How do you properly compare two non-overlapping rows? - page 5

[Renat Akhtyamov]

Evgeny Belyaev:

google : Market invariants.

Let's say a certain stock is currently trading around 10 roubles. Then an increase in its price by 1 rouble would be 10%. Suppose that after a few years the share price had risen to 100 rubles. The price increase of 1 rouble will be only 1%. One rouble for a 10-rouble share and a 100-rouble share are very different things, so investors are focused on percentage yields rather than monetary yields, that is, they are focused on relative values rather than absolute ones.

That seems to be what is needed.

Interesting option. But it has already been voiced above. A few posts about statistical methods. The gist is that a certain price range is treated as max and min. But I wrote about how the method gets out of hand when max or min is breached. And you need to be able to properly overlay the ranges like.

[CHINGIZ MUSTAFAEV]

Suffering from one's own ignorance is a holy cause)

[secret]

Aleksey Nikolayev:

Alexey, could you suggest some regression that is robust to outliers, preferably in the form of formulas?)

More accurately, not even to outliers, but to weak series correlation.

Because the usual ISC is bullshit on real data. I can tell more accurately by eye.)

[Evgeny Belyaev]

Renat Akhtyamov:
Interesting option. But it's already been sounded out above. A few posts about statistical methods. The gist is that a certain price range is treated as a maximum and minimum. But I wrote about how the method gets out of hand when max or min is breached. And you need to be able to properly overlay the ranges like.

You're out of the loop. Go grill out.

[Vladimir]

Evgeniy Chumakov:

There are two non-overlapping rows that are on 'different levels' (like in the picture above).

How can they be 'combined' so that they are side by side and overlap?

You can calculate an average in each row, then row_1 = value_1/mean_1, etc. But is this the right way to do it? Does the sample size affect the adequacy of the results... or should it be done differently? Or through normalisation of Max and Min ? Again sampling period? Actually what is the right way?

I think you know what I mean...

No, it's not clear. What does it mean to do any calculations (calculations, comparisons, graphing, quantising series, building neural networks,...) correctly, that is, to make a profit in the end - this is the question we are all looking for an answer to.

[Yousufkhodja Sultonov]

secret:
Alexey, could you suggest some kind of regression that is robust to outliers, preferably immediately in the form of formulas?)

Otherwise, the ISC is totally bullshitting on real data.

https://www.mql5.com/ru/forum/372456/page3#comment_23234856

[Aleksey Nikolayev]

secret:
Alexey, could you suggest some outlier-resistant regression, preferably in the form of formulas?)

More precisely, not even to outliers, but to weak series correlation.

Because the usual ISC is bullshit on real data. I am more accurate by eye).

There are formulas only for ANC) That's why it is very favourite and often used, despite all problems) For all others - fiddling with numerical methods...

If the correlation is weak, then either the correlation is weak (not significant) or it is not linear.

Try the Theil-Sen method to estimate the linear regression coefficient. It is quite popular and widely used in various sciences, but it is not often used in trading.

[Yousufkhodja Sultonov]

Aleksey Nikolayev:

There are formulas only for ANC) For this reason it is very favourite and often used, despite all the problems) For all others - fiddling with numerical methods...

If the correlation is weak, then either the relationship is weak (not significant) or it is not linear.

Try the Theil-Sen method to estimate the linear regression coefficient. It is quite popular and widely used in all sorts of sciences, but for some reason it is not often mentioned in trading.

Unfortunately, no one sees further than his nose. For a long time an ISC for non-linear regression has been developed - PNB is called. A lot of topics are devoted to this problem on the forum, but everyone stubbornly "ignores" them. PNBs were born and developed on our forum. It should be a shame to face difficulties when there are obvious solutions to the problem. Give them a foreign author, despite the fact that they are no match for PNB. Proven many times over. I'm getting tired of saying it publicly. I think in the days of Gauss and the two ISC authors, it was easier to promote achievement to the masses. I'm willing to compare PNB to the notoriousTheil-Sen methodhttps://www.mql5.com/ru/forum/372456/page5#comment_23244541

[Dmitry Fedoseev]

Yousufkhodja Sultonov:

Unfortunately, no one can see further than their nose. An ISC for non-linear regression has long been developed - PNB is called. Many topics are devoted to this problem on the forum, but everyone stubbornly "ignores" them. PNBs were born and developed on our forum. It should be a shame to face difficulties when there are obvious solutions to the problem. Give them a foreign author, despite the fact that they are no match for PNB. Proven many times over. I'm getting tired of saying it publicly. I think in the days of Gauss and the two ISC authors, it was easier to promote achievement to the masses. I'm willing to compare PNB to the notoriousTheil-Sen methodhttps://www.mql5.com/ru/forum/372456/page5#comment_23244541

Yusuf, you are the one who doesn't understand the elementary.

[Yousufkhodja Sultonov]

Dmitry Fedoseev:

Yusuf, you're the one who doesn't understand the basics.

What, exactly, don't I understand? Please enlighten me.