Data Filtering For Trend Channel Analysis
by Anthony W. Warren, Ph.D.
Trend-following methods typically utilize moving averages of closing price data for buy and sell signals.
Often, the signals turn out to be false due to short-term market fluctuations. Here, longtime STOCKS &
COMMODITIES contributor Anthony W. Warren, correcting one of the major drawbacks of moving
averages, introduces a trend-following method that smoothes the data for trend identification and
measures short-term price fluctuations to establish statistical boundaries.
The following article will teach you how to use a zero lag filter of the closing price data to create trading
bands. The trading bands are used with a long-term price filter for trading the intermediate trend of the stock market.
I recently developed a simplified version of the alpha-beta trend-following method that can be
implemented using custom formulas either in a technical analysis program such as MetaStock version 3
(see sidebar, "MetaStock") or using a spreadsheet program (see sidebar, "Trend channel spreadsheet").
Although the formulas are easily implemented, they make use of some data filtering techniques that are
not widely known in technical analysis. Consequently, here are both a tutorial on the design and synthesis
of data filters for trading systems and an application example that uses several data filtering principles to
implement trend channels for buy and sell trade timing. The formulas used are pulled from the
derivations found in the "Mathematica" sidebar.
But before I go any further, a few basic terms should be explained. First, input data is data such as a
series of closing prices of the stock market. A data filter is a mathematical means to modify the data; for
example, a moving average is a data filter. A moving average smoothes the data, reducing the extreme
swings of the daily closing prices. If you plot the closing prices along with a moving average, you will
find that the moving average is a smoothed version of the closing prices. A five-day moving average is
the total of the last five days' closing prices divided by 5: ...
DATA FILTER PROPERTIES
The two most important properties of data filters for technical analysis are noise smoothing and data lag
between the input data and the smoothed output values. Traditionally, moving average (MA) data filtering
has been the mainstay of most trading methods. Typically, if the main purpose is noise smoothing, then
the moving average length N is chosen as the smallest value that provides adequate noise smoothing, and
filter lag is an undesirable byproduct of the process. On the other hand, with trend following methods
such as, say, 39-week moving average crossings, the MA length N is chosen to obtain sufficient filter lag
to reduce false alarm buy and sell signals. In this case, the extra noise smoothing obtained with long MA
lengths is simply a result of the filtering process. The main principle here is that increased noise
smoothing requires longer filter lengths and thus more lag in the filtering process.