Creating a trading robot - page 19

New comment
 
Alexander_K2:

Uh-huh. The indicator is good. If that's what I think it is - it easily defeats a stochastic process, when the sum of many random variables gives a Gaussian distribution.

However, you need to add a second indicator (or maybe you have 2 combined into 1? I don't get it) - something like process entropy or kurtosis of returnee distribution - and everything will be fine.

There are still a lot of mysteries in the process. I think. I can't find my companion.

 
Georgiy Merts:

Who will do all this? Where is the blueprint? Or at least an outline of the code?

The swan, the crab and the pike

I could do it myself, but I have enough to do. Shared a preliminary plan).
 
Uladzimir Izerski:

Many more mysteries lie in the process. I think. I can't find a companion.

From experience - if you start your own branch and get people interested, you'll get offers of cooperation. Do not just repeat my mistake - the absolute unwillingness to work together, and a desire only to teach :)) - And all will be OK.

 
Alexander_K2:

From experience, if you start your own thread and get people interested, you'll get offers of cooperation. Do not just repeat my mistakes - a complete unwillingness to work together, and a desire only to teach :)) - And everything will be OK.

This thread is a great place to express your thoughts and ideas.

I believe that it will result in a very interesting project.

I hope for a long life of this branch, where we can discuss trading issues. Unfortunately, there were no such threads.

 
Uladzimir Izerski:

This thread is a great place to express your thoughts and ideas.

I believe it will result in a very interesting project.

I hope for a long life of this branch, where one can discuss trading questions. Unfortunately, there were no such threads.

Yes, we will.

When it comes to stochastic dynamics, probability distributions - I will help.

 
Georgiy Merts:

Erm... So, a 'trend-following' one is one that 'follows the price'??? Well, then almost any TS follows price changes and is a "trend follower". The only exceptions are "purely temporary" TS, the essence of which is to open at a certain moment in time, and close after a certain time interval, without looking at the price. The vast majority, almost all TS are "trend following".

But I don't see many people getting rich on their TS...

Yes, exactly "trend following". What may follow - see, for example, "Michael Covell. Trading by Trends. How to make money by observing market trends",

as well as the attached files below. The obtained results have not yet reached their potential, but as they say, there was a beginning, and the continuation will follow!

 
aleger:

Yes, exactly 'trend following'. For what might follow, see, for example, "Michael Covell. Trend Trading. How to make money by observing market trends",

as well as the attached files below. The obtained results have not yet reached a potential, but as they say, there was a beginning, and the continuation will follow!

It is good to draw on the experience of others.

But in order to push the suffering people we must apply new technologies. They exist. You have to know how to find them.

 
Alexander_K2:

I'll do it.

When it comes to stochastic dynamics, probability distributions, I will help.

There is no point in dwelling only on stochastic dynamics. It's a helper, but it's not a cure-all.)

 
Aliaksandr Hryshyn:
Trend following means trend following. Fine, let it be Zigzag, I don't mind.
1. we will consider a trend to be a price movement within one ZigZag knee of 1000 points (at your discretion). Parameters should be chosen accordingly (to your taste). We will indicate a trend: 1 - more than 1000 points up to the end of the current knee or the current knee is ending downwards not more than 200 points, and the next knee has the size from 1000 points; -1 - downwards movement, all opposite to the previous one; 0 - no trend, which does not fit into the previous conditions. This is something we should learn.
2) We will also analyze the market condition using the ZigZag indicator, as desired by workers. We will take several bends: their height (points), how many bars they use, ratios, differences of neighboring bends and whatever else we want. We may use several Zigzags with different parameters.
3. Preparation of data for training of the model on trend forecasting. Form a text file, it will be lines with patterns for training, each line will contain data from step 2 and step 1 (numbers -1, 0, 1), more time can be added for further verification of the correctness of all operations performed. The data from step 1 should be saved as they are in the history. The data from step 2, you should only take through the strategy tester, it will eliminate the influence of re-rendering on the result! From every bar the data should not be taken, do skip from 1/2 average number of bars for the Zigzag knee from step 1. You can adjust the number of bars, so that the number of each of the numbers from step 1 (-1,0,1) is approximately similar.
4. train the model using Yandex's Catboost algorithm, we need classification, Logloss metric will do. Don't forget to divide the set of lines into training and test ones, and a test sample can also be used to verify the method.
5. If everything is satisfactory, integrate the model in the Expert Advisor. We attach some strategy to the trend forecast.
6. Test it, check it, then according to the situation.

This is a rough plan without getting into various minor details.

It is desirable to adhere to the long-established terminology. The trend is a set of Zigzag knees, and the Zigzag knee is a wave.

The wave does not depend on the number of points. The question is the time range in question.

I already have the models ready for analysis.

Can you analyse several TFs at the same time?

I have an interesting idea, but I do not have enough experience in programming. (Don't think that I am complaining or that I want a freebie)).

 
aleger:

Yes, exactly 'trend following'. For what might follow, see, for example, "Michael Covell. Trend Trading. How to make money by observing market trends",

as well as the attached files below. The obtained results have not yet reached a potential, but as they say, there was a beginning, and the continuation will follow!

My friend, what kind of files did you display? Test in MT4 on five minutes??? With an average profit 0.0006 and losses - 0.0017 per trade???

Laughter in the room. You should also suggest a test on ticks.

If you do not know what timeframe you are using it for, you may use it for analysis. Better - on D1. With an average profit and loss not less than 0.005 per trade (moreover, it is better that the profit is not less than the loss). In other conditions the test on MT4 does not reflect the trading character at all.

Transfer this plummer to MT5, and run at least in OHLC 1M mode, or better yet "all ticks based on real ones" - and see the difference.

If that's the "beginning", it's clearly the "beginning of the end".

Because, again - from your definition, it appears that ANYTHING, even the most plum monkey, is also a "trend tracker". And the expert you propose is also a trend tracker, but it's of no use.

***

Case in point:

- Vovochka, I'm going to the shop, follow the fish on the table. (gone)

(returns, meets Vovochka on the street)

- Vovochka, I told you to watch the fish!

- There's nothing to watch out for!

- ???

- I was watching the fish very carefully, but then the cat came and ate it. I was watching it as long as the fish was still there... but now it's gone, so there's nothing to watch anymore.

***

This is the case with these "trend-following" systems.

1...121314151617181920212223242526...32
New comment