Do you know how to make canals? - page 7

[СанСаныч Фоменко]

Alexey Volchanskiy:

Friends, there's almost no trade, it's time to get into theory. Having drawn a fun picture, let's discuss working in a channel.

My humble opinion, the channel is an auxiliary tool and serves to confirm a signal received in some other way.

HP channel

All these channels, trends and so on are bullshit. it all looks good on history, and the future is hidden behind the fog of NESTATION.

Or we always remember about non-stationarity and look for tools against it, which we then apply, or we lose our deposit.

1. we trade according to patterns (a channel is also a pattern). We take TA + brains (experience) - the most promising, and maybe win. Or we take MO, automatically look for patterns... and then we have to find input data, which should again generate stable patterns for the target variable. The main problem is not the pattern search algorithm, but the ability to find input data for these patterns. Experience shows that there are about 30 such input data (multivariate). On this number, in principle, it is possible to search for multivariate channels. Is this necessary?

2. Statistics (Toolbox"Econometrics" in Matlab). GARCH. Convert original series to stationary, now in three steps. Till the end, NO ONE has managed to obtain a stationary residual from the model. And if the residual is non-stationary, there is always a situation that drains the depo.

[Alexandr Murzin]

СанСаныч Фоменко:

Until the end, NO ONE has managed to achieve a stationary residual from the model. And if the residual is unsteady, there is always a situation that drains the depo.

You have veiled the phrase "we will all drain" in such a difficult way.

[Forester]

СанСаныч Фоменко:

The main problem is not the pattern finding algorithm, but the ability to find the raw data for these patterns. Experience shows that there are about 30 such inputs (multicurrency).

Can you be more specific? Which 30 input data contain patterns?

[Maxim Dmitrievsky]

Here's the evolution of models from linear to garbage

maybe someone needs to use python to solve examples and see why channels don't work

The basic premise is that anything non-linear (non-stationary in the sense of the word) is underdetermined because it does not converge to the mean... everything is so trivial that it is hard to believe that econometrics has gone beyond Bernoulli's ideas or whatever... Gauss ideas.

http://www.blackarbs.com/blog/time-series-analysis-in-python-linear-models-to-garch/11/1/2016

[Maxim Dmitrievsky]

Viktor Korchagin:

that's how the trade test went

So automate it.

[Maxim Dmitrievsky]

Viktor Korchagin:

I'm not a programmer, the trade was done on sh4 so you can also trade with your hands slowly

so we can't trust your result as it might be random :)

[Aleksey Ivanov]

СанСаныч Фоменко:

Bullshit all these channels, trends and so on... It all looks good on history, but the future is hidden behind the fog of UNSTABILITY.

Either we always remember about non-stationarity and look for tools against it, which we then apply, or we lose our deposit.

1. we trade according to patterns (a channel is also a pattern). We take TA + brains (experience) - the most promising, and maybe win. Or we take MO, automatically look for patterns... and then we have to find input data, which should again generate stable patterns for the target variable. The main problem is not the pattern search algorithm, but the ability to find input data for these patterns. Experience shows that there are about 30 such input data (multivariate). On this number, in principle, it is possible to search for multivariate channels. Is this necessary?

2. Statistics (Toolbox"Econometrics" in Matlab). GARCH. Convert original series to stationary, now in three steps. Till the end, NO ONE has managed to obtain a stationary residual from the model. And if the residual is non-stationary, there is always a situation that drains the depo.

Completely agree. Channels, trends are by and large a posteriori and are simply our usual way of making sense of an already established story. Moving probability distributions need to be calculated - this will give more reliable information. But here too, non-stationarity confuses the cards.

[Artyom Trishkin]

Viktor Korchagin:

Your right...I'm not claiming anything))) but the point I've made, you can write an owl...it will be interesting to see the results

[Alexander_K2]

Aleksey Ivanov:
I totally agree. Channels, trends are by and large a posteriori and are simply our usual way of making sense of an already established story. Moving probability distributions need to be calculated- this will give more reliable information. But here too, non-stationarity confuses the cards.

Exactly. I've been telling how to do it on my thread for 2 months now. And some of the most sophisticated people are completely stupefied in these matters. They're clueless, to put it simply. It's time for them to play dominoes :))))

[СанСаныч Фоменко]

Aleksey Ivanov:
I totally agree. Channels, trends are by and large a posteriori and are simply our usual way of making sense of an already established story. Moving probability distributions need to be calculated - this will give more reliable information. But here too, non-stationarity confuses the cards.

This has been done in GARCH for about 15 years. But before that there are two more steps for increments: a trend model in increments and a volatility model (GARCH - clustering volatility first, but there are lots of other nuances). And then the moving density is calculated and modeled usually by t-distribution. If we look at the history of development of GARCH models, the efficiency of these models has improved dramatically exactly after probability density modeling. So, you can't do without it.