Making a Python trading system for MT. - page 14

 
Maxim Dmitrievsky:

If the lines weren't overdrawn too much, it would be fine.

I don't care about that at all. When trading, we're not interested in history, but in the future, to a certain extent).

History has always been rewritten )), and what matters is the interpretation of history now, and not what it was at the time of events. We can take an example from our recent past, about 20 years ago. We know much more about the past now.

PS Period 3000. Degree of polynomial regression -3, tried up to 5, but it does not change or give anything, the curve is about the same.

 
Yuriy Asaulenko:

It doesn't matter at all. When trading, we are not interested in history, but in the future, to an extent).

History has always been rewritten everywhere), and what matters is the interpretation of history now, not what it was at the time of the events. We can take an example from our recent past, about 20 years ago. We know much more about the past now.

If we consider from "now" then it will not be the same distribution, the tails will greatly increase and the average will shift. That is why I suggested to build from 200 regressions and to average the result, i.e. to look as if in dynamics. The link to Kalman regression above is about the same thing

The funny thing about non-stationarity is that whichever way you look at it, it doesn't become more stationary, sometimes it may seem
 
Maxim Dmitrievsky:

If we take into account the "now", it will not be the same distribution, the tails will increase a lot and the average will be shifted. That's why I suggested to build from 200 regressions and to average the result, i.e. to look as if in dynamics. Reference to Kalman regression above is about the same thing.

So all sorts of MAs as a regression and take into account from now, all the way through, at every point in time. And the regression line is sort of a hospital average over the period). - The present is on an equal footing with the past with equal weights - sort of good and evil being heeded indifferently).

By the way, the regression has settings for weights on the sample interval. At the nearest future we can increase the weights. We'll get a regression MA).

 
Yuriy Asaulenko:

It doesn't matter at all. When trading, we are not interested in history, but in the future, to an extent).

History has always been rewritten everywhere), and what matters is the interpretation of history now, not what it was at the time of the events. We can take an example from our recent past, about 20 years ago. We know a lot more about the past now.

PS Period 3000. Degree of polynomial regression -3, tried up to 5, but it doesn't change or give anything, the curve is about the same.

The image looks interesting. What is 3000 ?

 
Yuriy Asaulenko:

So all sorts of MAs as a regression and take into account from now, all the way through, at every point in time. And the regression line is like a hospital average over a period). - The present is on an equal footing with the past with equal weights.

By the way, the regression has settings for weights on the sample interval. The nearest future can have the weights increased. We will get the regression MA).

Well this average jumps on every bar, i.e. the centre of the bell will be wider or something like that, i.e. we need to take a bit more margin to make it work

That's what the posterior is for, but yours is more like an a priori, i.e. you assume that it is

in short it is time to sleep ))
 
Maxim Dmitrievsky:

Well, this average jumps on every bar, i.e. the centre of the bell will be wider or something like that, i.e. you have to take a margin to make it work

That's what the posterior is for, and yours is more like an a priori.

Here at MA it is wider, because she is very partial to the present and stays that way forever. Look at the first picture and estimate errors of MA relative to the price chart. And I've picked it up, it doesn't get any better than that.

Try to build an a priori on a regression. I'll see what you get.)) No way it's a linear regression predictor, maybe only short term, haven't looked at it.

 
Yuriy Asaulenko:

distribution relative to the regression line at 3000 counts. See post with picture.

The counts of what, excuse me?

 
Yuriy Asaulenko:

The price range.

You mean historical ticks?

 
Yuriy Asaulenko:

It is wider on the MA, because it is very partial to the present and will remain so forever. Look at the first picture, and estimate the error of the MA relative to the price chart. And I've picked it up, it doesn't get any better than that.

Try to build an a priori on a regression. I'll see what you get.)) There's no way it's a linear regression predictor.

Well, I'm not applying any of that, so I'm just observing what's going on.

I just have some vague ideas, that's all.

 
Yuriy Asaulenko:

No, this is Close 1M candles. I don't use ticks on history.

In that case, feel free to set it to 1 instead of 3000.

Forum on trading, automated trading systems & strategy testing

How to create Python based trading system for MT.

Yuriy Asaulenko, 2019.01.23 23:49

I don't care about that at all. When trading, we are not interested in history, but in the future, to some extent).

History has always been rewritten everywhere) and what matters is the interpretation of history now, not what it was at the time of events. We can take an example from our recent past, about 20 years ago. We know much more about the past now.

PS Period 3000. Degree of polynomial regression -3, tried up to 5, but it doesn't change or give anything, the curve is about the same.

That's it, gone - no offence)