Forecast in three mouse clicks

written by Sergey Tarassov


This small article demonstrates one very simple and very effective technology to get forecast based on a special type of spectrum. This multi frame spectrum algorithm is our "know how". It allows to resolve the most difficult question that arises when we deal with forecasting models based on fixed cycles - to define the longevity of cycles' life. The idea of cycles is the key part in the assumption that the stock market may have a cyclic nature. If we accept this assumption, we can compare the stock market to some water surface. Some factors affect the stock market, and it reacts in some way - similar to the waves on the water surface when you throw a stone into the river. Finally, the waves are dissipated - as they loose their energy; and the stock market no longer follows the same cycle. We can say that the market "forgets" the event that has been a cause of the beginning of this cycle and is waiting for another event to react on. The nature of the cause is not discussed here; it could be any of fundamental/inner factors; what is important for us is that any of these factors is able to generate waves, i.e., to start small cycles in the market's life. In other words, we believe that small cycles (not big ones) are responsible for the changes on the market. Therefore, our main target is to find the way to deal with these cycles, or the waves, keeping in mind that they are not infinite and have a rather short life. In my opinion, the best way to get the visual impression regarding the "life" of any fixed cycle is the wavelet; it will be discussed in some other article. Here I would like to show you the easy way of getting the forecast. Let's begin...


Suppose we have some price data for the analysis. Before doing anything, click on the "Calculate" button (to activate the program's features for creating the forecasting models). As an example, I take the first chart that has captured my eye; it has turned around to be the chart of Dell shares. I have downloaded these price data for  the last 3-6 years, starting from 1998 to middle of the year 2004.

Click #1: Click on this button:

You will get the Spectrum diagram that represents the most important playing cycles till 9, 2003 (the price points for May 9, 2003 and after that day are used to verify our model):


Be sure that the "Basic Interval" is set this way (it is a default setting):

If it is not, set it on "Last Cycles" item and click on the "Recalculate" button. (In this case, please change the title of your copy of this article to "Forecast in four mouse clicks".)

If all the above is done, it means that we have activated multi frame spectrum algorithm option (which is the subject of this article).

Click #2: Click on this button:



Click #3: Click on this button:

You will see the diagram that represents the superposition of the cycles calculated from the spectrum. The black line is the normalized price while a red line is the projection line calculated through these cycles; it can be prolonged as far as it seems necessary.

Drag the mouse on any piece of the diagram to see the details. The picture shows a part of the diagram near the border line between the learning and testing intervals:


You see that the red line follows the main tendencies of the black line very well on both intervals, learning and testing as well. To get the forecast, we need simply continue this line further (drag the mouse on the interval that goes beyond the testing interval).

The models created by this method give very good forecasting results very often. Usually it allows to see the future price movements several months ahead.

The following are some examples of forecast for several randomly selected stocks (they are in the alphabetical order according to my data base):


I would say that we can trust this model for several months ahead.



The forecast looks not bad though the peak in the beginning of March 2002 has been in reality much stronger than the Spectrum analysis shows.



This is not the best of the samples




A bit of theory

The main feature of the Spectrum analysis is that it allows to consider short  term cycles for the analysis, especially those that occur close to LBC (Learning Border Cursor). In other words, if we consider the strength of 10-days cycle, the price history 3 years ago may be not so relevant as the latest history. It is obvious because the wave corresponding to this cycle (if any exists) can be caused by some fundamental factor that has happened a month ago. Therefore, this wave is a kind of response of stock market to this disturbing factor, it is the way how the market communicates with the world. The multi frame Spectrum is more sensitive to short term cycles than to the long ones. Traders are not much interested in long cycles; the most exciting opportunities for the trader occur with the short cycles. But it is not easy to reveal the short cycle; there are so many factors to be considered. Thus, we need to be pretty accurate with these cycles.
For those who would like to play with this feature of the program and who prefers self-set options to the default ones, I would recommend to vary this parameter:

You can take it as the length of market's "memory" in respect to different cycles. 

As an example, let us look again at the Dupont shares forecast (see above). I would not consider it as a sample of good fitness to the price. Change this parameter to 7 cycles:

It looks like the projection line becomes better; the possible conclusion is that this stock has the shorter memory (or the shorter reaction interval to the disturbing factors):

I cannot say now whether the factors causing these cycles (fundamental or other factors) have been not so strong as in the other examples or this particular stock market is able to recover quickly. These are possible directions of the future research involving the fundamental analysis. What we can do now with the Timing Solution software is - to calculate the short term cycles and research the time intervals of their influence with the main purpose - to get the reliable forecast and create a base for successful trading.

26 February 2005