EURUSD on the 1 hour chart

Hello,

The following is a strategy for the EURUSD on the 1 hour chart.

Opinions are very welcome.

Hey notch. Your bitter comments could be taken as unpleasants but your academic bias is amusing. Can’t get angry with you 🙂

However not fair to say there aren’t worthly contributions on the forum. Strongly disagree

I, for my part, have learned a lot from the forum so far and take a lot with me from this threat, which helps me e.g. to develop strategies in much shorter timeframes and to pay attention to the GMT in the Datamanager.

I think everyone should first touch their own nose bevore someone ask the otherones to change the behavior.

Personally, I really like the forum here and enjoy reading a threat from people who shares own experience.

Many thanks to all those who count themselves to this art of peoples and want to count themselves in the future!

Marcel wrote:

The reality is the community hasn’t made any progress in 4 to 5 years literally.  I think there may have been a small leap forward after Ilya (for some reason) introduced the power of shorter training periods with random MC and a few other basic work-flow innovations.   There has literally been zero progress apart from this ‘innovation’;  perhaps finally working out how to deploy 99 EAs at once can also be considered a step forward for many.

Yo Notch.

Well, I found that method to be an extremely helpful and profitable discovery for my own journey, so decided to share it with others so we can advance as a community. Also I want to thank you for introducing the basics to me about 6-7 months ago. I am profitable for 5 months straight and have no intention of stopping.

Though, I’ve grown to become quite disappointed in this community. Even though I’m the only member to upload a strategy to the “strategies” section of the site, still I received some private messages telling me things like “I ran your strategy for 1 month and it lost me money! it’s nonsense!”. Even if you disregard the fact this strategy has been profitable for me for 4 months straight now on 4 different symbols (though re-optimised for some of them), the amount of true willingness to help others and cooperate is very low. Seems most people are more interested in grabbing “fully working” strategies wherever they can, disrespecting others, and refuse to learn or progress.

Anyway, It’s a pleasure to read your posts as always.

Ilya

The reality is the community hasn’t made any progress in 4 to 5 years literally.  I think there may have been a small leap forward after Ilya (for some reason) introduced the power of shorter training periods with random MC and a few other basic work-flow innovations.   There has literally been zero progress apart from this ‘innovation’;  perhaps finally working out how to deploy 99 EAs at once can also be considered a step forward for many.

Yo Notch.

Well, I found that method to be an extremely helpful and profitable discovery for my own journey, so decided to share it with others so we can advance as a community. Also I want to thank you for introducing the basics to me about 6-7 months ago. I am profitable for 5 months straight and have no intention of stopping.

Though, I’ve grown to become quite disappointed in this community. Even though I’m the only member to upload a strategy to the “strategies” section of the site, still I received some private messages telling me things like “I ran your strategy for 1 month and it lost me money! it’s nonsense!”. Even if you disregard the fact this strategy has been profitable for me for 4 months straight now on 4 different symbols (though re-optimised for some of them), the amount of true willingness to help others and cooperate is very low. Seems most people are more interested in grabbing “fully working” strategies wherever they can, disrespecting others, and refuse to learn or progress.

Anyway, It’s a pleasure to read your posts as always.

Ilya

Hello, Ilya,

Thank you very much for your contribution.

I’m already planning to publish one of my strategies in the documentation section…..I’m just not sure which one at the moment 🙂
I would also use your text as a text template, if that would be ok for you?

About the quality of the community:

You have to keep the community in mind that it’s a comparatively very heavy program and although the manual says that it’s not a “holy grail” many people ignore that and think that after 3 days around experimenting they can immediately form an opinion and are experts (also in this threat you can see at least one of this…..pretty much at the beginning of the threat 🙂 )……but under these measures there are a few people who have understood that the program was only understood after years of hard work……and I count you, Hankeys and notch as well as some others among them.

But now the question is again in which quality these experts want to deal with each other? Do we want to tell each other that we know better than everyone else or do we want to participate from each other and help each other to what we all want?
SUCCESS!

hi Gianfranco……i use  3 oos for develop strategies…and about 15,16 years of history data and much more..but (my opinion, i am not professional)1 year(oos) with bad results in h1..in my workflow…..  dont pass

Gianfranco

I have a slightly different approach ( presently). It is good if they fail on unseen data if they can be corrected with Walk forward.  Funny thing is that bulk of strategies that usually pass a walk forward test aren’t that good on Unseen data in their original form but they can be really good if Walk forward is made. Even if we have a powerful way to make new strategies all the time this cant be the goal to always generate new strategies it must be better to have adaptable strategies that can be curve fitted to recent market conditions and work going forward with a high likely ness to be profitable which stats we actually get from % of profitable runs. I find using MC during build increase the passed % from Walk forward and since this also increase the generated strategies per hour due to that Walk forward is slower I find MC usable again. It is actually not that hard to find strategies that are adaptable if You ease up on original strategy ranking demands and instead put them on WFO_ OOS. After 4 weeks I now have strategies on 20 instruments on H1,H4 and D1.

Yes i do 4 Walk forward tests .1 simulated Exact during build. 2 . Simultated Exact on other data or instrument ( same data length  6 months periods)  3. Simiulated Exact on Original data  1 min resolution. 4 .Optimizer  Exact.

This topic has become a very interesting discussion, not exactly related to the original topic and totally hijacked by some pseudo-scientists over here…but anyhow, still very interesting, so here is my 5 cents. I do not care if you will agree of disagree with me, as I see it all, nobody here is right about trading, there are only people who make money in the markets and people who do not, and I get a feeling we are all just learning here. Additionally, knowing it “all” will not make you a profitable trader 😉

given the non-stationary properties of financial time-series, it is irrational to expect to find a long life strategy that has a linear equity curve for the past 20 years; I NEED my 20 year equity curves to be increasing over time but to also perform poorly at the appropriate locations; even the most naive algo’ builder should intuitively know that a strategy that has performed well through every economic shock, news surprise or flash crash generally must be a lemon, right?

1. When trying to find good long term strategies, I always try to keep in mind that markets have different “moods”, there is bullish, bearish and ranging/stagnating market and we have high, low and “normal” volatility (ATR). So during strategy generation I always try to split my IS and OOS over those different market conditions, therefore I use slightly longer data windows for strategy generation only. Doing so I assume that generated strategies, should be able to deal with all different market “moods”. So yes it is possible to generate a strategy that is doing well in many OOS years outside the original generation period, also in different market situations including financial crisis periods. However, I must admit that it is very difficult to tell if this is based on inherent stability of the generated strategy or just based on pure luck. And the reason is lack of statistical significance of results during those (relatively short) critical market periods. See my next point.

2. About WFO and automatic re-optimization: It is not a secret that the method of generating strategies using a relatively small data window and verifying using WFO-like approach on a long OOS period is commonly used by many profitable traders. Personally, I know people running their own hedge-funds based on this method (seriously).

But, when you do automatic re-optimization just keep in mind that e.g.: optimizing (curve-fitting) 5 parameters and selecting the best settings
having only 50 trades in your optimization period is not very statistical significant and will get you nowhere. So IMO, you need to make sure that:
A. Your system has very low number of parameters to optimize (= low complexity systems).
B. You have enough trades in your optimization period to be able to tell that the resulting outcome is statistically significant.

So I’m asking myself if the provided auto-optimization code-example provided by notch is really usable or just another piece of curve-fitting garbage….

P.S.: For most of you who do not know what statistical significance is, just google for few examples of “P-value” explanation. In pseudo-science based trading the rule of thumb is that your minimal number of trades should grater than 50 + n * number of optimized parameters.

3. And there is this another thing: the stability of selected parameters after (re)optimization. Ideally when doing a WFO or automatic re-optimization I would like to
first verify if my optimization (parameter range) will operate in a stable region of my entire optimization distribution. When you do your automatic re-optimization and you select the best parameter set based only on the best results without looking at total distribution, the selected parameter set can still be this one “special” (orphan) parameter-set far from the reality (where “real” system performance is close to the mean performance of the entire optimization spectrum). Up to now I did not find a suitable solution for this issue than doing everything manually. Which is boring…

I wonder how gurus over here a dealing with those issues.

Just for clarification, in the ‘rule-of-thumb’ equation:

50 + n * number of optimized parameters.

‘n’ as used in many examples is 50 or larger, so if you have a system with only 1 optimized parameter your results become significant after 100 trades. In other words, after 100 trades you can tell that there is a significant chance, that what you see in your result is not just due to pure luck.

I think it would be very helpful and educative for most of the readers here, if you could provide a practical A-to-Z example of how you are using your auto-optimization code and how you are dealing with the issues I mentioned earlier in order to get some results, any results..

P.S.: I do not care much about your trade screenshots and life-style comments, in most cases this triggers a “false narrative” waring to me.

Wow, 20 years in trading is indeed something, but I think in those 20 years you have missed two most important lessons:

1. Respect towards other traders, especially newbies (people that need help), people that are as green as you were 20 years ago. You must understand that 99% of people here do not even understand your posts, because they cannot follow it. So indeed you are totally right, in that case you are just rambling 🙂 LOL
2. Some basic human communication skills without attacking others?

I will happy to teach you 1 & 2 for free, no strings attached! Instead you will show us how to get the information out of 1 sample.
Information from one data point which is totally buried in noise. So far I know only a handful ways to get info below the noise levels, like e.g.:
– oversampling = increasing of the effective resolution
– modulating signal and switching to frequency domain

But, this cannot be successfully applied to trading (time series). So I’m (and all other people here) are really interested in your way of approaching this.

SQx have functions that allow You to test almost any type of way to find good strategies thus all features that have been implemented by requests from traders that I think combined have 500+ years of experience . You can even test statistical significance  which I have done and it has no impact either. Throw away the “fake” books and use SQx to find ways to improve stability from strategies going forward. Hankeys strategies with RDD around 20 is very good strategies. For evry dollar they lose they make 20. They are easy to find it just take different long time depending on what putors You have.

I see that cooperating, or even exchanging new or ‘other’ ideas will be very difficult here….on the other hand we can see people are achieving some good results using SQ (X). So maybe it is a good sign and everybody should keep doing things their own way.

“There’s more than one way to skin a cat”.

Moreover, sharing your best strategies in public is also not the best way to go…if you will share your best strategies here, people will sell them tomorrow on mql5 market using a different name 😉

So what’s left? Exchanging the strategies with other SQ users? In a 1-for-1, strategy-for-strategy manner? In that way we can at least stay diversified, which is good for our portfolios. This, while accepting the risk you will exchange your good strategy for a shitty one 😉

What do you thing about this idea?

1 word: Aspergers

Ok thanks for sharing this. This explains few things to me.