Projection Line

 

What the projection line is

The Timing Solution software is designed to generate the projection line for any financial instrument. It means that we are modeling the future behavior of any financial data. We analyze the past information, search for repeated patterns or other elements, make a model and then test it. If the results are satisfactory, we create a projection line based on this model. 

It is a different approach than the one used in Technical Analysis (which is concentrated mostly on technical indicators). A typical task for Technical Analysis is to get enter/exit signals using indicators (as an example, the intersection of two moving averages). As somebody noted in one of the articles, the Technical Analysis professional is not interested in the future price movements at all, he needs to know only what must be done now. In other words, the forecasting horizon for TA specialist is just 1 trade.

Timing Solution is focused on forecasting. This software is very much interested in knowing the future to some extent. The forecasting horizon is much longer here, and no existing TA methods can provide the same thing. We test our forecasting models on all available financial data and got really statistical improvement for them.

 

 

How to use the projection line

Before discussing anything, we need to clarify some important issues regarding the projection line.

  1. The projection line is not the Holy Grail for the trader.  Each forecast has some degree of risk. Though it is based on the past performance data and statistical analysis, we simply do not know why the price moves this or that way. The projection line can provide some clues and guidance. However, it is totally the trader's responsibility to handle the risk concerned with the projection line.

  2. Usually we provide several models for each financial instrument. These models describe the financial instrument from different points of view. Each one reflects some "behavior patterns" that this stock market used to follow. To be successful, we should simply understand in what pattern mode the stock market is on.

  3. We have to accept the fact that sometimes the market cannot be described by any available models.  

 

Now let us look closer at the projection line. We believe that it may be a good idea to look at the real samples from the real life.

During the year 2005, we have published regularly our forecast for some financial instruments. You can see it yourself, on the website http://www.timingsolution.com. Let us analyze the forecast published at the end of December, 2005. The forecast horizon for all models is one month.  After this time, the usage of this model is too risky.

 

Dow Jones Industrial forecast

The upper diagram is the forecast published December 22, 2005; the other picture shows the real price movement.

We have created two alternative models for DJI. The blue line represents a dynamic model based on astronomical cycles, and the red line is for Spectrum model based on fixed cycles. The blue line is thicker; it means that within last 2 months this model has described the DJI movements better than Spectrum model.

Look at the upper chart. First ten days the blue line there described the real price pretty well (see the bottom chart). The discrepancy has started around January 6 - 7, 2006 (which is marked as "be careful!"). This discrepancy is a signal that you better be prepared for the change of rules of the game. From that moment, there are two possible ways: 1) the stock market behavior is unpredictable in regards to available models (or stochastic if not predictable at all);  2) the stock market since that moment will follow the inverse scenario - top and bottom turning points will trade their places. If the second possibility is the case, we can at least locate the turning points. This scenario has been realized in our case - see the turning point on January 10, 2006. After that date, we do not recommend to use this model - it is too risky.

 

Euro/USD rate forecast

Euro/USD is a very complicated financial instrument to forecast. We have found the very good Spectrum model back tested by Ben Price. The problem is that we still do not have the  alternative scenario for Euro/USD movements. All three available projection lines are based on Spectrum model, and any moment we can face the "inverse" or "melting" effects (see below the description of these phenomena).

So, the upper diagram is the forecast published December 22, 2005, and the next picture shows the actual price movement:

We have three projection lines here,  two of them are very close to each other, the most interesting is the green line. Between December 21, 2005 and January 3, 2006 the price was flat, the stock market looked like waiting something. The predicted turning point A (December 28) was pointing at a big price range only. In this case, what we have to do is just watching the turning points. The turning point B (January 3, 2006) was the beginning of a huge upward movement. The projection line caught the turning point C as well, though it promised the upward swing B-C be very miserable. However, in reality this upward movement was very strong. May be it is the reaction after long waiting, may be some fundamental factors are playing bigger role here than the price movement due to natural reasons.

 

S&P 500 forecast

The best model here is a dynamic model (see the green line). I think the coincidence with real price movement is pretty good:

 

 

Problems related to the projection line

Usually we provide several projection lines that correspond to different scenarios of price movement. If it is possible, it is a good practice to provide independent models, like one based on fixed cycles, while another one is based on astronomical cycles. Thus we analyze the stock market from different points of view. But the life is so diverse and changeable - there are no models that are able to describe the market behavior forever. Changes are unavodable, and the models have to be corrected and sometimes replaced.

Thus, while dealing with the projection line to forecast future price movements, you should keep in your mind these possible outcomes:

Sometimes the turning points are shifted. Looks like the cycles are "melting", their period is changing within the time.
I believe that this problem is related to the time metric problem: we do not really know in what time this stock market exists (and fixed cycles as well). Any university graduate will suggest to use Universal Time as the most commonly used time since the great Newton revolution in physics. The real Back Testing shows that the time measured by trade bars provides better results. And I am not sure that it is the best variant as well. The problem of the nature of time still persists.

As some experimental approach, I can suggest using the volatility time. In other words, we assume that the time is not even. When the volatility is high, the time flows fast; when it is low, the time flows very slow; when there are no trades, the time stops. Maybe, this understanding of time is more human than the physical time used by scientists (at least, this idea is very close to the idea of "real time" measured by events and their duration as suggested by Henry Bergson). Right now you can calculate the Spectrum for this kind of time.

 

What this projection line predicts

Obviously the projection line means the future forecast; the  question is - forecast of what? Definitely, we would like to make a projection line for the price itself or for any initial data. Unfortunately, we do not do that - for pure math reasons. It is due to the existence of trend in the most of financial data. Look at this example:

This is the price for IBM shares from 1960 to 2000. Though we measure this price in dollars, the dollars do not have the same value at different time. 

Trend is presented not only in the share price. Other types of financial data are a subject to trend as well. Look at the diagram for the currency pair, EURO/USD:

As you see, the Euro/USD shows a wide range of changes within last five years, from 0.8 to 1.44. And the most relevant to future forecast data level relates to the year 2004 and up. Again, though we use the same measure, its inner value is different for different times. Because we apply Neural Net technology to generate the projection line, we need to normalize these data, make them as flat as possible. In this case, a good replacement to price or price index would be the detrended oscillator. It allows to see the waves of price movements, and it is flat. Here it is:

The following diagram shows this oscillator (the red curve) together with the price chart (the black curve):

Compare these lines. The most characteristic points (turning points) of one coincide with the same points of the other; when one goes up or down, the other goes in the same direction. It looks like we have substituted the original diagram with an averaged one. The values for earlier data are over weighted while the original values of the latest data are under weighted. But now we can compare them and apply different math methods to analyze these data. 

If we need to reveal more short term waves, we can use the shorter smoothing period. See below the oscillator with smoothing period=10:

Thus, if we would like to make the forecast for future price movement, we need to decide what waves we are most interested in and make the relative price oscillator with the appropriate periodicity. So, when the program calculates the projection line for the price, it actually does it for some kind of the detrended oscillator.

Price movement is not the only thing that this program is able to create a projection line for. It also makes the projection lines for different indicators like Volatility index (Vx), Relative Strength Index (RSI), ADX and many others. All these indices are "targets" for Neural Net; it will generate the forecast for the chosen index/indicator analyzing its historical data.

To make the forecast for turning points, we have developed the special indicator. We call it "detrended zigzag". It is shown as a red line on this digram:

Any projection line for any indicator defined as a target for Neural Net module can be prolonged in the future as long as you need. Look at this example. This is Spectrum model for detrended zigzag index (the last available price bar here is Jan. 26, 2006);

The biggest issue here is to find the reliable forecasting model. This is the task for Back Testing module of the program.