Market prediction based on macroeconomic indicators - page 44
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
I did not think that it would end with the banal sale of books, so he would tell)))) Sorry)))) hilarious...
I would divide all modern economic mathematics into two parts
...
Seemingly subtle differences have very serious consequences.
...
Both use extrapolation of data as such. If you swap the cart with the horse, it remains a cart with a horse - an extrapolation. It doesn't matter whether we are initially looking for factors that influence a change in the future, or whether we are comparing the change that has taken place with those factors - the result is the same, a model that we think should predict the future, based on data from the past.
Personally, I would divide all trading approaches into two other groups:
With the first group everything is more or less clear. These are infinite combinations of predictors with different weights to try to explain the variable that supposedly depends on them - the price. The second group works differently. It is based on the identification of specific process properties, followed by the exploitation of these properties. For example, if the model is able to identify a strong process return, it simply finds markets with such a characteristic and starts trading "at the level pullback". At the same time, specific rules for identification of this magic "level" or specific trading techniques are not very important. What is important is that the process really has this characteristic and does not change it over time. At the same time, the result of each particular market entry is not important at all, for there is no idea about the forecast for a certain trade. What is important is the final statistics, which depends on the properties of the process.
Both the former and the latter use extrapolation of data as such. .
Here you are simply not aware of it. There is no extrapolation as such in machine learning classification models.
For example, forests.
Trees are built on a training sample, usually for a sample of over 3000 bars you get more than 100 trees - a forest of trees. Each tree is a pattern in TA terms.
On arrival of the next bar the combination of predictors corresponding to this last bar is searched among trees. It can be the 1st tree, the 50th or the last one - the one that fits best. Where is the extrapolation here?
It is based on identifying specific properties of the process, followed by exploiting those properties
It's a great idea if you can generate a training sample. The whole problem is the teacher. Otherwise, with the knobs....
Here you are simply out of touch. There is no extrapolation as such in machine learning classification models.
For example, forests.
Trees are built on a training sample, usually for a sample of more than 3000 bars you get more than 100 trees - a forest of trees. Each tree is a pattern in TA terms.
On arrival of the next bar the combination of predictors corresponding to this last bar is searched among trees. It can be the 1st tree, the 50th or the last one - the one that fits best. Where is the extrapolation here?
It is based on identifying specific properties of the process, followed by exploiting those properties
It's a great idea if you can generate a training sample. The whole problem is the teacher. Otherwise, with the knobs....
Why use your hands, Kohonen classifies perfectly, the signs just need to be presentedAgain about Kohonen cards
I do not understand the practical value in trading of classification methods without a teacher. We trade quite specific things: longs and shorts, breakout and rebound of a level...
And here? There's PCA, it reformats predictors and the new set has a number of very interesting features, and here?
I do not understand the practical value in trading of classification methods without a teacher. We trade quite specific things: longs and shorts, breakout and rebound of a level...
And here? There is PCA, so it reformats predictors and the new set has a number of very interesting features, but here?
I do not understand the practical value in trading of classification methods without a teacher. We trade quite specific things: longs and shorts, breakout and rebound of a level...
And here? There is PCA, so it reformats predictors and the new set has a number of very interesting features, but here?
What is learning without a teacher - automatic selection of patterns. Having patterns - we find points on the map where "certain things pile up: longs-shorts, breakout-bounce...", and then we wait for signals online to activate these points.
If there is no teacher, the meaning of the patterns is not clear.
What is a teacher?
A chunk of quotient corresponds to longs, and this chunk of quotient corresponds to shorts. When the model is taught, the sets of predictor values are divided into two classes corresponding to the teacher.
What about without a teacher? What is the significance of patterns?
What about without a teacher? What is the significance of patterns?