cancel each other out but well chosen algos that work together would be beneficial.

You can accomplish this with custom tailored genetic algorithm. Let it pick which instruments to put in the portfolio. Craft an evaluation function that relies heavily on alpha, has some penalty for churn. Run it several times and rank the top one or two solutions. Look at all of the solutions and pick the one you like best to rebalance your portfolio. There are several techniques to get this to work in a reasonable time.

Casey

1st off, Please look past any spelling errors as I am on a cell phone in a Costa Rican tree house in Monte Verde on holidays from Australia.

Ive worked extensively with both GA's and SA (simulated anealing). My experience is that GA is a superset of SA. Anything you can do with GA you can achieve with SA's and the latter are simpler and faster. Both provide stochastic optimization. Think of GA performing a conformal mapping overlay of the stochastic (random) process that is common to GA & SA.

I use SA to do non linear optimization of trading systems. Using the system total profit divided by std dev of equity as the cost functio for tuning the algo. This world well.

Back in early 2000's I coded up systems that individually tuned aprox 4 different algos and voted which one to use dependant on the risk adjusted equity return on backtresting/walkforward.

Results where mediocre.

I suggest that , rather, you might like to look at the market regime / behaviour and choose stocks / ETFs that suit the algo style that you have chosen (mean-revert, momentum etc).

If you take a helicopter view , really these two approaches are simply different mappings to a similar objective (preserve capital and maybe make money).

Yes, this is how we do it in our Integrated Predictive Analytics Platform a/k/a "Thunderbird APDE(t)". Integrate 7 Algorithm Categories for Embedding Intelligence, Big Data, and Data Science for deployment running in RT i.e., 1 – Crunchers, 2 – Guides, 3 – Advisors, 4 – Predictors, 5 – Tacticals, 6 – Perturbs, & 7 – Causals.

Reach out to me should you want to know more! We can help you should you like if it does not take too much of our time. George Ingram

Each algo is required to have a positive expectation from the decision making strategy, meaning N decision making strategies for a portfolio. This is the necessary condition. No matter of allocations will turn individual negative expectations into positive expectations. Ralph Vince's book "Mathematics of Money Management" answers your question in depth.

i tell you what you can do - achieve diversity trading several models on one market.