Machine learning in trading: theory, models, practice and algo-trading - page 2092
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
yes, give me two!
As for the subject, about the news - I foresee the outcomes:
- News influence on further price movement, still need to be marked manually
- The ZigZag is not suitable for this evaluation, it's better to use some trend indicator... Then why analyze the news? We work the old way, tester + GA or tester + MO
- the task of news analysis is a heuristic task, so to say, this is creativity.
as an alternative - just append news to some already marked dataset... just to try to increase accuracy
As far as I understand, there are no good free archivesAs an option - just add news to some already marked up dataset... just to try to increase the accuracy
As far as I understand, there are no good free archivesyes
But I would put it this way:
add news data as an additional filter to the existing TS
and try to tune the ready TS with the help of MO or GA
Imho, in this case we're going back to the classics - the price takes into account everything, i.e. in our case the charts analysis is more important.
yes, give me two!
As for the subject, about the news - I foresee the outcomes:
- News influence on further price movement, still need to be marked manually
- The ZigZag is not suitable for this evaluation, it's better to use some trend indicator... Then why analyze the news? We work the old way, tester + GA or tester + MO
- The task of news analysis is a heuristic task, so to speak, this is creativity.
In my opinion, the zigzag is quite a trend. It lengthens when it is in the direction of the trend and shortens when it is the opposite (compared to SB). You just have to compare knees with news and knees without news.
I was puzzled (in the sense of the approach to its formalization) by your idea about the influence of the news on the price until the moment of its release. It turns out that the influence is on two segments of time - before the news and after it, and it may well be different (opposite?).
In addition to the significance of the news (high, medium, low), we must take into account the difference between the numerical values of the forecast and the preceding one, forecast and real (past and past revised?)
on the news this "length" can get longer
But are we looking for trends or reversals?
And what if there is a return of the price? or it was a hairpin on the news?
Why TF15? The smallest available TF M1, well there is the logic of using D1, but M15 - well, you like the number 15, but it should not be a scientific basis for the process under study?
From my trading experience when I was into the news. The time of action is from an hour to 3 hours maximum, if the time of action is less than half an hour, the news is just not noticeable, and after 3 hours most of the time the news trend ends. Minutes will have a more accurate definition of the beginning of the zigzag, but also have a lot of noise.
In my opinion, the zigzag is quite trendy. It lengthens when it is in the direction of the trend and shortens when it is the opposite (compared to the SB). We just have to compare knees with news and knees without news.
ZZ "catches hairpins on the news".
I think that studs should not carry information, this is purely a technical point
and the same MAXD does not notice much or rather changes its value on the studs
I was perplexed (in the sense of the approach to its formalization) by your idea about the news influence on the price before the moment of its release. It turns out that the effect is realized in two intervals - before the news and after it, and it may be different (opposite?).
of course, part of the information has been already present in the market, and part of the information has been received with the news
some participants have more complete information
a part of participants does not analyze the news, but nevertheless they solve their current problems
i.e. in the time range it is very difficult to assess where the news have already begun, and where this news has just appeared in the market and influenced the price
this is about time, and about the price... everything is sadder there (or worse)
if we put aside all sorts of hypothetical trends, then in most cases, almost at any point in time (well, if you look at small increments in ZZ)
it is possible to open the deal both to buy and to sell, and the deal already opened earlier is not the fact, that it should be closed
Just to put it in perspective:
let's look at the SZ, what are we looking for for the ideal TS? - entry on a break, reversal on the next break of ZZ
then our TS should know the future? - that is not scientific in my opinion!
What should be scientific? - the TS should pass the top of ZZ and after the retreat of X points from this break in ZZ, make a decision to close or make a rollover
but, in this case, part of the ZigZag knees will carry extra information?
and how many pts should be "discarded" from the top of ZZ?
I think that unformalizable problems of using ZigZags, imho, yes it's ideal, but it does not take into account that for the market to work, someone has to sell and someone at the same time to buy - I mean those small price movements that we are trying to filter out with ZigZ knees - these are counter-orders, someone opened the wrong deal, but without these small movements, the market would simply not work
In my opinion, the zigzag is quite trendy. It lengthens when it is in the direction of the trend and shortens when it is the opposite (compared to the SB). We just have to compare knees with news and knees without news.
I was puzzled (in the sense of the approach to its formalization) by your idea about the influence of the news on the price until the moment of its release. It turns out that the influence is on two segments of time - before the news and after it, and it may well be different (opposite?).
In addition to the significance of the news (high, medium, low) it is necessary to take into account the difference between the numerical values of the forecast and the previous, forecast and real (past and past revised?)
Yes, what I've seen is a half hour before the news started. And usually a co-directional action, from low to strong.
Taking into account the forecast and the real values of the news, this just requires a good interpreting, there are many combinations. Co-directionality, multi-directionality, forecast coincided with the real one by a sign at least, or did not coincide. News is qualitative yes no, and quantitative. Quantitative in yes no is also difficult to put. Increased or decreased the simplest, but there is left at the same level. I.e. the effect of strong news may not be strong, and weak news may be strong, if not only the difference in the forecast of the weak news, but also a large change in the indicator in the news.
yes
But I would put it this way:
to add to the ready TS as an additional filter, the data on the news
and try to tweak ready-made TS with help of MO or GA
Imho, then we return to the classics - the price takes into account everything, i.e. in our case the charts analysis is more important.
That makes sense. This is an estimate of the FA. Which can be used for any purpose.
ZZ "catches spikes" on the news
I think that studs should not carry information, this is purely a technical point
and MAXD does not notice much or rather changes its value on hairpin tips
of course, part of the information was already in the market, and part of the information came with the news
some participants have more complete information
some participants do not analyze the news, but nevertheless they solve their current issues
i.e. in the time range it is very difficult to assess where the news have already begun, and where this news has just appeared in the market and influenced the price
this is about time, and about the price... everything is sadder there (or worse)
if we put aside all sorts of hypothetical trends, then in most cases, almost at any point in time (well, if you look at small increments in ZZ)
it is possible to open the deal both to buy and to sell, and the deal already opened earlier is not the fact, that it should be closed
Just to put it in perspective:
let's look at the SZ, what are we looking for for the ideal TS? - entry on a break, reversal on the next break of ZZ
then our TS should know the future? - that is not scientific in my opinion!
What should be scientific? - the TS should pass the top of ZZ and after the retreat of X points from this break in ZZ, make a decision to close or make a rollover
but, in this case, part of the ZigZag knees will carry extra information?
and how many pts should be "discarded" from the top of ZZ?
I think that the unformalizable problems of using ZigZags, imho, yes it's ideal, but it does not take into account that for the market to work, someone has to sell and someone at the same time to buy - I mean those small price movements that we are trying to filter out with ZigZ knees - these are counter-orders, someone opened the wrong deal, but without these small movements, the market would simply not work
Yes, what I saw was half an hour before the news started. And usually a co-directional action, from low speed to strong.
Taking into account the forecast and the real values of the news, it just requires a good interpreting, there are many combinations. Co-directionality, multi-directionality, forecast coincided with the real one by a sign at least, or did not coincide. News is qualitative yes no, and quantitative. Quantitative in yes no is also difficult to put. Increased or decreased the simplest, but there is left at the same level. That is, the impact of strong news may not be strong, and weak news strong, if not only the difference in the forecast of the weak news, but also a large change in the indicator in the news.
Yes, the case is clearly a murky one)
To put it crudely, but to put it mildly - there is something to think about)
On the 25th the roll-up course is supposed to end, but it hasn't started yet... It's 72 hours.
If it starts on Monday, that's 72\15 = 4.8 hours a day.
Here are people doing useful work https://rb.ru/story/ai-cough/
The model correctly identified coronavirus patients by cough in 98.5% of cases.