Discussion of article "Brute force approach to pattern search" - page 4
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
I think it's called a fractal in our trading science)?
Anyway, I mentioned zigzag there just to clarify my idea.
fractal too, zigzag fractal, it's just an attempt to identify some waves.)
At first I thought that the article is about bruteforce, and I read it because I myself thought slowly, how to bruteforce to identify all the patterns. I wrote in my article that if there are an infinite number of participants on the market or they have an infinite number of computing resources, then neither simple brute force method will reveal absolutely all regularities (of course, there is nothing infinite here, you can do with finite ones).
But then I realised that the article is not about that at all. The article is about what I do in my work. You write that a pattern is detected and exists for some time, then either persists or is inverted. The reasons for this is that you are analysing with a fixed window, a floating period and amplitude of a conditional sinusoid. In the figure below I have shown an example.
I have plotted a sinusoid with a random period and a random but increasing amplitude. Simplistically it is similar to what you write that the signal is periodic. Yes, it is periodic, but the period and amplitude float and not just float, but change almost randomly. Suppose we have a window for analysis as wide as in Figure 1. Then we have found the half-period very well and can trade on the reversal of the pattern. But if we use the same window to analyse further, we will get nonsense. Because the next half-period already consists of three windows and is also very noisy.
In order to analyse correctly, the window should be floating and adjusted to the current period in real time. I have attached a Gif, open it and see how it is visually done at my place. That is, there is always a certain pattern, we are looking for the period where it is.
At first I thought that the article is about bruteforce, and I read it because I myself thought slowly, how to bruteforce to identify all the patterns. And in my article I wrote that if the market has an infinite number of participants or they have an infinite number of computing resources, then neither simple brute force method will reveal absolutely all regularities (of course, there is nothing infinite here, you can do with finite ones).
But then I realised that the article is not about that at all. The article is about what I do in my work. You write that a pattern is detected and exists for some time, then either persists or is inverted. The reasons for this is that you are analysing with a fixed window, a floating period and amplitude of a conditional sinusoid. In the figure below I have shown an example.
I have plotted a sinusoid with a random period and a random but increasing amplitude. Simplistically it is similar to what you write that the signal is periodic. Yes, it is periodic, but the period and amplitude float and not just float, but change almost randomly. Suppose we have a window for analysis as wide as in Figure 1. Then we have found the half-period very well and can trade on the reversal of the pattern. But if we use the same window to analyse further, we will get nonsense. Because the next half-period already consists of three windows and is also very noisy.
In order to analyse correctly, the window should be floating and adjusted to the current period in real time. I have attached a Gif, open it and see how it is visually done at my place. That is, there is always a certain pattern, we are looking for the period where it is.
Here it is more accurate to say we are looking for the period of the strongest regularity, because as you said there is always a regularity, and not even one, but rather a certain large sandwich and layering these layers create the appearance of chaos. You have a fixed window anyway, you just try to find periodic processes in it. It still turns out that the final regularity is the regularity of the oscillation period of another regularity ) . You just move to a higher level ). So you can build fractal nesting dolls very deeply ) . This is also a good method, maybe even better than mine. Generally speaking, there is a lot of room for creativity ) . I proceeded from the fact that the straighter the line (balance curve), the second derivative of the number of transactions from the balance function tends to zero, given that we are trading with a fix lot ) and it is equal to zero where the middle of the next half-wave ) and this means the greater the chance of continuation in the near future or at least get a smoother reversal to be able to use reverse marting.
A little research shows that ugly pictures are quite enough. To do this, just look at a histogram plotted against the time-of-day sample for the tops of the zigzag. Pretty pictures should give troughs with high banks and bottom at zero level on this histogram.
The histogram below is plotted for EURUSD for the year 2020 for a zigzag with a 0.1% knee. There are troughs, but their bottoms are well above zero, suggesting quite a few ugly pictures when reversals happen "off the clock". If such bad days are also badly mixed with good days (which is quite possible), the result can be quite sad.
The histogram exactly repeats the typical intraday volatility pattern. That is, it doesn't seem to be true
You should get about the same from candlestick sizes or tick volumes. They are interrelated, zigzags, candles, volumes.
And since we know what to measure and the typical picture, we can evaluate in real time how much the market is "typical" right now (or reject dashing days in time), and what absolute values are expected.
You can even get your head out of your ass and (ahead of) the trend reversal by volatility changes.
histogram exactly repeats the typical intraday volatility pattern. So it doesn't seem to be true
You should get about the same from candlestick sizes or tick volumes. They are interrelated, zigzags, candles, volumes.
And since we know what to measure and the typical picture, we can assess in real time how much the market is "typical" right now (or reject dashing days in time), and what absolute values are expected.
You can even get your head out of your ass and (ahead of) the trend reversal by volatility changes.
There is a lot to think about. Although catching extremes is considered bad behaviour).
Maybe we should start a thread about seasonality, so as not to flood here.
The only regularity in the financial markets (in technical terms) is the elementary structure (M-shape),
which is constantly present on the chart, with no gaps in time,
and its dynamic trend (MSAD curve), which takes into account the actual frequency of this price fluctuation.
And this is on any scale, on any financial instrument.
Source: impulse equilibrium theory.
The only pattern in financial markets (in technical terms) is the elementary structure (M-shape),
which is constantly present on the chart, with no breaks in time,
and its dynamic trend (MSAD curve), which takes into account the actual frequency of this price fluctuation.
And this is on any scale, on any financial instrument.
Source: impulse equilibrium theory.
Alexander, your book is paid, that's one. There is not a single robot confirming the theory in the products, that's two. Only two indicators. If you are advertising your book here, it is not quite appropriate. If you have some theoretical information that will enable me or any other developer to make a system with positive mat expectation on the whole history, I will be glad to listen to you with pleasure. But for now it all looks like advertising, sorry. I'm not a fan of criticism myself, but here I can't help myself. If the information is too "classified" you can write to me in person, I will write a robot and give you if something there will be working.
Evgeniy Ilin, I can run your programme at myself according to your scenario to search for patterns.
...
In order to analyse properly, the window should be floating, and in real time adjusted to the current period. I have attached a Gif, open it and see how it is visually done by me. That is, there is always a certain pattern, we are looking for the period where it is.
How often do you look for a new dependency, do you determine that an old one has stopped working? Or do you use many dependencies at once, dropping out of the pool by metrics (or time)?
Evgeniy Ilin, I can run your programme at myself according to your scenario to search for patterns.
This programme is mainly designed for short sections. For the whole story I already know what the result will be. It will be, but the expectation is very low. And if you want to find something serious, it will probably take months of continuous work. It is difficult to estimate the time, but I think even with this programme it will take a lot of time, although of course it will be in automatic mode. And it is better to do it on a server on some kind of multi-core mode.