Discussion of article "Exploring Seasonal Patterns of Financial Time Series with Boxplot" - page 23

[Maxim Kuznetsov]

by and large, all optimisation boils down to finding (and/or selecting) an extremum in a multidimensional space.

Even! as long as the limits of derivatives are known and they are quite smooth, gradient descent is suitable.

but this is subject to the only nuance that is forgotten during "optimisation" - physical+logical interrelation must always be strictly observed.

"You can't just walk into Mordor."

[Maxim Dmitrievsky]

fxsaber:

Now what is expressed in the discussion from my side.

1. The result of any optimisation is a statistical study, even if there is no 100% interpretation.
2. Classical statistical study is the result of implicit optimisation on the interval of statistical study.
3. It is wrong to talk about a pattern without a result outside the statistical interval. You need OOS, at the very least.
4. Random luck, when the best Optimisation pass on OOS shows a good result, is possible. But it's better than nothing.
5. An EA is published that finds the "pattern" from the article in three minutes, showing a result just as good. The EA matches the one the author did not include in the article.
6. It is claimed that any Expert Advisor must contain a filter by time of day. Otherwise - serious omissions in the search for patterns.
7. It is claimed that trading deviations from the MA is one of the simplest and most famous TS.
8. A method is suggested that anyone can see a good result on OOS according to the research in the article. Whether it is luck or something else - without analysis.
9. The example of the posted Expert Advisor demonstrates that it is easier and more efficient to conduct such studies in the Optimiser without involving other tools.

1. this is hard to believe, because the result of optimisation is not a statistical study, but a study of a spherical horse in a vacuum, often (or rather, almost always) uninterpretable due to the curse of dimensionality.

2. "implicit optimisation" is a term invented by you personally, i.e. nothing like it exists and cannot be referred to. Statistical research is statistical research, quantification of events and estimation of probabilities, if you will.

3. we can talk about a regularity when the hypothesis of its existence (based on statistical research) is confirmed by a bot. I.e. statistical conclusions about the presence of a regularity are confirmed in practice. Here we do not touch the dark forest of confirmation and rejection of statistical hypotheses. As Alexei Nikolaev wrote, this would be a good thing to do. Also, there are absolutely no confidence limits defined for such tests when the original hypothesis allows for the "nothing is forever under the moon" option. In this sense, "regularity" is interpreted as a quantitative recurrence of similar events over the interval under study, the more the better.

4. it is just nothing, neither better nor worse, as it has no statistical significance, since confidence limits are not defined (see point 3).

5. The EA is copied from the article with the original principle found in the statistical study and simply optimised. Since the TS variant is as simple as possible, the optimisation is as simple as possible, with a high chance of hitting the target.

6. Let it be asserted, whatever. Probably agree.

7. Again, whatever, you're welcome to it.

8. You can see anything you want, no method is suggested. The usual well-known optimisation.

9. The example of the posted Expert Advisor demonstrates that fxsaber can do optimisation. Nothing else. Optimisation and statistical research are different things (see wikipedia) and such a comparison is just amateurish. Optimisation can be at the final stage of a study, nobody denied that. Doing research entirely on the basis of optimisation is another nonsense, it is not research, and everyone knows this very well, including Mr. Saber. Garbage in, garbage out.

I have only one question to ask you, why do you deliberately introduce some kind of distinction into our conversion and actively defend it, why make life so difficult for yourself and people. The question is rather rhetorical.

[Alexander_K2]

Maxim Dmitrievsky:

In fact, the people on this forum just don't know anything but programming.

Max, this is a forum for coders, that's all! I recommend you to post materials from the article on other resources. And by no means turn off the path of researching the market time and its structure.

And may grace come upon you in the form of cash.

[Maxim Dmitrievsky]

Alexander_K2:

In fact, the crowd on this forum just doesn't know anything but programming.

Max, this is a forum for coders, that's all! I recommend you to post materials from the article on other resources. And by no means turn off the path of researching market time and its structure.

And may grace come upon you in the form of cash.

I think there is a deaf phone effect here, people do not know each other and speak different languages

By your prayers, Alexander ))

[Реter Konow]

Optimisation in simple words:

Optimisation is the search for values of a group of parameters that provide the maximum value of a target parameter within a closed-loop system.

Thesearch can be considered as a research if it is a deliberate search of values and verification of the result in the target parameter.

If data on dependencies of values are collected and conclusions are drawn, it is a statistical study.(The genetic algorithm automates this and we don't notice the process).

BUT! - statistical study is not the essence of Optimisation, it is the ONLY part.

fxsaber wrote:"The result of any optimisation is a statistical study, even if there is no 100% interpretation."

But the result of any optimisation is the best values of the system parameters providing the maximum result in the target parameter. And statistical study is a tool, a means, a part of the optimisation process, which works automatically in the tester.

[Maxim Kuznetsov]

Алексей Тарабанов:

Oops. Can I have two more of these plots? If it's in a row, no offence.

Here's "a couple more plots."

[secret]

Alexander_K2:

In fact, the crowd on this forum just doesn't know anything but programming.

Max, this is a forum for coders, that's all! I recommend you to post materials from the article on other resources

You were immediately told on another resource that you are talking nonsense with a clever look).

The laws of mathematics, they are the same for all resources).

[Maxim Dmitrievsky]

Реter Konow:

About optimisation in simple words:

Optimisation is the search for values of a group of parameters providing the maximum value of a target parameter within a closed-loop system.

Thesearch can be considered as a research if it is a deliberate search of values and verification of the result in the target parameter.

If data on dependencies of values are collected and conclusions are drawn, it is a statistical study.(A genetic algorithm automates this and we don't notice the process).

BUT! - statistical study is not the essence of Optimisation, but ONLY part of it.

fxsaber wrote:"The result of any optimisation is a statistical study, even if there is no 100% interpretation."

But the result of any optimisation is the best values of the system parameters providing the maximum result in the target parameter. And statistical study is a tool, a means, a part of the optimisation process, which works automatically in the tester.

Without initial assumptions (statistical study) there is no optimisation, there is nothing to optimise. You can optimise your fantasies, but then the result will also be in the form of pink ponies

Don't confuse a general statistical study with statistics obtained during optimisation. Otherwise the tail starts wagging the dog.

It would be good to give links to citations.

[Stanislav Korotky]

If possible, I prefer to analyse in the terminal itself, so I took one of my indicators, supplemented it a bit to display increments in pure form in addition to cumulative past cycles, and got this picture (superimposed on the original from the article):

This is the average of hourly candles, not the median in M15, but there are similarities.

Ran WFO from 2016 (genetics) with these settings: optimisation on 12 months, forward trading on 3 (total 12 passes). Got the report:

inMaxAbsoluteDD and inMinTrades were not used.

[Maxim Dmitrievsky]

Stanislav Korotky:

If possible, I prefer to make analyses in the terminal itself, so I took one of my indicators, supplemented it slightly to display increments in pure form in addition to cumulative past cycles, and got this picture (superimposed on the original from the article):

This is the average of hourly candles, not the median in M15, but there is a similarity.

Ran WFO from 2016 (genetics) with these settings: optimisation on 12 months, forward trading on 3 (total 36 passes). Got the report:

inMaxAbsoluteDD and inMinTrades were not used.

Are there no hourly statistics? I couldn't see it

I.e. the initial hypothesis: during certain hours, the increments, on average, increase, i.e. during these periods the maximal profit will be reached. If we get hourly statistics of profits through optimisation, we can confirm/refute the hypothesis (without adding new parameters to the TS).

And if the optimiser finds other hours, it will be a fitting. This is very likely to happen when the number of TC parameters increases, i.e. genetics will trade in the wrong place and will not find anything from the hypothesis (as an option).