CFDs are complex instruments and come with a high risk of losing money rapidly due to leverage. -- % of retail investor accounts lose money when trading CFDs with this provider. You should consider whether you understand how CFDs work and whether you can afford to take the high risk of losing your money.

How to find uncorrelated Darwins easy - not only for newbies

Today I finished the setup of my new real portfolio. I used the excel sheet used for the investing strategies presented here and extended it with monthly porfolio sums. The sheet now contains 10 Darwins where I selected 5 for an investment.

When I calculated the sums about the strategy results of all 10 Darwins I was very pleased to find only 3 losing months out of 28, the worst month September 2018 showed losses of about 135 €, which is only 1.35 %.

Then I calculated it for the 5 selected Darwins and was disappointed to see 7 losing months, which is 25 % of all months. Even if the worst losing month showed only moderate 143 € (2.8 %) losses, I was looking for an improvement. I didn’t want to buy 10 Darwins, I wanted to stay at 5.

How can I find an uncorrelated Darwin?
The worst month was February 2019, so I looked to the other 5 Darwins in the Excel for the one which got the highest profit in this month. That was planned for all losing months to find a candidate which could cut the (moderate) DD of my portfolio.
The candidate with the highest profit in this month replaced the 5th from my first selection and the amazing result was that now I saw only 4 losing months out of 28 and also the profit (total return without paying fees) of all 5 was even more than 10 % higher than before and higher than with all 10 Darwins.


Why did I look for uncorrelation?
Years ago I learned to know a hedge fund manger who calculated correlation for each asset or strategy before he bought it for his fund.

Uncorrelation should smooth the drawdown of a portfolio. A typical correlation matrix from a portfolio generated from the tops of a filter implies more a higher correlation because all Darwins have to meet the same criteria.
Darwinex shows it only for portfolios, a typical one looks like the one of my demo portfolio:

As you see, there is no significant uncorrelation found as most numbers are positive.

The mathematical formula of Darwinex confirmed my idea that the replacement Darwin is really uncorrelated with at least two other Darwins:

And today I could also see that uncorrelation makes sense on the portfolio results. While the Darwins from my first selection showed more boring or even results (it is a holiday in many countries today), the uncorrelated one showed nice profit:

That’s why professional investors are looking for uncorrelated investments.


The real problem is that you don’t have much data to rely on. When data is not enough correlation can be sought through a quality analisys: a darwin with a trend approach trading the main pairs, a darwin with a countertrend approach trading also the main pairs.
The fact that in a particular month the Darwin A was uncorrelated with the Darwin B could just be a lucky occurrence.


Generally you can be right. Correlation of the pairs mentioned is also to regard. The you are using a scientific method which might not improve the results.

Darwinex looks only on correlation of the profit Chart, regardless of the traded assets, if I got it right.

In my case as described above the replacement eliminated the DD of 3 months out of 7 of 28, not only of one. The lucky occurrence here was more to have it in my analysis list because it would be hard to find it or a similar one without this list.

1 Like

Stating uncorrelation is the same as stating alpha.
Here you are considering 28 months so 28 points in you dataset.
Even when you have alpha the great part of a return chart is due to randomness and you need yearS to detect the alpha, the same for uncorrelation.
Ray Dalio studied uncorrelation for his All Weather Portfolio composed by stocks bonds and gold, he used 100 years of data.


Sorry, no, because correlation changes much faster and with a monthly review the short term period is much more important. The prediction period is one month and in the review correlation must be checked again.

With a 100 years period you might have 3 years tolerance before action and in 3 years a Forex account can be blown out easily if you don’t act.

As long as you only look at the survivors of the past back from now , you ignore that on start of your period there were much more candidates with the same promising outlook where significantly more than 50% are not in your survivor list.
That’s how investors lose their money and never understand why. You can’t buy the past success.

1 Like

I don’t want to change your mind and I don’t think I can even be able to… but I must insist that this is not a right way to be sure that the reduction of the DD in those three months was a real uncorrelation event or just a random occurency. Correlation without causality it’s very dangerous.
Try this…
link funny correlation


So you think it is very dangerous if Darwinex looks only for correlation on the two profit chart?

Permanent winners are always uncorrelated to permanent losers besides there is no causality but the results itself.
If winning or losing is not permanent the correlation will vary.

My approach is just a practical one looking only on results which are checked every month.

What I think is that using correlation is the first step, after you have found which darwins “uncorrelates” best, you must find WHY , second step. Third step is watching things going and always asking if the “why” is still valid.

1 Like

Even without looking for the reasons of the correlation the mathematical result is flawed.

On the other topic we were discussing about hundreds or thousands of results, here we have 28 measurements.
It is difficult to have thousands of values but in general every statistc with less than 100 has no value.

Investors lose money because they give significance to something that hasn’t: return of short trackrecords.

1 Like

Indeed, but if you know the “why” even a small sample has more value that a very long sample without a why


They also lose with long track records as they wait and wait before they go investing. I‘m sure there is more money burned on Darwins with long than with shorter track records.
Again: you can never buy the past.

1 Like

People burn money with everything, even with KO that has 100 years of trackrecord.
The point is not how to lose money but if a calculated number has some relevance.
A monthly correlation calculated on 28 months has a very high error.
The past is not so significant but the first step to use the past to forcast the future is a significant sample.

I just exactly do that and nothing else. :grinning:

The main point why it is not proved is not mentioned:
I don’t know whether there was no correlation in January 2019 or an uncorrelation, as the relevant review would have been end of January to define which Darwin was uncorrelated.

So nobody knows whether it would have been selected as uncorrelated, but the same is with the other 4 Darwins that it is not known whether they would have been selected.

The correlation is not even known for the 31st of March this year, which would have been the review date for the April composition.

After today the Darwinex correlation matrix show the following values:

Still uncorrelated, but not as far or extreme as yesterday.

1 Like