Discussing the article: "Developing Trend Trading Strategies Using Machine Learning"

[MetaQuotes]

Check out the new article: Developing Trend Trading Strategies Using Machine Learning.

This study introduces a novel methodology for the development of trend-following trading strategies. This section describes the process of annotating training data and using it to train classifiers. This process yields fully operational trading systems designed to run on MetaTrader 5.

The primary difference between trend-following strategies and mean reversion strategies is that for trend-following strategies, precise identification of the current trend is crucial. For mean reversion strategies, it is sufficient that prices oscillate around a certain average value and frequently cross it. It can be said that these strategies are diametrically opposed. If mean reversion implies a high probability of a reversal in the price movement direction, then trend following implies a continuation of the current trend.

Currency pairs are often categorized as ranging (flat) or trending. Of course, this is a rather conditional classification, as both trends and consolidation zones can be present in either type. Here, the distinction is more based on how frequently they are in one state or the other. In this article, we will not conduct a detailed study of which instruments are truly trending. We will simply test the approach on the EURUSD currency pair, which is considered trending, as opposed to EURGBP, which was examined in the previous article as a ranging pair.

Author: dmitrievsky

[fxsaber]

Thanks for the article.

I assume that there is a transformation that turns a flat symbol into a trend symbol, and after reapplying it again into a flat symbol.

Perhaps it can be done like this (very rough draft):

1. Broken into ZigZags, getting ZZ[i] as the relative change between the corresponding vertices of the ZZ.
2. Created ZZ_new[i] = 1 / ZZ[i] and built an "opposite" symbol based on that.

If the hypothesis is true, then the problem of markup for training completely goes away.

As a consequence, the correctness of TC pops up.

[Sergey Chalyshev]

Bullshit.

Before you can teach, you have to learn yourself.

You have to learn how to make a profit in the market.

It is a futile endeavour to follow the mouse, it doesn't take much intelligence.

Therefore, there is nothing to torture the neural network and teach it simple mathematical calculations.

[Ivan Butko]

Sergey Chalyshev mathematical calculations.

What do you want from non-traders

As they say "rubbish in, rubbish out".

[[Deleted]]

fxsaber #:

Thank you for the article.

I assume that there is a transformation that turns a flat symbol into a trend symbol, and after reapplying it again into a flat symbol.

Perhaps it can be done like this (very rough draft):

1. Broken into ZigZags, getting ZZ[i] as the relative change between the corresponding vertices of the ZZ.
2. Created ZZ_new[i] = 1 / ZZ[i] and built an "opposite" symbol based on that.

If the hypothesis is true, then the problem of markup for training is completely gone.

As a consequence, the correctness of the TS pops up.

You are welcome. As far as I understood from Alexey's presentation, it is necessary to find such a transformation of the series through the quotes flips that it satisfies the rules of symmetric trading. Correct me if I'm wrong.

But I don't quite understand what dependence between the transformation and generalising ability (search for working patterns) is.

[[Deleted]]

Sergey Chalyshev mathematical calculations.

Offer a deal markup that requires a lot of intelligence. Since there is an understanding of how much intelligence is required.

I agree with the highlighted, that is exactly what we are doing. Although it's not obvious to many authors.

[[Deleted]]

Ivan Butko #:

What do you want from non-traders

As the saying goes "rubbish in, rubbish out"

Rubbish in the head is rubbish in the comments, by analogy.

[fxsaber]

Maxim Dmitrievsky #:

what is the relationship between transformation and generalisation ability (finding working patterns).

1. For example, you have good ways of marking a flat symbol.
2. Then, to mark a trend symbol, you need to invert it by transforming it into a flat one and apply point 1.
   
3. At the same time, invert the received signals to apply to the original symbol.
4. Thus, the trend symbol is marked by the flat algorithm.

[[Deleted]]

fxsaber #:

1. For example, you have good ways of marking a flat symbol.
2. Then to mark a trend symbol you need to invert it by transforming it into a flat symbol and apply step 1.
   
3. At the same time, invert the received signals to apply to the original symbol.
4. Thus, the trend symbol is marked by the flat algorithm.

I will conduct such experiments later, then I will draw up the results :)

[Ivan Butko]

Maxim Dmitrievsky #:
Rubbish in your head is rubbish in your comments, by analogy.

You are not a trader.

This is a fact.

Accept it, you do not know and can not know in principle how to trade and how to automate it accordingly. And writing for the sake of money will not make you a trader.

[[Deleted]]

Ivan Butko #:

You are not a trader.

This is a fact.

Accept it, you don't know and can't know in principle how to trade and how to automate it accordingly. Writing for the sake of money will not make you a trader.

All right, Teacher. Then write it yourself. But do not copy free indicators, but write it yourself :)