Some signs of the right TCs - 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
use something that is clear how it works
Alternatively, normalise the input data in the NS training examples
here is an example in the articlehttps://www.mql5.com/ru/articles/497(read section "Normalization of input data")
There are no questions about the normalization algorithms.
Imagine there are two characters with digits = 5. One varies by 0.1% per day, the other varies by 5%. It turns out that the second one should be scaled down 20 times to fit the first one.
To do this I had to calculate the amount of volatility, not how much the price of one digit differs from the price of the other.
There are no questions about the normalisation algorithms.
Suppose there are two characters with digits = 5. One changes by 0.1% per day, the other by 5%. It turns out that the second one should be decreased 20 times to fit the first one.
To do this you had to calculate the size of volatility, not how much the price of one digit differs from the price of the other.
Then you developed the good old ATR indicator
in my opinion this is a task for optimization with a GA tester
There are no questions about the normalisation algorithms.
Suppose there are two characters with digits = 5. One changes by 0.1% per day, the other by 5%. It turns out that the second one should be decreased 20 times to fit the first one.
For that I had to calculate the volatility, not how much the price of one digit differs from the price of the other.
That is correct.
And how else would you calculate those same percentages? ATR is what tells you "how much the price of one symbol differs from the price of another". In my EAs, all levels are based on ATR, which I consider the only correct way to determine the difference in price.
The ATR calculation procedure may be different. For example, you can simply take a large ATR period. You can take the ATR from a senior timeframe. You can take the ATR only for certain time intervals... But the point remains the same - to detect differences in prices of different symbols, you should calculate the difference of ATR of these symbols.
I don't think anyone will suggest anything better.
That's right.
How else can you calculate those percentages? The ATR shows "how much the price of one symbol differs from the price of another". In my Expert Advisors, all levels are tied to ATR, which I think is the only correct way to find out the difference in price.
The ATR calculation procedure may be different. For example, you can simply take a large ATR period. You can take the ATR from a senior timeframe. You can take the ATR only for certain time intervals... But the point remains the same - to detect differences in prices of different symbols, you should calculate the difference of ATR of these symbols.
I don't think anyone will suggest anything better.
And the symbols are not just from forex, but of any nature - crypto, indices, futures, etc. Which normalisation rule should I choose?
To align heterogeneous numbers, use the logarithm.
Then if required, normalise using different methods.
Imagine that there are two characters with digits = 5. One changes by 0.1% per day, the other by 5%. It turns out that the second needs to be reduced by a factor of 20 to fit the first.
No, it doesn't. Compare the TS, not the tools.
Yeah, good luck will be needed...
"I'll put all the products in the code base." - another black box... Figuring out the IDEA, even with the source code, is a very difficult task... And using a strategy you don't understand is a waste of time!
You need to figure out which factor to multiply an arbitrary TsVP by in order to set, for example, digits = 5 with the same properties.
As an example, here are two characters with digits = 5.
Clearly EURSEK needs to be dominated by a number much smaller than one to get similar properties to AUDCHF with the same digits.
But if there is no AUDCHF as a reference point? And the symbols are not just from forex, but of any nature - crypto, indices, futures, etc. Which normalization rule to choose?
There is no single answer, each option has its own +/-
As options, divide into:
RMS,
RMS from price difference and regression,
RMS of increments.
There are no questions about the normalisation algorithms.
Suppose there are two characters with digits = 5. One changes by 0.1% per day, the other by 5%. It turns out that the second one should be decreased 20 times to fit the first one.
I had to calculate the volatility, not how much the price of one digit differs from the price of the other.
It's probably too late now, but it may still come in handy.
It is better to use scales in dollars rather than absolute prices. Ultimately we are interested in profit in dollars per dollar spent. That's why I've moved to analysis in dollars, and it is very convenient to compare instruments that way.
For example, take GBPUSD 5 digits, USDJPY 3 digits and AMD 2 digits. If we express the value in dollars, it will look as follows:
Suppose we trade an amount of 10000$.
Then for GBPUSD 10000/open*close-10000=candel price
For USDJPY 10000*open/close-10000=candel price
For AMD the formula is the same as for GBPUSD.
Then we get the average size of a daily candle in 30 days for GBPUSD=32.25; for USDJPY=31.45; for AMD=225.97. Now we can say that if we use the leverage on currencies 225.97/32.25=7, then we can count on profit, like on AMD without leverage. This is how easy it is to compare the volatility of assets.
This method naturally has more subtleties.