Statistics as a way of looking into the future! - page 5

 
m_a_sim писал(а) >>

can Close[i]-Open[i] be considered as an increment?

No, you have to count Open[i]-Open[i-1].

 
because Close[0] has not yet been formed?
 

Yeah.

Also, it is a way to insure against unauthorized "peeking" into the future - when the bar has not yet formed and we already "know" where it will end - the alma mater of all sorts of Forex grails. For the same reason, we cannot use time series like (Open+High+Low+Close)/4 or similar for making forecasts. They are perfectly "forecastable", which provokes interest, and when you try real trading, they lead to loss.

 

I would say that mathematical statistics is an indispensable element of a profitable ATC. There is more confidence in exact objective science than in various analytical approaches. I.e. for example to a system based on TA and positively tested for a significant period - less credibility, than to a system based on statistics, not necessarily positively tested, but determining a significant probability of profit in the next period.

So, mathematical statistics is like a mother for a trader. It will teach and feed you and "land" you from the clouds to the ground.

 

m_a_sim писал(а) >>
невооруженным глазом видно, как цена золота "тянется" за регрессией - в каком месте это видно? На каком баре смотреть?

 
Neutron писал(а) >>

Yeah.

Also, it is a way to insure against unauthorized "peeking" into the future - when the bar has not yet formed and we already "know" where it will end - the alma mater of all sorts of Forex grails. For the same reason, we cannot use time series like (Open+High+Low+Close)/4 or similar for making forecasts. They are perfectly "forecasted", than provoke interest, but at an attempt of real trading they lead to failure.

High and Low are predicted (without quotes) better than Open and Close due to their different nature from the point of view of statistics. The interest is promoted, but knowing the differences puts everything in its place. There is no peeking into the future per se, unless there is trivial peeking. It is appropriate to consider the increments as Close[i-1]-Cose[i].

 
Vita >> :

High and Low are predicted (without quotes) better than Open and Close due to their different nature in terms of statistics. Interest is provoked, but knowing the differences puts everything in its place. There is no peeking into the future per se, unless there is trivial peeking. It is appropriate to consider Close[i-1]-Cose[i] as increments.

Hai and loi are more predictable, it's true.

Приращение: y=(|Close[i-2]-Open[i-1]|)+(Open[i-1]-Open[i])

 
Neutron писал (а) >>

Prival, what is the deeper meaning of considering two values in the model - price and time? Wouldn't it be easier to limit it to one, or would the model suffer from that? Maybe there are some considerations in principle?

  1. These two quantities are included in the formula(t and price), so how can we get rid of them?
  2. The longer the time interval we forecast, the bigger is the error and it is due to

a) model inaccuracy

b) measurement errors

I've posted a picture here 'Tick builders. Optimization. DDE in VB (VBA)' is a simple one, but I think the meaning is clear.

The thing is we are trying to predict the price using a model with no random (stochastic) component, it's too coarse model and it doesn't allow to calculate error ellipsoid. The number of factors for analysis could be increased (consider gold price, metals index, oil, Dow Jones, etc.) but one factor I think should be that there is a Wiener process in the value formation. It is the one that I assume prevents the forecast horizon from increasing significantly, the elipse is increasing too fast.

 
Neutron писал (а) >>

Draw a straight line through this cloud using the method of least squares and see the tangent of its slope. So, at first glance, the result is good! But we need figures. As I understand it, the points are plotted on the axes.

For what instrument and on what TF is the prediction?

I will have to check my "predictor", I've never done that. I will try to show the results tonight.

 
Prival >> :

I'll have to check my "predictor" I've never done it before, I'm curious. I'll try to post the results tonight.

Can you tell me more about your "predictor"?