Machine learning in trading: theory, models, practice and algo-trading - page 1714

[Valeriy Yastremskiy]

Reg Konow:
Even without digitizing fund.data, besides price there are volumes, open interest, price levels and volumes, time parameters - session, season, etc... News, I think, is already digitized...

That is, the menu of market parameters is much richer than is discussed and used here, and the potential of MO is not fully realized in research.

I would gather the available parameters into a conditional system, organize the flow of their values into an MO algorithm that calculates the coefficients of "cross-sectional" dependencies relying on a statistical database.

Try to systematize the digitization and logic of accounting these data. I don't get it right off the bat) From what's in the numbers, stock indexes, gross income, Central Bank rate in the country, projected gross numbers, agency estimates, credit ratings, industry numbers, trade turnover between countries... And how do you use that? Technicalanalysis is a separate song... from fundamental analysis.

[mytarmailS]

Evgeny Dyuka:

That's cool, I'll check it out...

And this is what scientists are already doing ....

Partitioning of series by Principal Component Analysis (PCA)

quotes in a sliding window of size 10

there are 8 components

The algorithm sorts the components into slowly changing and rapidly changing ones, into those that contribute significantly and those that are weak, and sorts them

The strongest property of decomposition (any Fourier, wavelet, pca) is additivity

i.e. if I add all components pc1+pc2+...+pc8 together I get the same price back, this property is called additivity

And now think if we use the same indicators instead of components and add them together, what do we get? Exactly what we get at the output of the neural network when training it on a stochastic)

The indicators can already be used on top of the components - just imagine how many features can be generated by that decomposition

You can filter data, find and extract unnecessary (bad components)

You can change components and put them back or leave only them

You can speed up components, slow them down, or change their amplitude.

You can analyze each component relative to the other, or two or three of them.

You can forecast components separately to sum up in the general forecast (they are additive)

It is possible to forecast one components relative to other components

It is possible...

It is possible...

It is possible...

You can...

You can...

Which is what scientists do when they want to figure something out.

[mytarmailS]

Evgeny Dyuka:
Just a neural network working in the real market as an indicator and predicts the movement of the asset well. In addition, I have one more experimental one trying to provide entry points. Here are the last four signals from the last 10 hours, all signals are public.

Great job!!! What's new?

[Реter Konow]

Valeriy Yastremskiy:

Try to systematize the digitization and logic of accounting these data. I can't do it right away) From what I have in figures, stock exchange indices, gross income, the Central Bank rate in the country, forecasted gross figures, agency estimates, credit ratings, industry figures, trade turnover between countries... And how do you use that? Technicalanalysis is a separate song... from the fundamental.

Well, first of all, the digitized fundamental data become an object of technical analysis. What prevents you from inventing your own method of digitization? Another thing is that the custom digitization of information is subjective and does not fit into the general data flow. That is, the system of converting meaning into numbers, works only in the head, not going directly to the algorithm. This is the problem.

As for the available market indicators. All of these components must be placed in a system model for which the MOE will calculate correlations. Like so ...

[Igor Makanu]

Good article on Habra: Do neural networks dream of electric money?

I don't offer to read it, I left the link for myself

[Valeriy Yastremskiy]

mytarmailS:

The strong property of decomposition (any Fourier, wavelet, pca) is additivity

This is a property of any linear decomposition. If you can't get the signal back, then the decomposition is wrong. Question of logic. Any analog can be decomposed into sine waves. The goal here is probably different. To isolate, separate price movements by criteria of external influences. Then it makes sense. And the possibility of a more meaningful analysis of price behavior and forecast appears.

[mytarmailS]

Valeriy Yastremskiy:

This is a property of any linear decomposition. If you can't get the signal back, then the decomposition is wrong. Question of logic. Any analog can be decomposed into sine waves. The goal here is probably different. To isolate, separate price movements by criteria of external influences. Then it makes sense. And the possibility of a more meaningful analysis of price behavior and forecast appears.

What does it mean?

[Valeriy Yastremskiy]

ReTeg Konow:
Well, to begin with, digitized fundamental data immediately becomes an object of technical analysis.

We have different ideas about technical and fundamental. Fundamental = real data expressed in socially-established criteria. Gasprom's balance sheet is fundamental data. Stock sales and quotes are technical. Not convincing, just explaining as I understand it.

[Valeriy Yastremskiy]

mytarmailS:

What does it know?

The price movement is determined by many factors, isolate the main factors and tie them to the price change by decomposing it. In fact, this is what is done in the work. As far as I understand, the same types of price changes are distinguished and separated by means of the selection of their decomposition. You can not understand the factors and do not know their essence, but understanding that these factors can be separated and taken into account gives a lot.

[Evgeny Dyuka]

mytarmailS:

Great job!!! What new things have you added?

Lots of new stuff. New advisor, forecasts now 8 to choose from, new neuro gives entry points.