Yoghurt systems and canned systems or The relationship between trading tactics and the reliability of historical test results - page 2
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
What do long and short patterns mean?
The ones that last 'longer' and the perishable ones =)
An EA should not have any parameters that need to be optimized (picked up on history). IHMO optimisation on history is cheating itself. The only thing that I think is acceptable is if in the whole range of changes from - nocon to + nocon. The Expert Advisor remains profitable (all the runs) then it is worth choosing a parameter from the area that ensures a stable maximum profit.
I totally agree. This also means that the Expert Advisor is based on good patterns. Apparently, such an advisor is a long-lived one. What do you think?
I think the longevity of a system depends on the patterns on which it is built. The market is volatile and the patterns that worked before may not work now. There are different kinds of patterns in the market. Some last a long time, others do not reveal themselves much, and some are not regularities at all, but wishful thinking that we take for granted. Therefore, to create a reliable system, you have to immediately eliminate signal systems based on short patterns. You have to look for long patterns and catch the signals from them.
This is the main problem - how to determine if and how long the patterns we have found will work in the future?
This is the main problem - how do we determine whether and for how long the patterns we have found will work in the future?
It is probably worth testing. Right in the tester. If we notice in a month that something repeats, we algorithmize the idea. We tested it for a month. Optimized it. Then we go back to the history and run the same periods but with the same parameters. If adjacent periods are similar to the "benchmark" one, it is good, we move on. If in one, we are still on the plus side, that is also good. And so on. In short, the depth of such a successful run can be considered the "longevity" of the pattern. I am sure that the vast majority of visible patterns are lost even on adjacent test periods. They are very short.
Do you think the method is suitable? :)
Rоbеrt Pаrdо_Dеsign,Testing and Ortimisаtiоn of Trading System Rus
Encyclopedia of trading strategies.
In these two books the question of optimization and trading systems creation is well described. Here on the forum you can get to the general formulations and explanations, but it is better to work out a certain approach yourself.
1. If we know the patterns, we can at least teach someone else.
2. If these patterns are mathematical, they are 100% transferable to a computer.
3) It is possible to construct a knowledge base from the patterns we know.
3a Reliability of recommendations of the knowledge base of regularities must give not less than 95% prognosis.
4. based on the knowledge of regularities, it would be possible to certify traders)))
etc.
Question - so? How many real working traders have been trained?
Conclusion: "we" do not know the patterns, but only create another fetish, something like god-seeking and god-building.
It's probably worth checking. Right in the tester. We noticed on a month that something was repeating - we algorithmised the idea. We test it on a month. We optimized it. Then we go back to the history and run the same periods but with the same parameters. If adjacent periods are similar to the "benchmark" one, it is good, we move on. If in one, we are still on the plus side, that is also good. And so on. In short, the depth of such a successful run can be considered the "longevity" of the pattern. I am sure that the vast majority of visible patterns are lost even on adjacent test periods. They are very short.
Do you think the method is suitable? :)
That's the thing, it may work well on history, but in the future it may not work at all or work a bit and that's it. I've seen many different TS (in my own and not in mine) that produced excellent results on history, but didn't work in the future. And vice versa, the result on the history was mediocre, but in the future, on the real, these TS plundered money like crazy. This raises the question, what criteria are used to determine the serviceability of TS in the future? For me this is the main question now. I cannot make a profitable TS on history, even on OOS - not a problem. But how do I understand whether this TS will work in the future? And the most important - how long?
Let's see if we know the patterns:
1. If we know the patterns, we can at least teach someone else.
2. If these patterns are mathematical, they are 100% transferable to the computer.
3) It is possible to construct a knowledge base from the patterns we know.
3a Reliability of recommendations of the knowledge base of regularities must give not less than 95% prognosis.
4. based on the knowledge of regularities, it would be possible to certify traders)))
etc.
Question - so? How many real working traders have been trained?
Conclusion: "we" do not know the patterns, but only create another fetish, something like god-seeking and god-building.
Wow. I wrote, the patterns are changing. Also... Obvious patterns on price (or with the simplest constructions) can all be very weak. There are other, deeper and non-obvious patterns. For example, what is an indicator? An indicator is some transformation of market information in order to represent it in another form. Maybe on this other form some non-obvious regularities will appear?
And why do you have specific validity figures? And in general, it's not about prediction here, it's about the steady repetition of something, even if it's not repeated every time, but every other time, for example. That doesn't mean that the pattern is short, it can be very long at all, but it's just not absolute. On it, as you know, you can make money too. MM - here's some profit for you even from such a pattern.
Conclusion?
That's the thing, it may work well on history, but in the future it may not work at all or it may work a little and that's it. I've seen many different TPs (for myself and not for me) that produced excellent results on history, but didn't work in the future. And vice versa, the result on the history was average, but in the future, on the real, these TS plundered money like crazy. This raises the question, what criteria are used to determine the serviceability of TS in the future? For me this is the main question now. I cannot make a profitable TS on history, even on OOS - not a problem. But how do I understand whether this TS will work in the future? And the most important - how long?
That is why I described the test of workability and longevity of the sought pattern in order to assess how stable it is and to conclude how long it can last in the future. That is, it is as if we are accelerating on a springboard and jumping. The length of the run-up and the height of the springboard determine how far we can fly :)
For the 21st century, that's a bit weak.