Upgrade August 18-19, 2005
Changes in Spectrum module Many changes have been done in this upgrade. The most important of them is the logarithmic scale for the Spectrum module. This approach allows to reveal short term cycles better, that is extremely important for short term forecasts. The Spectrum diagram now looks like this:
Thus, we can see the 10 days cycle and 2 years cycle as well on the same diagram.
Changes in Timing Solutions module
The very interesting feature is added to this module now. It allows you to vary the indicator to be forecasted. Here is the new look of Timing Solutions dialog box:
As you can see, you can define yourself the Target, or the indicator you try to forecast. The program will use this indicator as the output of the Neural Net. Thus, using standard automated solutions, you can predict any indicators such as:
and many others.
Let me show you just one example. As a target, I use the detrended zigzag. Suppose we want to create the model that catches the turning points:
The critical changes parameter for this zigzag is set to 5%. You may ask: why 5%?
Just because the 5% zigzag provides the wings with a period of 28 days (or one transaction in 14 days):
Now run any of automated Solutions.
For example, run the Solution N1. The program will reveal the most important astronomical cycles and use these cycles as inputs for the Neural Net (FAM model). This Neural Net will predict the detrended zigzag index - because it is its output.
The following is the result I got for some price history file (it is Dell company):
Usually this model catches the nearest turning point because the forecasting horizon for astronomy based models is about one month.
To be sure that I am not dreaming, I took some other data file from my sources (it happened to be Disney's). This is the result provided by Solution N1:
By the way, for this data file (Disney), the Solution N3 works very well.
If we choose the Solution N3, the program will perform these operations:
- creates the detrended zigzag;
- calculates the spectrum for this zigzag, and
- extracts the most important cycles from this spectrum;
- uses these cycles as inputs for Neural Net and
- uses detrended zigzag as an output.
Now the program performs all these operations automatically, by one mouse click. I recommend to play with different modifications of the detrended zigzag (see the next item).
Advanced Detrended Zigzag Some changes have been done in detrended zigzag module. We developed a special kind of zigzag that identifies the turning points more precisely, I would suggest to play with this indicator.
Let me show how it works.
The previous detrended zigzag looked as straight lines connected turning points (see the red diagram on the bottom):
Now there are two parameters added that allow to modify this zigzag. I think the most interesting of them is curviture.
Try to set this parameter as 2 and click on "Ok" button. Then set it to 5 and create one more zigzag.
All these zigzags are shown together:
As you see, the higher curviture is, the more we are focused on turning points pointed by this indicator. Now our zigzag looks this way:
Play with this indicator as a target (output) for Neural Net module. Also it is interesting to use this index for Spectrum module. I applied this experimental technique for FAM models for some stocks and for Spectrum models. Results are really interesting.
Fast buttons in Neural Net module. In Neural Net module, there is a new button now that allows to perform the standard operations faster:
All the best,
Sergey Tarassov
August 18,19 2005
Toronto, Canada