Testing real-time forecasting systems - page 65

 

Buy again, target 5702, stop at 5612 area

 

Let's move the stop to 5635 according to the indicator https://www.mql5.com/ru/code/9310:

Further, in order not to clutter the thread, I will write only the initial stoploss, and whether or not to transfer it to a new level (according to the updated values of the indicator SL_to_Bar), your business.

Forecasts will be published once a day (if possible) before the opening of the market.

 

The position closed on the take:

P.S. Trading is done with 0.1 lot.

 

As of today, the picture is clearly unequal:

Therefore, I would not open any positions.

 

Holiday is over and it's time to get back to work:

  • A few thoughts on entropy. Have come to the final conclusion that we should focus on the maximum values of this parameter as a criterion for selecting the future implementation of the process. The idea of selecting the entropy "level" depending on expected news does not seem to work.
  • The forecast for the next week from 16 to 20, so far the same: EURUSD, M15, forecast for 500 samples (100 samples about a day). I'm testing an alternative process identification module. So I don't recommend to use it as an additional material for decision making. The behaviour of the forecast was a surprise to me myself.

Estimation of "hitting" the trajectory in the range gave me a value of about 20, in other words, I need to calculate at least 20 trajectories in a certain way, and the future implementation with a high probability will be among them. So, not to make a mistake - I calculated 33 trajectories and ... and was surprised. These trajectories "converged" on three, i.e. "self-organized" three clusters:

Entropy of trajectories (by number):

First cluster (two trajectories), entropy 145.7

Second cluster (30 trajectories), entropy 108.02

Third "cluster" (only one trajectory), entropy 149.14

Objectively, the first and second cluster are equally likely.

Who chooses which trajectory? Just curious to see an estimate of the plausibility of the prediction, not by algorithms and mathematical methods, but by human evaluation (as Bender said) :o)

PS: by the way, out of curiosity, "rolled back" two days and made a prediction, quite curious, the system estimated the chances of the trajectory to the 1.6 level as very high, but today, the chances have been reduced and the 1.6 level is considered unlikely in the near future.

Files:
d.rar  16 kb
 
grasn >> :

Thanks for the info. A typical state at the point of achievement. What next?

As a virtual trader - I also see two scenarios. three more weeks up. or already a collapse.

Out of this point we go. and not just on this pair.

A good indicator of the process now is gold and oil.

 

grasn, the first and third trajectories seem more plausible!

Thanks for the benchmark for next week.

 

to Sorento

"спасибо за инфо. Типичное состояние в точке достижения цели."

Not a professional mathematician or physicist, which of course is bad, but just an amateur. But it seems to me, to reason about bifurcation points just like that is a bit abstract, because these points are largely determined by the chosen model, and in the link, there is no mention of any model.

"As a virtual trader - I see two scenarios too. three more weeks up. or already a collapse.

From this point we leave. and not just on this pair."

We may be close to a downward reversal, the most likely trajectories are exactly that, but show different downward dynamics.

 
mpeugep >> :

grasn, the first and third trajectories seem more plausible!

Thanks for the benchmark for next week.

Thanks, but just in case - the identification module is being tested in its raw state, I don't recommend trusting it completely.

 
grasn >> :

to Sorento

these points are largely determined by the model chosen, and the link doesn't say anything about any model.

>> Exactly! Can you build a model from the points?