Looking for patterns - page 36
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Attach a dash for trend filtering.
I meant the ATR calculation like Ma uses averaging. There it turns out that if there are large values at the end of the averaging period and then they leave the period, at that point the curve drops, although on the leading edge everything is calm. That is, those values no longer exist in reality, but they affect the behaviour of the indicator, which is wrong.
Chinky's suggested a good option here, you can use it as a trend detector as well. You could rework it a bit.
You are right about volatility.Good afternoon!
Once again, the original idea of the thread has gone a little sideways. It's no longer "Looking for... ", but creating ***. I understand if we create an EA for testing, i.e., we check the pattern and confirm its performance, or not. And we should check the next one, not create something further. In my humble opinion, the Expert Advisor should be created only after at least 20-30 patterns have been found. But I think that the Expert Advisor is no longer the subject of this branch - it will be something like "An innovative system of decision-making ..." ... :):):)
Regards, RomFil
P.S. And the ideas of regularity in pair trading? I have one very good idea for any timeframe. Ready to share with the condition of instrumental verification by an EA.
Romfil, you are right. Look at the situation. We have found a certain pattern in Rails, but like all other ideas, it works in one place and not in another. That is, we need to make some segmentation of the market by features: trend-flat, different amplitude and frequency. I want to make such a separator, then it will help us to test any strategy more deliberately.
We will continue testing different strategies. And as you suggested, we will make Expert Advisors based on indicators to test the strategy. Let's test your strategy.
About Genghis's idea. Look at the picture. If we make 2-3 instances of the indicator with different minimal distance and combine them into one, then we see small movements, but we also know in which global trend we are working.
That is, we are now creating a universal tool for checking ideas. And improving stops in the EA is also such a tool.
Vladimir, hello there!
I did not understand the question a bit. Regarding the timeframe, I am used to considering the daily movements, on H1 everything is visible, and you can test it quickly. The period of 20 years - I take the maximum time interval, excluding the daily movements on the hour timeframe till 1999. The source of quotations is a standard DC.
Spread the idea around, we can change something.
I partly agree, but in my opinion, stops and takeovers are the last thing to be played with. I could go on and on about strategic thinking and the right way to do things, but I think you can tell me more than I can... - so I won't talk about it. I just wanted to note that we have begun to deviate from the intended path ... :)
I can't tell you the idea in a few words. Tonight I will try to describe the greater part of it and post an indicator of the "idea".
Regards, RomFil
I just tried in the quotes archive in a regular brokerage house, after the history from Metaquotes the GBPUSD watchtips reached April 30, 2019. And that's it.
Tried it on H1 till 2015 and did not go any further. On higher timeframes the pound story goes back to 1992.
Just wanted to point out that they have started to stray from the intended path ... :)
What's the point of patterns if you don't try to analyse them in practice?
I agree with you specifically on the fact that not 1-2 parameters are required, but many. So I ask again what do you think about using big data in automated systems?Or are we talking about a different extraction method? Software?
Oh I see, it's a tambourine dance! I don't know what the gimmick is in this software, haven't figured out the intricate retrieval of quotes until the end. But try to download pentaminoes from Metaquotes server and you'll get the long history on H1 as well.
Then update the chart
What's the point of a pattern if you don't try to analyse it in practice?
And specifically regarding the fact that not 1-2 parameters are required, but many - here I am in agreement. So once again I ask what do you think about the use of big data in automated systems?There's endless talk about "big data" because of the different understandings of the term. There is no common understanding of the word "big" ... :)
But actually my personal opinion: it's not the amount of data that's important, it's the decision making system!!!
I have already answered this question in a private correspondence (you must have understood my phrase in a wrong context), I will repeat it here (an extract from the context of a dispute on one of the websites on the Internet):
Ok ... If you are interested, then let me give you my opinion:
1) Indicators are yesterday's century! The terminal screen is the main source of information. But what is on the screen? There may be zone indicators and so on.
2) My idea of the whole thing is as follows: (1) there is a graph with neural network predictors, some 200-300 of them, maybe more - these predictors on the history itself + with coming of each tick modify their readings by adjusting to the target function defined by AI itself; (2) all these predictors + a screen (not only one, but a time series of screens from different times and in depth) with indicators or without them are fed to the input of a multilayer very-very deep network (this is existing; the network must be large, about tens of millions of neurons); (3) then you set a criterion for the AI and it finds the best possible target function on the basis of inputs.
Well, that's it ... :):):) This is only in a general sense, but each element should include a whole complex of calculations and information processing.
In general I can say, that even now using deep neural network with 120 inputs and one output I can predict direction of next bar in 70-75% of cases. And to be honest it's not the limit, I can squeeze out a couple more percent. But this is only the direction. The magnitude of this direction can be tried to determine by the idea that I outlined a few pages of the thread in the past.
Regards, RomFil.
P.S. So this is the volume of data - is it "big" or is it still not enough?
Oh I see, it's a tambourine dance! I don't know what the trick is with this program, I haven't figured out the tricky part about getting quotes yet. But try to download pentameters from Metaquotes server and you'll get the long history on H1 as well.
Then update the chart
Here's another correction ... :)