Econometrics: State-space model forecasting - page 10

New comment
 
MetaDriver:
Don't make me laugh. I disagree less than 100%. That's the context in which you called trading on bar colour a "myth". Your sentence takes you completely out of this context, because then you cannot guarantee any "positive expected payoff", because it only exists in the tester (on history), and is only known retroactively on the real market. ))

Well, okay, I agree to trade with 100% accuracy - prove it
 
EconModel:

There's not enough knowledge that immediately.

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

And why check the error for stationarity? For a normal distribution.
 
EconModel:

What kind of tail does it have?

?hist, ?density, ?quantile

A test at 100 bars is ridiculous. Do it at 2-3,000 at least.

 
EconModel:

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

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


I can see everything.)
 
EconModel:

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

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


Avals:
frequency distribution to see if the error is normally distributed

Here is the unit root test # Augmented-Dickey-Fuller Unit Root Test:

Value of test-statistic is: -7.0173


Critical values for test statistics:

1pct 5pct 10pct

tau1 -2.6 -1.95 -1.61

As I understand it, the residual is stationary.

 
EconModel:

Here is the test for unit root # Augmented-Dickey-Fuller Unit Root Test:

Value of test-statistic is: -7.0173


Critical values for test statistics:

1pct 5pct 10pct

tau1 -2.6 -1.95 -1.61

As I understand it, the residual is stationary.

I don't know what these tests show. The distribution shows either an overshoot and thick tails like the price increments, or a fixed small variance without outliers beyond a few sigmas.
 

The question is an old one. What to do about the forecast.

So far I see one working suggestion - to trade one bar. Really no spread. Doubtful.

 
Avals:
I don't know what these tests show. We can see from the distribution - either like price increments - spiky and thick tails, or a fixed small variance without outliers over a few sigmas

I will try to draw it. I just don't know how to do it, I haven't used it. You can't attach a picture to the algorithm. And there are a lot of tests for the unit root, specially invented. and so: thick-not quite thick....
 
EconModel:

The question is old. What to do with the forecast.

So far I see one working suggestion - to trade one bar. Really no spread. Doubtful.

You open towards the forecast if the target is greater than X*spread. On the next bar, if the forecast is the same as the position, we hold (saving spread over re-entry). If in the opposite direction, we close
 
Avals:
you open towards the forecast if the target is greater than X*spread. On the next bar, if the forecast coincides with the position - you hold (we save the spread compared to re-enter). If in the opposite direction, we close

Yes, this is a refinement of the metadriver.

Pardon, so this result is according to the anonymous tester. Then we should double-check in the tester and see....

1...34567891011121314151617...26
New comment