Algorithmic ''centrifuge'' - page 22

 
Aleksei Stepanenko:


All new is well forgotten old. In this thread too). I started to write in mql with one purpose - to implement classical thechanalysis. I couldn't find any understandable implementations and here we go) All strategies have been developed long ago, the profitability is not the point, but the way to apply them in a certain market in certain situation. So far any robot is inferior to a human in quality of solutions, because its field of view is limited, but a robot is more efficient, you have to use it and not try to reinvent the wheel, i.e. new strategies. Neural networks are lame and have not even been taught how to use them properly. But research in this field is interesting and it must be followed. If you look at what algorithmic funds earn, there are many arbitrage and elementary trend following strategies. As for arbitrage, interesting are such researches as Maxim's about seasonality. As for the trend - everything was described many years ago) if you don't get greedy, everything works. If you get greedy, you need to keep thinking))
Z.I. Personally, apart from the TA class, I now trust the simplest regression check. Nikolay Semko knows all about it) is very helpful.
 
Aleksei Stepanenko:
Peter, is there any way you can describe your thinking logically? I would like to highlight the basic criteria and ignore the fact that the strategy is not 100% accurate. If you are talking about proportions and percentages, between which values? Between which waves (knees)? Please draw some pictures to illustrate your idea more precisely. Because you should make some suppositions and assumptions when writing the code. Otherwise you may not get off the ground for a long time.

I'll go into more detail later when I've thought more carefully about the idea.

 
Wizard2018:

It is not possible in principle. It takes a long time to explain, there are fundamental reasons. There are two states in the market, yes. And it is very important to be able to identify/recognise them. But they are not trend-flat. On a trend-flat market and not only charts - for example SB :))), "divide", from lack of understanding - "what is going on". And then, having fallen into the trap of such "division", they try to bash their forehead through the wall next to the open door, separating one from the other. "On history" no problem, but here and now...

How impossible, if you look at the chart and separate the two freely. Another thing is that you are not aware of exactly how you are doing it. But, the criteria for distinguishing a trend from a flat is very simple. It is a difference in forms, proportions, ratios of values and nothing more. There is no mysticism. Another thing is that it can be problematic to program.

Anyway, the trend/float division is very convenient for trading. Trend - hold a position, flat - flip.

I repeat that the trend/float dynamics division is done on history. And it would be a shame not to solve this problem...

 
Wizard2018:

It is not possible in principle. It takes a long time to explain, there are fundamental reasons. There are two states in the market, yes. And it is very important to be able to identify/recognise them. But they are not trend-flat. On a trend-flat market and not only charts - for example SB :))), "divide", from lack of understanding - "what is going on". And then, having fallen into the trap of such "division", they try to bash their forehead through the wall next to the open door, separating one from the other. "On history" no problem, but here and now...

I'll cut in with a couple of lines) As I'm implementing the following paradigms of incompatible states:

Buy/Sell

Buy/Sell/Square

Buy/Sell/Square/Fog ... four lines have accumulated)

 
Aleksey Mavrin:
Everything new is well forgotten old.

Thank you, Alexey! Your thoughts are interesting.

About neural networks, I think it's similar. Neural networks are good at classifying intelligently fed data, but the problem is in the data, not in the networks. I have a question about trend following strategies, becauseforex market is more flat than trend following? And on the regression, is it for plotting channels?

Here a question has arisen about the separation of the trend and the flat, may be there are thoughts?


Tag Konow:

I repeat that the trend/float division is done on history. And it would be a shame not to solve this problem...

And I would like to share in the reality with the least amount of delay.

 
Wizard2018:

It is not possible in principle. It takes a long time to explain, there are fundamental reasons. There are two states in the market, yes. And it is very important to be able to identify/recognise them. But they are not trend-flat. On a trend-flat market and not only charts - for example SB :))), "divide", from lack of understanding - "what is going on". And then, having fallen into the trap of such "division", they try to bash their forehead through the wall next to the open door, separating one from the other. "On history" no problem, but here and now...

Well, and talk about it. Everything is more interesting than percentages and proportions...

 
Реter Konow:

ZS. Once again, the trend/float divide is what we do on history. And it would be a shame not to solve this problem...

There is no difficulty in determining what is trending or flat at the moment. The difficulty is in predicting a trend and a flat.

 
Nikolai Semko:

Determining whether a trend or a flat is in progress is not difficult. The difficulty is in predicting a trend or a flat.

At all?

 
Алексей Тарабанов:

At all?

For me, yes. Everything has been implemented for a long time.

if only this ancient indicator of mine:

https://www.mql5.com/ru/code/10882

 
Nikolai Semko:

For me, yes. Everything has been implemented a long time ago.

even if it is this ancient indicator of mine:

https://www.mql5.com/ru/code/10882

I agree completely. With a caveat - for a real TS construction we are even more interested in predicting the trend size than the trend itself, or in other words, in predicting the end of the trend.

Aleksei Stepanenko:

Thanks Alexey! Your thoughts are interesting.

About neural networks, I think it's similar. Neural networks are good at classifying reasonably presented data, but the problem is in the data, not in the networks. I have a question about trend following strategies, because forex market is more flat than trend following? And on the regression, is it for plotting channels?

Here a question arose about the separation of the trend and the flat, can you think?

And I would like to share on the fly in reality with as little delay as possible.

I'm extremely surprised. I always thought the matter of distinguishing between a trend and a flat is a cornerstone issue and everyone who starts writing MTS finds it difficult to decide for himself.

If the indicator described by Nikolay is 66%(!) accurate in saying whether the flat is over or not, what else do you need?

If you do not know anything about the indicator, you cannot do it properly, if you do not know anything about it at all. The volumetric analysis gives some advantage if it is properly prepared.

Therefore, in my analysis system I implement the following types of classical TA with a note on what I base them:

1. trend. fractals, regression.

2. Graphical. fractals

3. Candlestick.

4. Levels. fractals. volumes.

5. Volatility, regression

6. Fundamental switches. my brain.

By the way, no one has mentioned the graphical one here yet. The riddle from me, is the triangle a flat or a trend ? And the wedge with the flag ? :)

But till H1 it is clear that any TS yield in jumping 100% per year with moderate risks, I haven't reached it yet, I am working on it. I have to use only one indicator, i.e. i need two indicators, roughly two main indicators, and if i count fractals, i need two more indicators.

Who is looking for more profitability - come to the zoo to the monkeys, unless you are an HFT fund. They haven't come up with more than that yet, alas.

Neuronetworks don't know how many generations of development it takes to get to the human level, but when it happens everything will change so much, that the market most probably will be absolutely different, as it has changed with the advent of algo-trading, and it cannot help changing, because players make the market.

I wonder if anyone has tried to feed NS not just prices but fundamental data from around the world, news, tramp tweets, analyst predictions, etc. I think it's an interesting idea.