In the first post of the DVI series, we concluded with a simple system that performed pretty well in a 26 diverse ETF universe. In the past I have found more strategies with similar or performances, and then the real challenge (and fun) for me starts. How do we trade this strategy?

To further pounder this, I guess I have spent +90% of my research time to find nice strategies. At the beginning of this year I forced myself to stop searching and to start trading based on the 3-5 most promising strategies. I am glad I did as I learned a lot about trade execution, setting up my infrastructure, the discipline of trading etc. Today I still trading with a relatively modest account of 25k$ (more will added when I feel I have mastered the routine), but I still do not have a framework telling me how to should split my money and divide across strategies, how many strategies etc.

My objectives aim to generate 10-15% uncompounded a year with ~10% DD. As a consequence of the ~10%MaxDD requirement, most investors will agree to diversify their investments to reduce risk. This comes straight from the text books on Modern Portfolio Theory (MPT). But how does this diversification work with mechanical strategies? Can we treat mechanical strategies as individual assets as described in MPT? And if so, what would be a good way to select and combine mechanical strategies into a portfolio of strategies? How do we decide to activate and deactivate strategies?

Maybe these questions are all very easy to answer for people with more financial education/investor/trading experience but I have not sorted them out yet (I hold an MSc degree in Mech Engineering and an MBA). So I invite all readers to help me sort these out, please comment!

In the mean time, I am reading and experimenting to deepen my understanding. Some interesting topics I have found are the developments around the Adaptive Time Machine by David Varadi, the latest book by Ralph Vince (The Leverage Space Model – LSPM) and it’s implementation in R discussed over at FOSStrading.

Here is what my current line of thinking is.I am going to try to validate this in the coming weeks and see if I can work this out into a workable setup for my own trading.

- when taking a position, one should never trade below a certain minimum value because of transaction costs. Let´s say 2500$.
- returns coming from trading strategies behave like returns from individual assets and so MPT or LSPM models hold for them
- define the tools to be used to experiment
- use MPT or LSPM to find an optimum allocation of money to trading strategies maximizing reward vs risk of drawdown

a) experiment with number of strategies added and their correlations

b) investigate the difference when applying a strategy vs individual ETF’s or applying a strategy vs a universe of ETF’s and threat the return stream as an individual stream

c) investigate on/off switching of strategies based on their historical performance - translate into a practical setup

I started off on the 2nd post on the DVI system, but I felt that I had to clear this out of my system first. I will post the 2nd part on the DVI system pretty soon.

Happy Trading!

QD

Hi QD,

I’m a long time reader of most of your blogroll posts and this is my first comment on your blog…let me say it sounds very promising.

Having said that the Portfolio of Trading Strategies is a very well discussed issue – simply check how many posts you can find on AB newsletter – and everyone is approaching it from a different angle and with different solutions.

Unfortunately it sounds like a never ending story that’s why most people tend to use an equally diversified approach where they divide the overall account into as many sub-accounts as the number of trading strategies trading a single system each. Of course I believe this is the less efficient solution but it let you start trading without going too much into exoteric issues.

Some solutions I have been considering are:

1. using volatility-adjusted sub-accounts: let’s say you are trading 2 systems only with the first having a volatility double then the second. You would then trade 1/3 of the account for system 1 and 2/3 for system 2.

2. trading a portfolio of systems like you would be trading a portfolio of single stocks: you should then have a ranking/scoring algorithm selecting the best systems/signals to be traded when money is not enough to trade them all and select the most promising ones.

3. using an approach like the SOTM http://marketsci.wordpress.com/state-of-the-market/ where you could combine different signals (ideally long/mid/short-term) into one with a weighted average (maybe using a system peformance stat as weight).

Having said that I believe the pragmatic approach is still the best one: when combining systems into a portfolio you are actually creating a new system of itself which require to be evaluated as usual without too much wondering about the best ideal combination which may not apply to your specific portfolio.

I’m really looking forward to hear your thoughts,

Paolo

Hi Paolo, thank you for your reply!

You bring in some interesting approaches and my fingers are itching to further investigate these.

Inspired by your post, I am considering to take the 3 strategies I have posted and see if I can simulate the alternatives you are suggesting i.e:

1. Determine individual strategy performances

2. Fixed amount per strategy in subaccount and determine optimal weighting

3. Same as 2 but with rebalancing every x months

4. Volatility adjusted amounts per strategy in subaccount as “variable” variant of 2

5. Ranking/Sorting of systems (this is were my gut feeling was taking me)

I will post my progress as it gets in. Stay tuned.

The SOTM approach I tried in the past with adding signals of 10 different RSI(2) systems. The result was not very promising so that is why I left that trail.

As a seperate topic, you know that I am using Amibroker. As you mentioned the AB newsletter, I was wondering if you have any ideas on analysing a portfolio of systems in AB. Ideally I would like to be able to use trade by trade information when running Backtester in Amibroker. So e.g. switch off a system in backtest when recent x trades have not performed well and switch on later again. This seems not easy to backtest, let alone several systems at the same time…. any thoughts?

Michel

Michel,

glad you found it interesting.

As a side note to your “adding signals of 10 different RSI(2) systems” I would avoid combining signals from the same kind of indicator (see also http://marketsci.wordpress.com/2010/08/03/indicators-as-concepts/) favoring the combination of signals from different kind of indicators.

Regarding the analysis of a portfolio of systems in AB, I have a short list of the most relevant AB’s newsletter posts I can send you if you want – just drop me an email if so. There is also a Multiplex function by Paul Ho available at http://www.amibroker.com/members/library/detail.php?id=1227 but I’m ot very confident with it.

Unfortunately Systems portfolio testing is one of the few weeknesses of AB.

Btw the most accepted and promising solution is described here http://finance.groups.yahoo.com/group/amibroker/message/143966

but any custom money management (like your trade performane threshold) would probably have to be implemented in custom backtester code. It could be not required if using equity line logic (example at http://finance.groups.yahoo.com/group/amibroker/message/134556 by Howard Bandy).

Paolo

Hi Michel,

Same as Paolo, I find that your blog is very promising. Well done and Welcome to the world of trading-related blogging!

When I read about LSPM, one of my first thoughts is that it could be applied to a portfolio of strategies – which I think is a great way to gain diversification and reduced volatility. Haven’t gotten round to implementing/testing anything serious on that but I’ll be watching your progress with interest.

Another angle I’m keen to investigate is similar to the Adaptive Time-machine I suppose, but with trading regimes. Similar to the idea on David’s blog with monitoring of the volatility as a “system” filter, ie have a stable of systems to switch on/off (or variable allocation) based on a meta filter used as trading regime (volatility or else)

Jez

Hi Jez,

Thanks for visiting my Blog and your encouraging words!

I will keep you updated on LSPM implementation, I first need to get some more experience in using R

I like your idea on trading regimes….another interesting thread for researching…would I only have more time… 🙂

Have a nice weekend!

Michel

Allocating models across a portfolio is nicely explained in Rishi Narang’s book “inside the black box”. There are several approaches, each well summarized.

http://www.amazon.com/Inside-Black-Box-Quantitative-Trading/dp/0470432063/ref=sr_1_1?ie=UTF8&s=books&qid=1281367980&sr=8-1

Carl

Hey Carl, thanks for the tip. I have added the book to my reading list.

QD