Strategic foresight systems - page 49

New comment
 

The development of my system is ongoing. Now the forecast for the first date would look as follows.

It is noticeable that it has become less error prone. I now have promising opportunities to further improve the prediction algorithm, but I am in desperate need of a next-generation processor (at least 10 times faster), as the computational performance of a desktop computer is severely lacking for the parts that are not partitioned out. Good luck!

 
Try using, for example, Matlab. You can do some things much faster there.
 

Can your forecasting system give a prediction (picture with visible numbers on the right) for the Euro ?

Thanks !

 
tuma88:

Can your forecasting system give a prediction (picture with visible numbers on the right) for the Euro ?

Thank you !

As an experiment, the Euro forecast for the week ahead - 10 closing prices of 12 hour counts.

1.36986 1.3774 1.38367
1.36583 1.37757 1.38559
1.35863 1.37418 1.38722
1.3611 1.37583 1.38763
1.35956 1.37359 1.38514
1.35864 1.37679 1.39177
1.36065 1.37691 1.38847
1.35988 1.37611 1.38839
1.35874 1.37584 1.38731
1.35558 1.37225 1.38757

Looks like a corridor move.

 

1.3910 on the Eurochka is my prediction. You have 39177 .

A week until March 8 means .

Let's see how it goes .

 
tuma88:

1.3910 on the Eurochka is my prediction. You have 39177 .

A week until March 8 means .

Let's see how it goes .

I don't have 39177 listed. The left and right columns are range boundaries. The forecast is the middle column. Accordingly, at the end of the week it is the last value of 1.37225. And what you wanted to look at, I do not understand - a difference of 7 points?
 

Correcting the message of 01.03.2014. Unfortunately, there was an error in the system, so I am re-posting the forecast.

1.37047 1.37818 1.38456
1.36552 1.37821 1.39027
1.35869 1.37351 1.38861
1.35968 1.37497 1.3899
1.36048 1.37673 1.38842
1.3614 1.37647 1.39128
1.35878 1.37523 1.39167
1.3553 1.3747 1.38961
1.35441 1.37619 1.3924
1.35525 1.37925 1.39269

Work on the pronouncing system continues :) Where have all the mat forecasting enthusiasts disappeared to?

 
-Aleksey-:

Correcting the message of 01.03.2014. Unfortunately, there was an error in the system, so I am re-posting the forecast.

1.37047 1.37818 1.38456
1.36552 1.37821 1.39027
1.35869 1.37351 1.38861
1.35968 1.37497 1.3899
1.36048 1.37673 1.38842
1.3614 1.37647 1.39128
1.35878 1.37523 1.39167
1.3553 1.3747 1.38961
1.35441 1.37619 1.3924
1.35525 1.37925 1.39269

Work on the Prono system continues :) Where have all the mat forecasting enthusiasts disappeared to?


I do not understand why I should make predictions about the value of the currency pair, for example 1.35525. What are we trading? We have orders to buy and sell, and we have a value, a specific figure, apart from that, I do not see what error this value has been calculated with. Maybe, the error is so large, e.g. 100 pips, that the forecast is not even a forecast anymore.
 
I forgot to specify that the forecast is made for a 90% probability of finding values within the confidence interval, i.e. 1 in ten closing prices is entitled to be outside the confidence interval. The left column shows the lower confidence interval and the right column shows the right, while the centre column shows the predicted values themselves. They are shown in the figure. Thus, an error upwards is 150 pips, and downwards - 250 pips, for a total of 400 pips. We trade the forecast value, because it is the most probable, while the probability to be at the range borders is lower (i.e. according to the forecast sum the actual value will be closer to the prognostic values more often, and less often will be closer to the borders of these values). The error is large, I'm working on it. How to trade a forecast, there are variants, for example, when the distance from the current price to the forecast value is equal to the distance from the forecast price to the limit of the confidence interval (stop). It is clear that in this case the probability shift will be more in our favour than if the intervals from the opening price to both limits of the confidence interval were equal. The reason for setting a stop at the border of the confidence interval is such that either the least probable scenario took place (with the probability of 10% the price passed it (showing a possible rapid change in the forecast model), or the algorithm incorrectly detected the forecast model). In both cases, the position cannot be held further. In the example above, having opened the position on the opening day of the 3rd day and set tp to the forecasted 1.37925 and stop at 1.35525, we could gain 33 pips. Something like that, i.e. probabilities are traded.
 
-Aleksey-:
I forgot to make it clear that the forecast is made for a 90% probability of finding values within the confidence range, i.e. 1 in 10 closing prices has a right to be outside the confidence range. The left column shows the lower confidence interval boundary and the right column shows the right column and the centre column shows the predicted values themselves. They are shown in the figure. In other words, it is 150 pips upwards and 250 pips downwards, for a total of 400 pips. We trade the forecast value, because it is the most probable, while the probability to be at the range borders is lower (i.e. according to the forecast sum the actual value will be closer to the prognostic values more often, and less often will be closer to the borders of these values). The error is large, I'm working on it. How to trade a forecast, there are variants, for example, when the distance from the current price to the forecast value is equal to the distance from the forecast price to the confidence interval boundary (stop). It is clear that in this case the probability shift will be more in our favour than if the intervals from the opening price to both limits of the confidence interval were equal. The reason for setting a stop at the border of the confidence interval is such that either the least probable scenario took place (with the probability of 10% the price passed it (showing a possible rapid change in the forecast model), or the algorithm incorrectly detected the forecast model). In both cases, the position cannot be held further. In the example above, having opened the position on the opening day of the 3rd day and set tp to the forecasted 1.37925 and stop at 1.35525, we could gain 33 pips. Something like that, i.e. probabilities are traded.

Yes, a very important clarification.

So, the confidence interval = 400 pips. It is generally accepted that the confidence band is the range where the price will fall in your case with 90% probability. By definition, any price within the confidence range is considered equally likely to occur. If we trade within a day, a 400 pips prediction error is many times greater than a daily price movement, i.e. the error is greater than the potential profit and we are back to the sad fact of fifties.

If you have some considerations on the distribution of the error inside the confidence interval, you can improve the system and narrow the confidence interval. But that would be a different system.

1...424344454647484950
New comment
Reason: