Machine learning in trading: theory, models, practice and algo-trading - page 2858
![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
Well, there is an approximation of the curve by splines. It is necessary to take splines of the first order (broken line). Then select thenecessary segments.
Are you talking about interpolating with splines? I've never done that.
First order - isn't it a straight line between two points - eh darkness me.
And, how to define again "necessary segments", which should be selected?
The closer you get to the end, the less significant the level is due to the fact that they account for a smaller and smaller share of the sample. The probability that it's just a random artefact increases.
It's still interesting to understand what the size of the ZZ cutoff depends on.
I actually think there's a pattern with respect to the penultimate segment...
Right now the main topic in MO is ChatGPT. Maybe even write counsellors) They say it will be much more useful than Google.
Is it the main topic? I think it's a toy, it generates all kinds of Chernigovshchina.
There's no way to get new knowledge from it, by generalising what it has memorised.
Search engines are also now based on neural networks, and give more relevant information.
Are you talking about spline interpolation? I've never done that.
First order - isn't that a straight line between two points - eh, I'm dark.
I advise you to read Tibshirani from page 295.
And, how to determine, again, the "necessary segments" to be sampled?
This is determined by the problem being solved. For example, you can try to take all the segments as predictors - each segment is given by two numbers (X of the origin and slope coefficient).
It is still interesting to understand what the size of the ZZ segment depends on.
I generally believe that there is a regularity concerning the penultimate segment....
It's quite possible. It can be influenced by the penultimate segment and the ratio of the sums of knee heights of different directions, the presence of news, etc. etc.
search engines are also now neural-networked, and give more relevant information.
I think google's getting tense about this thing. While it does not replace it, but its prospects have become more vague.
Is it the main one? I think it's a toy, it generates all kinds of Chernigovshchina.
There is no way to get new knowledge from it by generalising what it has memorised there.
Search engines don't generate anything either, but there's no way without them on the Internet, and this thing is the next stage in their development. They are already writing simple TCs in Python with its help, and in time they will move on to more advanced ones.
I think google is already stressing over this thing. It's not replacing it yet, but its prospects have become more dim.
Search engines don't generate anything either, but there is no way without them on the Internet, and this thing is the next stage in their development. They are already writing simple TCs in Python with its help, and in time they will move to more advanced ones.
They are already writing simple TCs in Python with its help, and in time they will move to more advanced ones.
I will try to write TC in python with help, interested 😀 however I'm sure it will be ridiculous. Inserting a couple of snippets will be much faster and more reliable.
A search engine should just search, but this thing generates original text. It doesn't do it very well yet, somehow streamlined or something. I asked to solve an equation that has no solutions (by Fermat's theorem) and it "found" solutions by adjusting the equation and "not noticing" the requirement of positivity) But you can continue the dialogue with her and she eventually admitted the error) When asked the same thing with a direct instruction to use Fermat's theorem, she immediately said that there are no solutions).
I'm too lazy to mess with this thing, and I don't have access to it - it's hard to get it in Russia, I have to work hard.
A search engine should just search, but this thing generates original text. It doesn't do it very well yet, somehow streamlined or something. I asked her to solve an equation that has no solutions (by Fermat's theorem) and she "found" solutions by tweaking the equation and "not noticing" the positivity requirement) But you can continue the dialogue with her and she eventually admitted the mistake) When asked the same thing with a direct instruction to use Fermat's theorem, she immediately said that there are no solutions).
I'm too lazy to mess with this thing, and I don't have access - it's hard to get it in Russia, I need to mess around.
I couldn't find it on google.
There were some examples with requests like "write code for an EA in Python to cross averages for such an instrument from such an exchange using such an api". I think even something meaningful was obtained in the end. Surely you can find an answer to such a question in Google, but here communication is in a more human language and it seems even possible to continue the dialogue with clarifications.
1.At the same time, we most often skip those that have a complex structure, for example, even such rules as "On Thursday at 16 o'clock sell, if the daily bar is growing".
This is not a complex structure - it is a coincidence that will not repeat/repeat in the future (P=0.5). That's the conclusion I've come to. All "complex" non linear relationships without theoretical justification are fiction. Just an algorithm stretched over the data. Market participants are not aware of this "complex" relationship. Which means this relationship has no effect on price movement.
It is another matter if there is a theoretical basis for "Thursday at 4pm sell if the daily bar is rising". See below
2. What you mean by "real reasons" is not clear here.
Well, it's an old example. It is not like this now, but the meaning will be clear to you.
From the point of view of a single economic entity, the main activity of Russia can be characterised as an exchange of natural resources for the benefits of civilisation. In this case, the basic monetary unit for such an exchange is the dollar (because it is the universally accepted world currency). That is, the proceeds for natural resources are in dollars. At the same time, all intra-country activities are linked to roubles. And, importantly, all domestic government spending is also in rubles. In order to get secured money for its spending, the state takes away a part of profit from rent companies in the form of MET tax. Thus, in this chain there is a forced exchange of dollars for rubles by rent companies. Compulsion is always a good thing, because compulsion means that there is a party ready to throw market bids and thus move prices.
Like the payment of any taxes, the payment of MET is strictly regulated. For us it will be important that the tax is paid once a month, the tax for the previous month is paid before the 25th of the current month.
Further, it is known that everything is always done at the last moment. Exchanging quid for rubles in order to pay taxes is not an exception (besides, there are other reasons to hold quid longer). Therefore, it is very likely that companies/individuals subject to severance tax will exchange dollars for rubles in the period just before the 25th ofeach month.
So the idea is to short dollar-ruble before the 25th of each month.
Any calculations of the theoretical interest rate can also be attributed here.
3. Any strategy has unfavourable periods that can be protracted, so you need to have a zoo of such strategies, and to create it manually is either to be a genius or to live a very long time. I just got into MO after developing such a genius strategy, which took about two years, and every sneeze of it I could logically justify, and when I put it to work I just came across an unfavourable period for it, which was smoothed out on the adjusted data in the tester.
Returning to the second point - I am looking to find similarities in the behaviour of binary predictors that will continue to be effective in the foreseeable future. To do this, I want to identify specific predictors. I concede that we need to take into account the cyclicality of the relationships between predictors, which any of the automata models I know of do not do.
The point I'm trying to make is that predictors are the point. And when they are normal - MO is no longer necessary.
I'm leaving the discussion - it's just that 10 years ago I taught my neural networks on OpenCL, I spent a shitload of effort and time on it, and I didn't get any profit from it. God willing, you will succeed.