Testing real-time forecasting systems - page 45
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
What do you build with? Deductor, or do you have your own tool?I use this:'Predicting time series with Deductor Academic 5.2'.
Both the number of runs and failures give reason to be very optimistic about the TS.
It's too early to make final conclusions, but the main movement seems to be catching. Here's another forecast from now for the day:
While writing and making the trade:
>> we'll see. A large-scale test will finally decide everything. That's what I'm getting ready for.
>> Yay! That's the third or fourth failure out of about 80 launches! This needs to be injected! :о)
grasn, I'm not sure that this particular prediction should be interpreted as a failure. The movement and levels seem to have been shown correctly, and the discrepancy was due to timing (it happened much earlier). In this regard, shouldn't we pay attention to the fact that the forecast includes, in fact, 2 coordinates - price and time, and for time it would also be nice to give some tolerance in the form of a confidence interval? I don't know the kitchen, but the determinism of time is alarming ;-). And by and large, the non-linearity of the predicted time should not affect profit - the levels remain.
Grasn, I'm not sure that this particular forecast needs to be interpreted as a breakdown. The movement and levels seem to have been shown correctly, and the discrepancy was due to timing (it happened much earlier). In this regard, shouldn't we pay attention to the fact that the forecast includes, in fact, 2 coordinates - price and time, and for time it would also be nice to give some tolerance in the form of a confidence interval? I don't know the kitchen, but the determinism of time is alarming ;-). And by and large, the non-linearity of the predicted time shouldn't affect profits - the levels remain.
Yes, you're absolutely right, time is the most important component of the system, on which stop levels in particular depend. I am working on it in the system right now.
By the way, here is the forecast for the next week (H+L)/2 EURUSD M15 (each reading is 15 minutes on the chart).
PS: but will need clarification, especially before the "surge", and especially on Monday on the occasion of the possibility of a strong gap.
A little addition - added a "new feature", let's see how it affects the quality of predictions :o)
By the way, here is next week's (H+L)/2 EURUSD M15 forecast (each timeframe is 15 minutes).
Sergey, i.e. do you expect the week to move upwards, but by the end will be to the closing level of the previous week? So a wider and stronger downward channel will be formed? Or will it be a reversal pattern for an extended move upwards? What statistics do you have about it and do you make a monthly calculation, let's say a more rough one, to determine the direction of the movement (without any trajectories)?
Sergey, i.e. do you expect the week to move upwards, but by the end of the week it will be to the closing level of the previous week? So a wider and stronger downward channel will be formed? Or will it be a reversal pattern for an extended move upwards? What statistics do you have about it, and do you make a monthly calculation, let's say a rougher one, to determine the direction of the move (no trajectories)?
I don't use TA and can't say what channel will form. But the forecast is roughly like this, haven't recalculated yet. In fact "went up" a bit faster than I thought it would. But we should not forget that this is an implementation of the process in a statistical sense, the main thing is to assess the overall level of displacement. And it does make sense to recalculate.
I haven't built for a month. There are still some questions, but the next step in system development is forecasting of M15, M30, M60 on one and the same timeframe (some nesting).
I don't use TA and can't say which channel will form. But the prediction is roughly like that, I haven't recalculated it yet. In fact it has "gone up" a bit faster than I thought it would. But we should not forget that this is an implementation of the process in a statistical sense, the main thing is to assess the overall level of displacement. And it does make sense to recalculate.
I haven't built for a month. There are some questions still, but next step in system development is forecasting of M15, M30, M60 on one and the same timeframe (some nesting).
Go for it. >> go for it. I'm certainly following your progress with passion.
Come on. Go for it. I'm certainly following your progress with passion.
That's the thing, there hasn't been any outstanding progress. Right now, the price is hovering at the very edge of the trajectory, but it looks like it's going to go further. We'll see. On principle, I will not recalculate it :o)
Right now, the price is fluctuating at the very border of the trajectories, but it looks like it is going to go further.
Sergei, why do you build a boundary with one RMS and not 3, for example, for 99.97% probability (if the distribution is normal)? Maybe one sigma is not enough for a reliable prediction?
PS. I understand that the question is rather rhetorical, but hopefully the price will stay within one sigma...
That's the thing, there are no outstanding gains yet. Right now, the price is fluctuating at the very border of the trajectories, but it looks like it is going to go further. We'll see. I will not recalculate it because of a principle :o)
It seems that the real price schedule is halved relative to the forecast.
One forecast matches the timing, the other the price or even more likely the structure.