1 (edited by zpoken 2015-06-27 21:12:54)

Topic: Combining several strategies, Curvefitting?

A question for anyone. I am always avoiding too many logic slots so I will not do curvefitting. But I like to have many strategies and then combine in the Strategy Portfolio" to do good combinations. Would you consider that as a way of curvefitting too? I know there is a lot of if´s, because you can combine in a lot of different ways. One way to argue is that combining many separate strategies will give you a better edge. I dont think there is a simple answer, but it would be interesting to hear your opinion.

Re: Combining several strategies, Curvefitting?

Hi zpocken,
less slots is no garantee for  no curvefitting.
Curvefitting exists all time.

To avoid curvefitting it is good to make Robustnesstest and Tests with different currency pairs.
Only one or two slots.

You can combine strategies in a portfolio. But it will not better if the single strategies are curvefitted.

thomas

Re: Combining several strategies, Curvefitting?

Hi Tnickel
yes, you are right. Curvefitting is there all the time and I meant that I use as few slots as possible to hopefully find more robust Strategies.

I was twisting my head about if I find one very robust strategy with 8 slots (probably impossible) and I compare with running 8 separate strategies with just 1 slot where each one of them are robust (all in the same instrument).

I think you gave me the answer already. It all comes back to robustness, and using one or two slots.

Re: Combining several strategies, Curvefitting?

I suggest that you attempt to use 'Multi Markets' on simple or complex strategies to see what happens with whatever you develop.

Both may  pass or fail the test.

Re: Combining several strategies, Curvefitting?

zpoken
And the most importend test is to check the strategy with unseen data.
You can generate from 2006-2012 and check with data from 2012-2015.

thomas

Re: Combining several strategies, Curvefitting?

And the most importend test is to check the strategy with unseen data.
You can generate from 2006-2012 and check with data from 2012-2015.

The Generator's OOS does exactly that.

Re: Combining several strategies, Curvefitting?

@ zpoken - to your 1st post:

I proceed in a similar way as you. I trade a portfolio of several simpler strategies,  assembled together with the help of Strategy Portfolio. This is a very important tool for me that I use all the time, even if it currently works only in a basic manual/visual mode. In the "drawer" I have a cache of several tens/hundreds of (mostly generated) simpler strategies. These are used when I need to replace a problematic strategy in the trading portfolio or when I need to add another strategy (when the account equity grows). Thus you can perfectly fine tune your overall RM/MM.

The weak point of the current state of the current Strategy Portfolio is that you need to search for fitting portfolio of strategies manualy/visually, which is often rather exhausting.

For this reason I wish Mr. Popov will be willing to add the "automated" Strategy Portfolio feature in next FSB releases. This would allow to generate a Repository of optimal portfolios with required RM/MM parameters.

Re: Combining several strategies, Curvefitting?

Still an example from a Czech server, how even several weaker individual strategies can optimally corelate and create a tradable portfolio when put together:

http://www.financnik.cz/komodity/zkusen … rijmy.html

You need not read the text, just see the 3 examples of equity curves - first three are always individual strategies and bellow is the resulting portfolio. Most extreme is the last example.

Re: Combining several strategies, Curvefitting?

Bold,
the examples on the Czech server,  is it backtesting only (I dont understand Czech)?
Anyone have live experience from running several strategies that have been combined  in the portfolio or in a similar way?  Is it a good way of thinking?

Re: Combining several strategies, Curvefitting?

I do have a live experience with running several (from 2 up to 30!) strategies simultaneously. See e.g. here:

http://forexsb.com/forum/post/30120/#p30120

From here I can write about existing strengths and weaknesses.

The examples from my previous post should mainly stress my opinion that the automation of creating the Strategy Portfolio would be much helpful in finding a really well diversified and robust trading system.

If you have a look at the last example before the end of the article, I bet that the majority of traders would never trade any of the 3 individual strategies. But still, if put together, they were able to create an agreeable equity of the whole portfolio. Now imagine that for example the 1st ugly equity represents a trendfollowing system, the 2nd ugly one a breakout (or false-breakout) system and the 3rd ugly one a system for choppy markets with lower volatility. Like in real markets - no individual strategy brings profits all the time; every strategy has  its strong and weak periods. But when you put them together you can achieve a perfectly diversified and robust overall strategy.

Now back to FSB. Most of the time I have used the Generator for searching new profitable strategies. I run it many times, with different settings. Thus I obtain hundreds of individual strategies. In order to be able to work with them in a more detail later, I need to strongly reduce their number first. You can imagine that I only keep strategies with "nice" reasonable equities; the rest of strategies with "ugly" equities are thrown away.

Theoretically this proces has sense = if you take several nice strategies and make a well diversified portfolio by puting them together, you should obtain a robust winning overall system. Unfortunately, my experience says that this is valid only in backtests, but not necessarily also in live trading.

Therefore I now tend to combine strategies generated by FSB with "nice" equities with clearly defined hand-made strategies that often show rather "ugly" equities, to create a more robust overall trading strategy. Here I would see the big advantage of the automated creation of Strategy Portfolio - a human brain and restricted time for work cannot compete with powerful computer processors.

Re: Combining several strategies, Curvefitting?

I have been doing a bit like you building strategies in three groups (Trend, Consolidating and Breakout) but never been able to run them long enough to know if they have been working good together.

Re: Combining several strategies, Curvefitting?

I use the combination approach, I am not skilled enopugh to make pefect strategies, I look for some good numbers in a strategy and do not worry too much about all of the metrics......

I try to have many strategies trading and the combined equity line is where the action is..... if there is an upward slope, I am quite pleased.

For some reason, I find that I have a ton to learn re strategy development...  and I really appreciate your discussion in this thread..

Happy Weekend!!