Machine learning in trading: theory, models, practice and algo-trading - page 2676

 
vladavd #:

Well, subtract the trend, normalise the amplitudes and you're left with a sine wave(s).

Using the same conditional commodity as an example: we stock the warehouses for production - the price rises, we stock up - it stopped growing or started to fall. After a while, having sold out stocks, we return to the market and sell again - the price rises again. It is clear that in reality everything is much more complicated, but the basic model is exactly the same.


Well, there is an understandable factor here - we sell goods and the price goes up. It can be taken into account in the formula and track the dynamics. And price change is a reaction. In Forex, then the signal will be something similar, but not the chart itself, which simply reflects the exchange rate and does not carry information about the future

And TS are built not on forecasting, but on errors (inefficiencies) of various kinds. But there is no signal in them, as well as in the rest of the chart. They can be filtered out somehow, if they exist, but it is very difficult in an efficient market
 
Maxim Dmitrievsky #:
Well, there is an understandable factor here - we tare goods, the price goes up. It can be taken into account in the formula and track the dynamics. And the price change is a reaction. In Forex, then the signal will be something similar, but not the chart itself, which simply reflects the exchange rate and does not carry information about the future

I don't see any fundamental difference between a concrete factory that trades sand for production and an importer of electronics that trades currency to buy goods abroad. Yes, different markets, but the physics of the process are the same.

"Something similar" from different participants adds up to a common signal. The goods are bought not by one player, but by many. Some part of this set acts on one time horizon, thus drawing a more or less stable signal in the price.

 

Maxim Dmitrievsky #:

And TS are built not on forecasting, but on errors (inefficiencies) of various kinds. But there is no signal in them, as well as in the rest of the chart. You can somehow filter them out, if they exist, but it is very difficult in an efficient market

Difficult, no argument. Is it easy to find inefficiencies in an efficient market?)

 
vladavd #:

I don't see any fundamental differences between a reinforced concrete factory, which trades sand for production, and an electronics importer, which trades currency to buy goods abroad. Yes, different markets, but the physics of the process is the same.

"Something similar" from different participants adds up to a common signal. The goods are bought not by one player, but by many. Some part of this set acts on one time horizon, thus drawing a more or less stable signal in the price.

So the signal was before, and the rate changed after. Here is the lead-lag relationship. Then you should go to fundamental analysis, but there is nothing to extrapolate by sinusoids there, but I would like to, of course. And fundamentalists are not happy with their work 😀
 
vladavd #:

Using the same conditional goods as an example: we stock the warehouses for production - the price rises, we stock up - it stopped growing or started to fall. After a while, having sold off stocks, we return to the market and sell again - the price rises again. It is clear that in reality everything is much more complicated, but the basic model is exactly the same.



The goods are purchased at certain price levels, for this purpose time series are analysed (price changes at suppliers), then we buy from who is cheaper and then sell to the one who has a higher price. And the warehouse does not need to be maintained 😄. I see it like this ))

 
vladavd #:

Difficult, no argument. But is it easy to find inefficiencies in an efficient market?)

Usually they are found on a hunch and don't live long ).
So, since the topic is about MO, I like all sorts of going through the options, sometimes quite good to find them
 
Give up on sine waves if you know what to do with them)
 
mytarmailS #:
Leave sinusoids alone if you know what to do with them.)
That's what we're asking what to do with them.
Personally, they're not in my market picture.
 
Maxim Dmitrievsky #:
So we're asking what to do with them.
Personally, in my picture of the market, they are absent.
You detrend the price, even with Mashka and approximate/approximate it with two sinusoids using ISC.

The resulting simple model is already predicted in some way.
 
mytarmailS #:
Detrend the price, even with Mashka, and approximate/approximate with two sinusoids by ISC

The resulting simple model is already predicted in some way.
There's one, loess regression, I think. It does it better. Detrending should be done leaving maximum information, I already wrote above