Machine learning in trading: theory, models, practice and algo-trading - page 982

 
Aleksey Vyazmikin:

Why is it such a fashion to test on older data than the training?

About the strategy, I propose to work together on my basic trend strategy - look for an entrance, and the exit is predetermined - a trawl on the Donchian channel.

i don't care what i'm testing with

I've already played with trawls, i'm not interested.

 
Maxim Dmitrievsky:

I don't care what to test on

I've already played with trawls, I'm not interested.

Why not interested? I mean, what's the problem? The signal for the output is defined - the appearance of the beam ZZ. It only remains to deal with the inputs.

 
Aleksey Vyazmikin:

Why aren't you interested? I mean, what is the problem? The signal to exit is defined - the appearance of the ZZ beam.

It's a feather in the cap strategy because there are no fundamental patterns.

The only thing that works is training on ticks of different brokers, like arbitrage. It's true it is too labor-intensive to train on ticks.

 
Aleksey Vyazmikin:

Why is it such a fashion to test on older data than the training?

About the strategy, I propose to work together on my basic trending strategy - look for an entry, and the exit is predetermined - a trawl along the Donchian channel.

The best way is to trawl in the middle and to the right and to the left are untrained areas.

 
I cannot make up my mind about it:

I don't care what to test on

I've already played with trawls, I'm not interested.

If there are steady trends there, they do not last long.

 
forexman77:

If there are steady trends there, they do not last long.

I agree, the option is to train on indices while they are going up, but the trend is that they may fall soon) or to train pair trading between correlated indices

 
Maxim Dmitrievsky:

I agree, as an option to train on the indexes while they are rising, but now the trend is that they may soon be falling...) or to train in pair trading between indexes that correlate.

This is my point: trend systems do not work in forex, maybe something in between. Plus figure out how to extract patterns from the price series and use them as predictors. Just haven't come up with anything yet.

Now I tried to increase the predictors by three times, but nothing has improved. Then I tried to exclude the predictors by their importance in several steps - again, nothing happened. Almost the same.

 
forexman77:

I mean, trend systems don't work in forex, you probably need something in between. Plus, I have to figure out how to extract patterns from the price series and use them as predictors. Just haven't come up with anything yet.

It's funny to look for patterns in an unsteady process, when they are constantly changing ) I haven't even tried

NS works smoothly on cycles, when they change it stops working ) don't even need any predictors.

Picking the right instrument, situationally, may give a profit. But indexes are no good now

I think it's a good idea to teach pairs trading...

 
Maxim Dmitrievsky:

It's fun to look for patterns in an unsteady process when they are constantly changing )

Some I think you can look for. Just some special approach is needed, not like everyone is looking for in RSI. For example, the same cycles as an example.

 
forexman77:

Some I think you can look for. Just need some special approach, not as everyone is looking for in RSI. For example, the same cycles, as an example.

i just take the top more or less cooing bots in the market and try to understand the stratum and teach about the same )) there's neuronics too

there's also neurons there too. of course they mostly suck on martin.