Create and Test Forex Strategies
You are not logged in. Please login or register.
Forex Software → Expert Advisor Studio → Robustness - Multi Charts
I am unclear as to what is best or what is most suitable for settings re number of pairs in Multi Charts.
If we have three pairs..... that is a certain degree of robustness and 4 pairs is another degree.....
But what is satisfactory?
Does anyone have any research..?
Thank you
Ha.... 
I was pondering the same question, thought to myself that Dave must have an answer for that.
I used 4 additional pairs with my newly uploaded data, it didn't have no results for the strats it found. I did widen the data amount as well, it really shows a difference, much harder to pull the cat out of the bag.
What I'm thinking is that we should look at the correlation between pairs, there's no significance in confirming the strat on a highly correlated pair as it would pass it easily, wouldn't it?
Someone more knowledgeable in math and stats is very welcome to chime in.
I just have no idea...... That is why I asked in here.
I think that Strategy Quant attempts to pair EurUSD and GbpUSD when they make their strategies..... at least from what I have seen in the forum there.
For some reason EA Studio has set 3 as the default and I am thinking that is sufficient.
I am running 3 instances over night and I have those set at 4.
I am running 60 minute charts on one, 75,000 bars...... if I get anything to pass the 4, I will think that to be excellent. I probably should have left it at 3.
Maybe Miroslav has some ideas about this subject, I hope so. for some reason he set the default at 3. maybe that is good enough.
A lot of what we hear is just some sort of rumour that is repeated so often that it becomes established fact. I have searched on the net re robustness and curve fitting and not seen a definition as to what establishes robustness.
I would like to see some sort of arithmetic proof. LOL
I had one run overnight with a few results where 3 of 5 were good......
I added a few more pairs and then 5 of 9 were good...
So..... a bit of this depends on how many pairs are being examined as well as the correlation...
Strategy Quant combines two or three pairs in their strategy development to get robustness.. and I have done that in MT5.
I like how you can test many pairs together in one strategy in MT5, like you can have 12 at a time.... but unless you are a very skilled programmer, they all execute the same trades and you lose out on a lot of entries.
From what I have seen....... by combining pairs in the development of the strategy, you lose out on trades as there is juggling to get them all the same trades......
At this time, I prefer how EA Studio works.... make a strategy for each pair and see what happens against a group because that allows more trades per pair.
(I know that my enthusiasm for EA Studio is hard to see LOL)
Here is a discussion that may be of interest
I have this feeling that the way EA Studio processes the strategy with the validation at each step and the final Multi Charts is doing exactly what we need.
The proviso being that we watch what happens and keep pruning the portfolio and adding new strategies as time passes.
I am not so sure that we will find a strategy that will live forever, we have to expect some losers and we have to expect that performance will not always live up to test results.
The only protection is doing the due diligence on a daily basis and attempt to improve as time passes.
I just finished a simple test.
What I did: generated a bunch of strategies on 20k bars, selected out most interesting ones, then put them on unseen data (amounting to 25% of generated data, 5k bars) and evaluated and categorized them further.
I ended up with 8 strats, which passed my unseen data test, and 9 strats, which failed on unseen data.
Then I ran them through Multi Markets to see what MM tells me about those strats.
I used 5 pairs, 4 in addition to 1 base pair.
Most important note is that they all failed to pass MM with flying colours.
Basic stats below (how many markets failed):
Certainly, sample size is small to have meaningful conclusions, but still a little insight.
To have a rule that a strat has to pass all markets, means there would be no strats to trade.
Allowing 1 market to fail, would net 2 tradeable and profitable strats.
Allowing 2 markets to fail, would net 8 tradeable strats, half of them unprofitable.
Nothing conclusive, but still worth thinking about.
I usually add as many pairs as I can to MM
Also, to a certain extent MM can be validated by using different time frames
I do not think we are going to be able to find what we want. something that will work in all cases.... it is still a crap shoot, but MM helps to cut down on the problems.
My thinking is that trading is like insurance...... use a huge number of strategies and the smallest possible risk for each, the law of numbers should look after us.
I was just thinking that the sample may have been a bit small and that if the time frame is fast that the bars could be a little bit different between old and new data.
Also, was there any optimization on the new data...
Sure, sample size is too small, but it does provide some info. I have been digging deeper and I'm seeing the same results.
New data is used only to divide the generated strats into profitable/unprofitable ones. That's how I can evaluate MM objectively, does it validate correctly and in what region number-wise.
I'm sounding like a party-crasher but it's not so.
I have had fairly good luck, but I use small stops. large stops have caused me grief.
Like, on a daily chart 85 or 90 pips tops.
I am unclear as to what is best or what is most suitable for settings re number of pairs in Multi Charts.
If we have three pairs..... that is a certain degree of robustness and 4 pairs is another degree.....
But what is satisfactory?
Does anyone have any research..?
Thank you
I used 4 in addition to the base pair, spectacular failures, to be honest. So I researched it a bit and made a selection of pairs with different correlation patterns in relation to the base pair. Much better! I hope though to find strats which pass the test, quite a mountain to climb...
I am going to run with three and see how that works out for a few months as I do not really know what is good.
I have many that work well from FSBPro with two or three and the requitements in EA Studio seem to be tougher so I am going to take a chance.
If they turn out to be really bad, then I will have to rethink.
Since recent updates I think that the Multi Markets is much more reliable as we can set pretty strict requirements with all of those variables.
There are not many that pass Multi Markets like 2700 out of 2,774,000.
Are you using the same criteria to pass multi-market tests as you do for your base pair, or are you loosening them a little? In my limited time using EA Studio, I have loosened the requirements, and still find very few strategies that can pass 3 pairs in MM (with 11 pairs tested).
I loosen them in Multi Markets.
I have no idea how long you run the EA Studio. for me to get some good results I usually leave it for 2 million turns...
I didn't see any numbers from anyone else.
I am still learning how to use the Studio, seems there is a lot to learn as to how to get items in the collection that are worthwhile.
If you make the requirements too tight.... then few results........ you can loosen the requirements and then use the filters to develop your collection.
Forex Software → Expert Advisor Studio → Robustness - Multi Charts
Powered by PunBB, supported by Informer Technologies, Inc.