Econometrics: one step ahead forecast - page 38

New comment
 
-Aleksey-:
I am interested in this question. Even if you mature for tester checks and the result is 51/49 simplistically in trades and 51/49 in pips to the plus, you will have to wait about 100 trading days to realize the meager stat. advantage. To increase the number of trades and shorten this timeframe - you have to move to a smaller timeframe. There manually using your favourite programme will be inconvenient, because it is often. Actually, my question is, are you going to implement all the E-views algorithms you are using in your trading robot's code, if you find some suitable model? Are you willing to spend a couple of years on this?
It's been implemented. The code is attached to the article.
 
faa1947:
This has been implemented. The code is attached to the article.
Thanks, found it in the code. You run E-views MQL program - that's handy, no questions then.
 
-Aleksey-:
There you have export of quotes and reading of results. E-views program itself should be started manually or it works in automatic mode, reading data and writing the results after a certain period of time, or not in time? How is the automatic mode organised? Just not familiar with the program.

Starts up automatically. Read the article. There is a test result in the MQL4 tester there.

If it's not there, the question was about transferring everything E-views does to MQL code or C++ library...

The question is about porting in any case. Only now we don't know WHAT to port. You have to build the model first.

 

faa1947:

It is a question of rescheduling in any case. Only at the moment you don't know WHAT to transfer. The model has to be constructed first.

Now again it is not clear, if there is automation - why the question of porting?
 
-Aleksey-:
Now again it is not clear, if there is automation - why the question of portability?
EViews is a terribly slow wagon.
 
faa1947:
EViews is a terribly slow wagon.
Can you give an example in terms of slowness in figures - interesting...
 
-Aleksey-:
Can you give me an example in terms of slowness - interesting...
If we're talking about the specific EViews call scheme in the article. Running a model on 118 bars takes me up to 5 minutes. Optimisation is not an issue
 
faa1947:
If we talk about the specific EViews calling scheme in the article. Running a model on 118 bars takes me up to 5 minutes. Optimisation is not an issue
And when you move to a smaller timeframe you have to somehow predict for a few steps, otherwise, on the minutes the one-step prediction may be in the pre-spread. How do you plan to solve this problem - both speed and multiple steps?
 
-Aleksey-:
And when you move to a smaller time-frame you will have to predict a few steps somehow, otherwise a one-step forecast on the minutes may end up in the spread. How do you plan to solve this problem - both speed and multiple steps?

The number of prediction steps is a result of calculation, not desire. Kotier might not be predictable at all at a given point and you will have to wait until it becomes predictable.

The threat from the other side. Forecast error becomes comparable to candle length. This is already showing up on H1.

 
faa1947:

The number of prediction steps is a result of calculation, not desire. Cotier may not be predictable at all at a given point and will have to wait until it becomes predictable.

The threat on the other hand. The forecast error becomes comparable with the candlestick length. It is already showing on H1.

What do we care about the error value? It may be 1000 points. You can regulate its use and size yourself - you can enter not by bar opening, but by a pending order near the error boundary. On the average, in a great amount of transactions, the price will be closer to the predicted value, i.e. it will move away from your order to the profit. Only the border of the confidence interval would go up or be horizontal - to buy. But you should make a forecast for at least 500-1000 steps to estimate the boundary of the confidence interval in order to use this method. I.e. minutes at any rate, otherwise there will be very few deals. And you need performance, otherwise the locomotive will run away until you do the math :)
1...313233343536373839404142434445...139
New comment