Elliottware: machine learning extension of Elliott Wave principle
Instruments for Elliottware Trading
In theory we can use anything. However I am thinking to use Rapid Miner in those analysis. However this may be very heavy in practical way, because of the data isolation and preparation with the Rapid Miner models. Well it is really very heavy on that by now. I want to use this on something easy and not really sophisticated.
I think it would be a good idea to make some testing with the MT4 native backpropagation nerual net.
You can download the BPNN family of indicators.
You van use them as a panel of experts assisting you with your prediction. Remember this is not a trading system this is a prediction toolset.
So the toolset includes by now:
Elliottware prediction Toolset:
1. The standard BPNN Neural net
2. The smoothed BPNN Neural net (it uses EMA smoothing, check the code)
3 The JJMA smoothed BPNN Neural net (I use the Kositsin library, JJMA)
4. The SSA end-pointed BPNN Neural net
5. The Caterpilar SSA BPNN Neural net
Short term prediction toolset:
1. The BPNN Cycle Period. This instrument was implemented for the first time in my thread about Digital ASCtrend in the TSD forum. This was to be used for timing and short term prediction for timing in the direction for the bigger trend.
2. As an analogy I use the BPNN Smoothed Cycle period. It uses the EMA smoothing. In theory it may eliminate some noise.
3. And doing so I wanted to use the JMA smoothing instead of EMA smoothing for Cycle adapted BPNN Neural net. This makes a hybrid indicator combining:
-Cycle adaptation (indicating the relevant time period with 1.5 Dominant cycle adaptation)
-JJMA smoothing to smooth the Data
-Neural net: perceptron with backpropagation to make the prediction. Check the book mark in order to know how to modify the neurla net parameters.
Comments
The BPNN happens to be intresting because it uses the Improved Resilient back-Propagation Plus (iRProp+). The author of BPNN, gwpr writes also that "The main disadvantage of gradient-based optimization methods is that they often find a local minimum. For chaotic series such as a price series, the training error surface has a very complex shape with lots of local minima. For such series, a genetic algorithm is a preferred training method."
However even with the Support Vector Machines, that are specially designed to face the problem with the local minima, I do not get any better results yet. The task is really very complicated.
The use of Elliott Wave for training rande identificatio (I hope) would assure me that I am not looking the solution in the wrong forest.
Improved Resilient back-Propagation Plus (iRProp+). The method is described on this address
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.17.1332
I think I am going to add another hybrid mod between the MESA adaptive indiactors SATL and FATL with Neural Net projection.
That may be cool.
Here I would like to add another suggestion of the use of BPNN advanced mods. We can try not only to prediction direction of the market action. We can also put on the same chart two others one would be with the high prices and another with the low prices. In that way we could obtain a probability channel.
Here those shots are only for example. And this is not a trading suggestion. I just would like to show a possible combination of three BPNN predictors. However I did use the Elliott wave principle in order to caclulate the training set of the neural net.
It is funny to not that those predictions actually worked ;). This instrument can be used but it is just an instrument usless without integrated trading approach