Econometrics: State-space model forecasting - page 9

New comment
 
MetaDriver:

There's no need to trend. This is your left-wing whim. Trade one bar to start with. Without the spread, that's enough. If there is a clear profitability in these conditions - there will be something to talk about further.
Yes, it's an interesting idea.
 
MetaDriver:
Well, all my forecasters are initially one-bar. It's another thing: it's too far from 100%...) And where else would you make a "pure experiment" with bar colour (trade direction)? How else can you provide a 100% prediction of the direction without preliminarily reading the history?
Real. I agree not 100%, only b positive MO
 
TheXpert:
Better yet, tell me, is there a curwafitter in the stash? )


I do)) The aftar has it, I don't know.
 
EconModel:
Yes, the thought is interesting.

Have you checked the error for a normal distribution? This is a standard econometric requirement for predictive models
 

Test the grail in the tester. Quality. Expensive )


Avals:
I have it)) The aftar has it, don't know.

Bummer... Don't tell me that on a clean graph...

Otherwise I'll have to tear up my template that it's impossible on pure graphics.

______________________

Here's TC's picture, it shows perfectly that it's a fit.


 
Avals:

is there a graph of the error distribution? If it's as thick-tailed as the incremental distribution, then f*ck it.)

P.S. you can even put them on the same graph for clarity


> summary(forecast.residuals)

Min. 1st Qu. Median Mean 3rd Qu. Max.

-3.353e-03 -3.636e-04 3.888e-05 5.263e-05 5.883e-04 3.181e-03

What's his tail?

There seems to be a weekly cycle.....

 
EconModel:


> summary(forecast.residuals)

Min. 1st Qu. Median Mean 3rd Qu. Max.

-3.353e-03 -3.636e-04 3.888e-05 5.263e-05 5.883e-04 3.181e-03

What's his tail?

There seems to be a weekly cycle.....

frequency distribution to see if the error is normally distributed
 
Avals:
Did you check the error for a normal distribution? It's a standard econometrics requirement for predictive models

Well, normality is a bit much, but stationarity....

Missing..... Will do.

The choice of models was by AIC...

 
FAGOTT:
real. I don't agree 100%, just positive MO

Don't make me laugh. I don't agree less than 100%. That's the context in which you called trading on bar colour"myth". Your sentence takes you completely out of that context, because then you cannot guarantee any "positive expected payoff" at all, because it only exists in the tester (on the history), and is only known retroactively on the real. ))
 
Avals:
frequency distribution to see if the error is normally distributed

I don't have enough knowledge to do that straight away.

Actually, as I recall, you have to check for a unit root.

12345678910111213141516...26
New comment
Reason: