Machine learning in trading: theory, models, practice and algo-trading - page 1791

 
Selling books on forex, inexpensive))
 
Maxim Dmitrievsky:
Sell forex books, cheap))

Hahahahahah )))) what a bastard))


I have an idea, tell me, is it hard to write a platform in Python? No trading, just charts to watch it comfortably, to interact with them and so on

 
Maxim Dmitrievsky:
Sell forex books, cheap)
ahahahaha
 
mytarmailS:

Hahahahahah )))) this is some kind of a shocker))


I have an idea, is it hard to write a platform in python? No trading, just charts to watch it comfortably, to interact with them etc.

The platform? I don't know, I just do data analysis. It's not hard, but why?
 
By the way, these are very good books. Pure empiricism. There you can get ideas on how to work with BP for MO. I was just kidding about the sale, of course.)
 
Maxim Dmitrievsky:
only why

If you want to teach AMO interactively, just point entry points right on the chart and let it learn, the same way you can punish, there are also cool ideas how to use it, you can even make a product for all traders.

Maxim Dmitrievsky:
Just kidding, of course)

Yes, I thought so ;)

 
mytarmailS:

The same way we can punish them. There are also great ideas how to use it, we can even make a product for all traders.

Yes, I think so ;)

Well it will be an open source product) if to make it for web at once. I don't know, I think it will take a lot of effort.
 
Valeriy Yastremskiy:

The question then is how to choose the right model and then the meaningful parameters.

Obviously, it depends on what you need from the model and how well it does it. The same model can be used in different ways and its correctness may depend on the way it is used. For example, take the aforementioned Ornstein-Uhlenbeck, with which you can:

1) Look for price plots well described by it and use the model to predict prices.

2) One can recall that this model is often used to describe a random fluctuation in price near some equilibrium level. To do this, we need to rewrite it in the form:

p(n)-p0=a*(p(n-1)-p0)+d*e(n), where the parameter p0 means this equilibrium price. On the other hand, the equilibrium price can be calculated based on some fundamental considerations and try to trade the divergence-convergence of these two equilibrium prices (if there is one).

Valeriy Yastremskiy:

And SB (pseudo) doesn't really fit my logic, the Wiener process with discrete time, then. It looks more like Brownian motion).

Here we should start with "correcting names", which Confucius said)

SB and Wiener process are very similar concepts (mathematical models). The essential difference between them is in the device of time: for SB it is discrete, and for the vinor process it is continuous. At the same time, from a wint can be obtained SB by taking it at discrete moments, while a wint can be obtained from SB by a limiting transition.

Brownian motion is a real physical process to which many different mathematical models can correspond. Confusion arises because these models are also called Brownian motion.

 
mytarmailS:

I had to learn eexcel and round off the balance sheet data, and I was stupidly unable to read the file from R, I must have killed an hour to spit and fix it in eexcel

At least we discovered something new :) In fact, this is the format of the standard report, and I have not changed it. In general, if you don't know Excel well, you could just replace everything in Notepad :)

mytarmailS:

This is what I got

You can check if everything is OK, at least by eye, I hope so ))

Balance chart is different - a little too much work.

mytarmailS:

I had a cup of coffee and thought, interpolating is not right, there may be a strong bias if the TS has not traded for a long time, damn, I need ..... to count the balance on each candle, as in reality, otherwise there is no way, otherwise unrealistic cuts will be....

Can you count the balance on each candle? Because I'm not in the mood for coding and thinking :)

So I originally saw it, that you need dates of opening and closing positions. It turned out that in my strategy, I have a strange additional order opening and then closing it relatively quickly :) I.e. the volume happens to be two lots, which also complicates everything. So, yes, I can just keep the balance on each new bar.

Added balance with date, the situation at the time of opening a new bar. If on the last bar the column "Type" was "Buy" or "Sell", and on the current bar it is "NONE", it means that the position was closed completely on the last bar, and may be only partially closed, then only the figure in the column "Lot" will change.

Files:
Balans.zip  1040 kb
 
Aleksey Nikolayev:

Obviously, it depends on what you need from the model and how well it does it. The same model can be used in different ways and its correctness may depend on how it is applied. For example, take the aforementioned Ornstein-Uhlenbeck, with which you can:

1) Look for price plots well described by it and use the model to predict prices.

2) We can recall that this model is often used to describe a random fluctuation in price near some equilibrium level. To do this, we need to rewrite it in the form:

p(n)-p0=a*(p(n-1)-p0)+d*e(n), where the parameter p0 means this equilibrium price. On the other hand, the equilibrium price can be calculated based on some fundamental considerations and try to trade the divergence-convergence of these two equilibrium prices (if there is one).

Here we need to start with the "correction of names" that Confucius said)

SB and the Wiener process are very similar concepts (mathematical models). The essential difference between them is in the device of time: for SB it is discrete, and for vinor process it is continuous. At the same time, from a wint can be obtained SB by taking it at discrete moments, while a wint can be obtained from SB by a limiting transition.

Brownian motion is a real physical process to which many different mathematical models can correspond. The confusion arises because these models are also called Brownian motion.

cps. Interesting thought. Select/separate sections of a series by the degree to which the model describes the section. But how can we tell at a glance whether the model describes the series well or poorly? You can't get the correlation at a glance. But there is something in it. The question / task is not in the prediction, but in changing the behavior of the series.

Terms and their unambiguity make life easier)))) I have SB initially in the range from minus to plus infinity in infinite time and only then the rules. The Wiener one was immediately in the rules)))) apparently that's why it is closer.))