 |
 |
 |
 |
|
 |
 |
13
To view or adjust the program defaults, hit the Program Defaults button on the first screen. There are many
parameters on the Program defaults screen, some with drop down boxes for choices other than the defaults.
We recommend that you use the defaults until you are comfortable with the Techsignal™ program.
The one default that is adjusted most often is the number of data points to analyze (the default value is
2,000). The Interpolation default is set to yes so log interpolations will be used for studying stock trends,
while ROC (rate of change) is a better choice with economic data like the CPI and the M3 money supply;
ROC data usually produces considerably higher Bartel’s test numbers, too (see Bartel’s Test in the Analysis
Grid section below).
There is a Restore Defaults button found through the Program Defaults button on the Main Menu Data Input
screen; this Restore Defaults button restores “all” of the program defaults in case you make a mistake.
Remember the “all” since you will restore every parameter back to Techsignal™ Default settings by using
the Restore Defaults button.
Which Data to analyze
The default data to used for analysis is the Close or period closing value. If the data consist of OHLC, you
also can analyze one at a time, the Open, the High, the Low or a Median Price which is the High + Low
divided by two. Most use the Close or the Median price.
How Much Data to Use
Statistical testing of periodic signals is very sensitive to the number of data points being examined. Too few
and the data can not be tested. Too many and the test begins to get confused and poor results can result,
particularly as you exceed 10,000 data points in your analysis. Sometimes there is simply too much motion
over time or too much trend, and you would need to first detrend the data. Instead, Techsignal™
recommends compressing the data into fewer data points by changing typically from daily to weekly data
points and therefore better statistical analysis. For a point of reference there are approximately 252 trading
days in a year in the U.S. stock market; so, 10,000 data points is roughly 40 years of data, and 20,000 data
points takes you back 80 years to just before the stock market high in 1929.
In general, results begin to get compromised as your data series exceeds 3000 data points. Too few data
points also produces poor results particularly with fewer than 1200 data points. Experience has shown that
1800 to 2000 data points is about optimal in most analyses (roughly 8 years of data). Techsignal™ will
allow you to analyze up to 20,000 data points (80 years). Cycles will likely be found in almost any data
series, but it is a question of how reliable the cycles are. Techsignal™ provides statistical test results to
help pinpoint which are the stronger, more persistent and reliable cycles.
If you find that 2000 data points is not working well, you can try loading more data to discover if something
bigger is going on such as a larger cycle you haven’t picked up with only 2000 data points (about 4 years).
You can load more data points (if you have that much data) from the Main Menu Data Input screen by
manually changing the number of data points on the left hand side from the 2000 default or whatever is
showing (double click the box when done), or by entering the Program Defaults from the same screen on
the right, and changing the default.
This process of using a larger data series to define your cycles we call definitizing a cycle, that is, finding
the presence of a strong, persistent cycle within a longer data series, and one which still passes the rigid
Techsignal™ statistical testing. Cycles in stocks which have proven over time to be more persistent and
reliable are those of approximately 14, 26, 55, 70, 100, 131 and 200 trading periods, with variations
depending on the data series you ask Techsignal™ to examine. For example, after you run a cyclic
analysis and find periods of say 51.3 and 99.6 trading days based on 2000 data points, you might rerun the
same analysis on 4000 data points and find cycles of 50.1 and 102.5 trading days. These are likely the
same cycles.
When you find longer cycles in larger data series, i.e., in the range of 150 to 250 data points, [THIS NEXT
PART IS VERY IMPORTANT] you usually then need to adjust the phase of the cycle in the current time
frame, by moving the plotted cycle forwards or backwards a bit (using arrow keys on the left side of the
Graph screen to actually move the cycle plot). The reason we phase cycles is that older data is sometimes
too far back in time to be in synch with more recent data. Why? Cycles can “wobble” or expand and
contract a bit over several years, enough so to throw off the match up of recent cycle highs and lows with
recent price highs and lows in the current time frame, even though the cyclic periods are statistically valid
 |