Edge effect on the way to the GRAAL - page 5
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
Am I correct in assuming from the figure that the network guesses the direction 30% of the time?
Have you tried working with a collection of nets? For example with 3 or 5 to refine the decision.
Or with a pair of nets: one guesses only upwards, the other only downwards.
By the way, why exactly 3 (or 5, I'm confused ;) ) input neurons. I just met networks with 4, 7 or 15 inputs :)
p.s.
I once did an experiment. I memorized all the history I had and searched for the most similar situations to the current one
using vector distance method (normalized vectors, of course). In 60% of cases, history repeated itself :)
But it still depends on prediction range and vector length.
No not correct, Grid is guessing 10-15% of what is shown in blue. The red shows the training sample. I don't use the Grid Committee - I haven't felt the need for it yet. If predictive ability of isolated NS proves to be insufficient, then I'll work with committee.
By the way, about pre-training. I can rigorously show that retraining the NS in n steps, is equivalent to increasing the prediction horizon by a factor of n. The consequence of this is a power dependence of increasing the predictive ability of the NS. For example, if NS right after training correctly predicts 10% of signs of price movement direction, then in one step after training the prediction ability decreases to 1%, in 2 steps - 0.1% and so on, and this is a medical fact! Obviously for the price-type time series retraining at each step is extremely relevant.
No it isn't, Grid is guessing 10-15% of what is shown in blue. The training sample is shown in red. I don't use the grid committee - I don't feel the need for it yet. If predictive ability of isolated NS proves to be insufficient, then I'll work with committee.
By the way, as for retraining. I can rigorously show that retraining the NS in n steps, is equivalent to increasing the prediction horizon by a factor of n. The consequence of this is a power dependence of increasing the predictive ability of the NS. For example, if NS right after training correctly predicts 10% of signs of price movement direction, then in one step after training the prediction ability decreases to 1%, in 2 steps - 0.1% and so on, and this is a medical fact! Obviously for time series like price series retraining at each step is extremely relevant.
Have you actually succeeded in predicting anything or are you just pouring water in front of gawkers? If yes, how much, what and what do you use in the output of your neuristics? Have you studied the predictability of the series? Besides, you still haven't answered my question: how do you trade without knowing the movement of currency in one direction or another during a certain period of time?
Desperado, wavelets belong to the weak approximators and it's not good to use them for prognosis as well as SSA and statistics with its spectrum analysis, regression and other tricks.
Have you studied the predictability of the series? Besides, you still have not answered my question, how do you trade without knowing the movement of currency in this or that direction during some period of time?
I dont have words about your whole comment, but about predictability of a series, it is an interesting point, do you have algorithms (ideas, thoughts) for the stated estimate?
Have you actually succeeded in predicting anything or are you just pouring water on the gawkers? If yes, how much, what and what is the output of your neuroscience you use? Have you studied the predictability of the series? Besides, you still haven't answered my question: how do you trade without knowing the movement of currency in one direction or another during a certain period of time?
Desperado, wavelets belong to weak approximators and it is not good to use them for forecasting, like SSA for example, or statistics with its spectrum analysis, regression and other tricks.
If it's not a secret, what's good to use then?
About your whole comment I have no words yet, but about the predictability of the series, it's an interesting point, do you have any algorithms (ideas, ideas) for the stated estimation?
The Hurst Index, for example, gives quite a good estimation of the market condition and its predictability.
If it is no secret, then what is good to use?
It is far from being a secret. Non-linear dynamic models, including neural networks, MGUA, and radial basis functions. Many things have been created now.
TheHearst indicator, for example, provides a good estimate of market conditions and moments of predictability.
The Hurst index does not reveal internal non-linear regularities existing in BP, it is an integral index and has a strong affinity to the correlation coefficient between adjacent samples in the series of the first difference of the initial BP (this can be proved strictly). Thus, all that can be constructed using this characteristic are first-order autoregressive models or their derivatives. The problem that is solved using the NS apparatus, as you correctly pointed out above, is broader and is not limited to linear autoregressive models and certainly not to a way of estimating hidden patterns. I recently dealt with "box-counting" a method of quantifying the predictability of real financial instruments, I think we should be talking about something similar.
Who told you that there are hidden patterns in Forex?) Everything is open and accessible there. Another thing is whether you have enough information to study a series in depth. But this is more about fundamental analysis. From the point of view of technical analysis, everything is available to you.
Another thing is whether you have enough information to study the row in depth.
There is such a thing. Moreover, even if there is enough information, there is no guarantee that it will not already be obsolete and useless... In short, it's all about compromise!
And who prevents you from using fundamental analysis with your forecasting system? For example, the news in SaxoTrader is delivered in real time.