Machine learning in trading: theory, models, practice and algo-trading - page 657

[Forester]

Maxim Dmitrievsky:

log(close[i]/close[i-15])

Where to compress what, why?

I also thought that through subtraction.

[Forester]

Maxim Dmitrievsky:

log(close[i]/close[i-15])

Where to compress what, why?

It's more of a ratio, not an increment.

[Maxim Dmitrievsky]

elibrarius:
I also thought that through subtraction.

the logarithm is not defined with negative arguments, why subtract

and SanSanych with division, I showed him that he is wrong about the removal of emissions by such logarithm

[Yuriy Asaulenko]

SanSanych Fomenko:

This hypothesis of an efficient market is nonsense.

It's a moot point.) But I'm not talking about market efficiency, I'm talking about the observer.

For example, a car pulls up to a T-intersection. Where it will turn we absolutely do not know. And the driver knew it at least half an hour ago. Although the process is completely deregulated, for the observer the process is random - any prediction is absolutely impossible. We can say absolutely nothing about its further behavior by the wiggle-shift of the car and other patterns.

This approach can be called informational. Any information is random for the recipient, otherwise it is not information).

[СанСаныч Фоменко]

Dr. Trader:

...and you can't do it at random.

That's for sure. I killed more than 2 months and no result.

I've got an idea to create some indicators and use them for manual trading.

I believe garch can do a lot, but the amount of time I have to invest in order to understand it is too much, and there's not much of it.

I can't agree. You have to take the road, not the vegetable garden. It may take many times longer, but it's on the road.

[Forester]

Maxim Dmitrievsky:

the logarithm is not defined with negative arguments, why subtract

and SanSanych with division, I showed him that he was wrong about removing outliers by such logarithm

For negative numbers you can *-1, then logarithm and then *-1 again

This way you can center and smooth out the outliers.

[Maxim Dmitrievsky]

elibrarius:

For negative numbers, you can *-1, then logarithm, and then *-1 again.

let's compare with subtraction)

[Maxim Dmitrievsky]

Maxim Dmitrievsky:

Let me compare it to subtraction.)

That doesn't make any sense.

regular

log

[Forester]

Maxim Dmitrievsky:

It doesn't make any sense.

regular

log

did you do that?

double delta=close[i]-close[i-15];

double log_delta=(delta>0?log(delta):log(delta*-1)*-1;

should go around zero.

[Maxim Dmitrievsky]

elibrarius:

did you do it like this?
double delta=close[i]-close[i-15];

double log_delta=(delta>0log(delta):log(delta*-1)*-1;

should go around zero.

for(int i=start;i<rates_total;i++) 
     {
      bool invert = false;
      double pr = close[i]-close[i-ReturnsPeriod];
      if(pr<0)
       {
        pr = pr*-1;
        invert = true;
       }
      double pr2 = log(pr);
      if(invert) pr2 = pr2*-1;
      ReturnsBuffer[i]=log(pr2);
     
     }