Where is the line between fitting and actual patterns? - page 32
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
We talk about wasps-schmos, samples-memples, but no one said a word, and probably no one even thought, but do such approaches (division into Sample, OOS, etc.) apply to all TCs without exception? If not applicable to all, which ones are not applicable? I posed a question to Reshetov leading up to this topic, but he didn't see fit to answer.
Let's try to get to the bottom of this.
In terms of time spent in the market of each single trade (each single trade, because TS may conduct several parallel trades simultaneously) TSs may be divided into two types:
1. TS with unknown time of each trade. This type includes all TS, in which it is not known in advance when there will be the next entry signal into the trade (or whether it will be at all), and in some systems it is not known in advance when there will be an exit signal.
2. TS with the maximum trading time known in advance. This type includes systems in which there is a signal to enter strictly periodically, for example on every bar or on a certain day of the week at a certain time. There is always a signal for the exit, because the maximum lifetime of each trade is predefined. The condition of exiting a trade can be the end of the time limit or the reaching of stops.
What thoughts arise when we divide the TS into such types? What follows from this, within the framework of the topic? I will listen to the opinions, and then I will express myself.
In my opinion they can be adjusted in the same way. Yes, for the former the time spent in a trade is blurred, but it is also limited. The main thing, as Figar0 pointed out, is the analysis of results. Of course, if the main profit of such a system accounts for several deals that were in the market as long as possible (according to the fact of testing), the probability that it is a fitting is high. If the main profit was obtained in deals with average holding time (i.e., there were many such deals - stat reliability), then it is more reliable. Yes, for systems with maximal time of position holding such analysis is not necessary, but the adjustment due to tp>>sl or sl>>tp can be implemented in them as well. Only a significant increase in statistics (number of deals on history) will help here. Because, for the statistics on the system to be reliable, there must be a lot of implementations of each outcome of the system - both positive and negative. There are more outcomes in systems with explicitly unlimited time of position holding, so additional analysis is needed to prevent rare outcomes from distorting the results.
In short, statistical validity is important and in some types of systems it may be lower than in others for the same number of trades. Confirmation of statistical validity may require additional analysis of results. And in general, most of the ways to confirm that the system "does not fit", such as multiple instrument operation, wide optimal zone, etc. - is the way to increase statistical reliability, which is the main proof of the system's performance. Other than that only common sense and insider knowledge :)
In my opinion they can be fitted in the same way. Yes, for the former the time spent in the trade is blurred, but it is also limited. The main thing, as Figar0 pointed out, is the analysis of results. Of course, if the main profit of such system lies in several trades that were in the market as long as possible, the probability of it is high. If the main profit was obtained in deals with average holding time (i.e. there were many such deals - stat validity), then it is more reliable. Yes, for systems with maximal time of position holding such analysis is not necessary, but the adjustment due to tp>>sl or sl>>tp can be implemented in them as well. Only a significant increase in statistics (number of deals on history) will help here. Because, for the statistics on the system to be reliable, there should be a lot of implementations of each scenario of the system - both positive and negative. In systems with obviously unlimited time of keeping a position there are more scenarios and therefore additional analysis is needed to prevent rare scenarios from distorting the result
I agree. It is possible to fit both types. But only the second type is capable of looking for patterns.
To begin with, it is necessary to clearly define what a pattern is. What everyone is looking for after all. Knowing clearly how to look (rather than what to look for), it is easier to look. So, the red girl, when she goes looking for plantain, is strictly directed to the roadside, as she knows that that is where plantain grows - it is a pattern, and she does not need to know, and she does not know exactly why she should look for plantain by the roadside. That is, she has trained herself, optimised herself, in such a way that she will be the most successful red maiden in the whole village at looking for plantain. But plantain does not grow on all the roads, this pattern is not always manifested, but it does not cease to be a pattern, and the road-plantain can be used to collect medicinal herbs (read: earn money). Also, this pattern does not work for the whole planet (FX), but only in certain countries and at certain latitudes (FX tools). Moreover, I intentionally do not talk about the time of year (this is a throwback to mushroom growers), because there is plantain in winter under the snow (at any time of year), just a red maiden not appropriate to shovel snow by the roadside (higher spreads, high commissions, regional bans).
From the wiki: Legitimacy is a necessary, essential, constantly repeated interrelation of phenomena of the real world, defining stages and forms of formation process, development of phenomena of nature, society and spiritual culture.There are general, specific and universal laws.
From myself: I agree with the wiki. Moreover, in order to speak about the presence/absence of regularities, we need a clear fact of the presence of one phenomenon and the phenomenon that follows it. Phenomena that can be measured, felt, sometimes smelled, and measured the timing of the investigative phenomenon. After all, a phenomenon that was, is and will be does not mean that there was a cause for it, and if there isn't, you can't exploit that investigative phenomenon and identify when it occurred and predict when it will end.
Let's return to TC types. Let's start with dissection of the first type. We will investigate a simple TS with the intersection of two MAs. All the TS derived from it have the same properties. I will gather my thoughts and continue. In the meantime, we may pare down this post.
No, it's better to start with the second type. Giggling already after reading the previous post? I'm chuckling myself.
In general, the second type of TC is all facets of the same cutlet, along with Flowing Pattern Theory and Instrumental River (this is with Alexei's lighter hand).
There is a causal phenomenon, followed by an investigative phenomenon. Causal and causal patterns. Once a pattern is found, we can use different variations of this pattern. Optimization of such TCs consists in searching for cause-and-effect relations in the patterns. Since patterns have a beginning and an end in time, it's easy to calculate the necessary Sample period size, so that the number of trades would be 300-400 at least. OOS (20-30%) is distributed by Sample and used to control optimization. Thus, real can follow immediately after optimization. The number of trades is a constant, hence all stats are simplified, it is easier to select the right one and the probability of fitting is reduced.
If patterns are found using this approach, they will operate both when increasing the size of figures, and when decreasing them - the patterns do not depend on it (as opposed to that, look at TS by MA). If no pattern can be found - then suck it and dry off. Clearly - either there is or there isn't.
What you say is not true of all patterns. For patterns, yes - they have a clear lifespan, longer than which it makes no sense to be in a trade. For trend-followers - no. In addition, the market has its own internal time and its course may be different and not coincide with the astronomical one. I.e., the stages of "life" of a regularity and the real market process, standing behind it, may have different duration in time, which is unknown in advance and can be detected in the course of a trade. We can synchronize with them (phases) according to events on the chart - i.e. signals. Like one signal - the process started, the second signal - the process is over in a simplified version. Although it is possible to identify its other phases.
Now on MA, the first type. Yes, they lag, but that's not what we're talking about now. The MAs do not show any pattern. It follows from the fact that we do not know when the next crossover will occur (the market is not a sine wave, it is not stationary). We also know that at any time interval we can find such a combination of MA that will yield a profit.
So, we optimized the MA and found such a combination of 200-100. Ok. We ran it in the real sector. We waited for the crossing and entered the market. But the market decided to decrease the size of figures and there is no signal to exit. We got old and died without waiting for it. But it doesn't mean that we were wrong to enter, but nobody needs this "rightness". We won't know if we optimised correctly or not. And nothing will change with different optimization variants, with addition of stops and trailing stops, the problem will remain the same. It is impossible to trace the cause-and-effect relationship between one phenomenon and another.
If regularities are found during this approach, they will work both when the size of figures is increased and when they are decreased - the regularities do not depend on it (for contrary, look at TS by MA). If no pattern can be found - then suck it and dry off. Clearly - either there is or there isn't.
Take two sets of numbers. Combine them into one, for example, by alternation - number from 1, number from 2, etc.
Now we will look for regularities - we will surely find, for example, after 1->2, after 7->5, etc.
Let's take the same sets again, but first shift the 1st set one number to the left, and then join them again by alternation.
What kind of patterns do we have now? Apparently somewhat different. Even cause and effect may be interchanged here and there :).
I want modestly to share some information. I should say at once that I am far from the programmer profession, but my curiosity eats me, as well as any of those present here with a selfish purpose. It is called differently, but I will say it is a bar placed by its range (HL) in the previous candle (HL), i.e. fading or correction of the current trend.The question - how often the next candle after the "placed" will be the opposite colour from the first. My answer is about 50/50 for the daley, with a deviation of about 0.3%... I took another time frame and got the same ratio: 45 034 lines (minutes), positive outcomes - 2224 with a possible 4471.Hourly for the year 408 / 812. Daly since 2001 - 164 / 337.
And again a question - what is going on? Is the whole system set up and is it really artificially generated? Or am I completely clueless about excel :o). I can send the "book" to someone, maybe check it out... I am personally shocked... I just wanted to stay out of the market and occupy myself with something...
What you write is not characteristic of all patterns. For patterns - they have a definite time of life, longer than which it does not make sense to be in a trade. For trend-followers - no. In addition, the market has its own internal time and its course may be different and not coincide with the astronomical one. I.e., the stages of "life" of a regularity and the real market process, standing behind it, may have different duration in time, which is unknown in advance and can be detected in the course of a trade. We can synchronize with them (phases) according to events on the chart - i.e. signals. Like one signal - the process started, the second signal - the process is over in a simplified version. Although it is possible to identify its other phases.
Take two sets of numbers. Combine them into one, for example, by alternation - number from 1, number from 2, etc.
Now we will look for regularities - we will surely find, for example, after 1->2, after 7->5, etc.
Let's take the same sets again, but first shift the 1st set one number to the left, and then join them again by alternation.
What kind of patterns do we have now? Apparently somewhat different. Even cause and effect may be reversed here and there :)
I want to humbly share some information. I must immediately say that I'm far from the profession programmer, but I'm, like any of those present here curiosity eats with selfish purposes. It is called differently, but I will say it is a bar placed by its range (HL) in the previous candle (HL), i.e. fading or correction of the current trend.The question - how often the next candle after the "placed" will be the opposite colour from the first. My answer is about 50/50 for the daley, with a deviation of about 0.3%... I took another time frame and got the same ratio: 45 034 lines (minutes), positive outcomes - 2224 with a possible 4471.Hourly for the year 408 / 812. Daly since 2001 - 164 / 337.
And again a question - what is going on? Is the whole system set up and is it really artificially generated? Or am I completely clueless about excel :o). I can send the "book" to someone, maybe check it out... I am personally shocked... I just wanted to stay out of the market and occupy myself with something...
I confirm. somewhere in the code base there was even a script calculating the probabilities of candlestick patterns.
It worked out on a large enough 50/50 range with small offsets obviously not sufficient for building a TS.
i got quite a large 50/50 range with small offsets, clearly not enough to build the trader.
i.e. it may wager 10 times pocket and then 5 times profit.
It would be statistically reliable and normal. :)