Stocks & Commodities V.

9:10 (403-408): Trading The Regression Channel by Gilbert Raff

Trading The Regression Channel

by Gilbert Raff

very trader has had the experience of selling a stock or commodity too soon during a rapid price

reversal, only to realize in retrospect that this was a consolidation within a trend. Or just the opposite:
perhaps the trader watched profits evaporate when the "consolidation" became a downhill avalanche.
A vast array of methods exists to describe price movement, ranging from trendlines to Fourier analysis.
The ideal description of a trend would provide clear recognition of its start, trading range and
termination. The ideal tool would permit the trader to know exactly when to enter a trade, when to exit,
when to trade contrary to the trend and when a new trend had started. The ideal tool would also give a
good sense of whether price action was likely to be worth the risk of any trade at all.
While we don't expect perfection, most traditional description of trends fail miserably to measure up to
our ideals. The regression channel is a technique I've developed that provides many of these stated needs,
including the ability to predict reversals within the trend and, frequently, to project the end of a trend
weeks in advance.
To start, let us review two classic descriptions of trend. The first describes an uptrend as a sequence of
higher highs and higher lows. In Figure 1, a daily price chart of Blair Corp. (BL), an uptrend begins at
point A and ends at point B. From point B to point C, there is first non-trending action, then a lower low
at C. Presumably, trend AB is over. Figure 2 shows the same security later. Much to our chagrin, exiting
at C was a mistake.
The second classic description of trend defines it with a trendline. Figure 3 represents the uptrend defined
by the line joining lows C, D and E. At F, a new low breaks the trendline decisively. Where do we go
from here? Future price action is shown in Figure 4.
DEFINING THE REGRESSION CHANNEL

Article Text

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

FIGURE 1:

Figures

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

FIGURE 2:

FIGURE 3:

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

FIGURE 4:

FIGURE 5:

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

These results were unsatisfactory, so I undertook the development of a better method. First of all, to
describe price direction and strength I chose linear regression, the least-squares best fit line between two
prices.
In Figure 5, the linear regression of price between two intermediate-term swing points is drawn as line Z.
From Z, a quantitative description of strength and direction of trend AB is available as the slope of Z, the
rate of change of price over time. The larger the absolute value of slope Z, the faster price is going up or
down. A low or zero value for Z suggests that the market is not worth the risk of a trade.
Between A and B, the market drifts away from the Z line both above and below, but at least during time
AB, the market always returns to the trendline. To describe this quantitatively, we define the channel
range: the extent of movement away from the trendline the market will tolerate before trend AB
disintegrates.
The channel range in this example can be described by finding the point furthest below or above Z in AB.
In Figure 6, we've chosen the high at the point that we refer to as RCH (short for regression channel
high). Returning to point A, draw a line parallel to Z that intersects RCH. This line is the regression
channel top (RCT). This line has a constant distance for Z called the channel range (CR). Now return to
A and draw a line parallel to Z at CR below it. Note how the regression channel bottom line then predicts
the later regression channel low at RCL. If we have done our work accurately, we will find that one
extreme will predict subsequent highs and lows and incorporate the market action in between.
This is the essential strength of the regression channel. At some point in the development of a price
move, the market's direction can be described by Z and the price variability can be described by CR,
sometimes with an amazing degree of accuracy for many months. Once the channel is constructed, we
then project Z, RCT and RCB lines into the future.
If the regression channel is to be of practical use, it must exhibit predictive behavior. Do the lines drawn
at RCH really predict price turns? In Figure 7, we see that four months after RCH, Blair turns right on the
RCT line (at point C). What else can we observe here? Note that through all six months, the line Z
repeatedly serves as an important support and resistance line. It does so for the last time just after C,
when support fails and the trend referred to AB ends decisively. During these six months, several
non-trending periods by simplistic definitions appeared, when the market was really running in a
well-defined channel.
But what about price move CD? Is there any way regression channel analysis can help the trader avoid
going over the cliff? Superficially, the fall is so abrupt, it seems hard to avoid. Yes, it can be predicted,
but to explain how, we must examine both the short- and the long-term trends.

This is the essential strength of me regression channel. At some

point in the development of a price move, the market's direction
can be described by Z and the price variability can be described
by CR, sometimes with an amazing degree of accuracy for many
months.
A complex of short-, intermediate- and long-term trends lies beneath the surface of the daily market

Article Text

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

FIGURE 6:

FIGURE 7:

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

FIGURE 8:

FIGURE 9:

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

action we've plotted. The trader must explore them all to understand where he is on the "map" of the
market. The idea is to analyze different time scales to get clues about important or confusing events. For
my purposes, since I prefer to trade the intermediate trend, I omit the intraday level and use short-term
swing points to indicate those on the level of a daily chart. In Figure 8, within the now-familiar AB
channel, short-term channels have been constructed, showing obvious utility to the short-term trade. For
example, the point RCL becomes an attractive entry to a trade managed by a trailing stop-loss just below
short-term regression channel line RCL-B1. Note also the tendency for the short-term channel to
converge toward and break at the AB intermediate-term regression line Z. This is known as channel
intersection. We will always look to reactions at intersections between channels of different time frames.
Is there any way we could have anticipated the waterfall slide from point C to point D? To the short-term
trader, with a daily chart perspective, no, but the short-term channel bottom with a trailing stop still takes
the trader safely out with a profit at C1, comfortably before D.
But move CD can be predicted by the intermediate-term trader who wants to ride the AB channel to its
end without constantly trading in and out.
The answer lies in looking at a long-term chart. In Figure 9, we see the chart of Blair Corp. for the 15
months preceding point C. It is now clear that point C is the first break above the RCT line for 1990.
When a market breaks above its long-term regression channel it is often dangerously overbought and
liable to result in violent corrections. Channel AB appeared to have been on a collision course with the
long-term RCT for the last six months. Channel intersections can be anticipated to be points of volatility
and price reversal. By taking these factors into account, an astute trader would take his profit at point C1
and be satisfied.
Market indices and commodities show similar behavior. In Figure 10, the weekly cash Standard & Poor's
500 index is shown between 1981 and 1991. The 1987 crash occurred after the only move above the RCT
in 10 years, and the results are analogous to line CD in Blair Corp. Note also how the October 1987
bottom predicted the October 1990 bottom.
AVOIDING PITFALLS
How can we know when we've drawn a valid channel? A valid channel will demonstrate predictive
behavior. It should define and predict swing highs and lows within its market time frame. At channel
intersections, resistance or volatility should be the key. Before I trade a regression channel, I look for
evidence to validate my construction.
For example, let's begin a new regression channel analysis of AT&T on April 9, 1990 ( Figure 11). Before
drawing daily regression lines, step back and examine the long-term chart of weekly data using a
regression channel spanning all the data (Figure 12). At the moment, prices have just broken above their
long-term channel and may be in for a sharp correction. This is only a hypothesis at this point, but
certainly that was not obvious from Figure 11.

Article Text

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

FIGURE 10:

FIGURE 11:

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

FIGURE 12:

FIGURE 13:

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

The prediction of changes in price trends is me final and most

intriguing application of regression channel analysis. This may
seem far-fetched until we realize mat we are applying the principle
of channel intersection to channels of different time spans.
Now return to the daily chart and load the largest screen of historical data possible. Then go back and
construct the regression channel for the preceding market trend (Figure 13). Why use point C to measure
the channel range, rather than D? Point D was a price spike, a very large, abrupt price move (the
"minicrash" of October 13, 1989), in which prices may exceed their true range briefly. This is particularly
important in volatile commodities.

et's see what we get by accepting the channel range | described by point C. At E, prices pushed above

RCT to an overbought extreme, then crashed (D). This is a reasonable channel construction, since it
explains market action and that is the criterion for predictive behavior, the most important key to
recognizing a valid channel. The more confirmations we get from the market's turns at channel lines, the
more confident we are that we can rely on the channel and trade it.
At point F, prices break the preceding channel and bring us up to date. Is there sufficient information to
construct a correct new channel now? We can't tell yet. My approach is to use the previous method and
see if the results fit the market's action. If so, we'll stick with them, but if not, we'll revise, and at some
point we'll have a definitive channel with predictive behavior If this occurs early in the trend, we have a
powerful tool to trade. If it occurs late, just before a trend change, we'll have to wait for another
opportunity.
In Figure 13, begin at point B and construct Z1 by connecting B to the latest trading day with a regression
line. Since the trend is down, I measure the channel range first by picking B as the highest high. This
doesn't work out well, since it excludes H, and there appears to be no valid reason why. I then refit the
data, using H as the extreme, and the result is Figure 14. Based on the evidence of the long-term chart
and the daily chart, would I be willing to use this chart and be ready to short AT&T? Not yet First, I
would require proof from subsequent market action that the channel has ample accurate predictive
behavior. If the channel is right, this is not long in coming. Let's extend our channel lines to the right and
see what happens.
In Figure 15, we see prices attempting to break out of our new channel. Point Z1 acted as a support line at
I, at which point the RCT acted as a resistance line at J and then prices broke to K. At this point, several
confirmations of our channel have occurred, and I would be willing to short AT&T. Figure 16 shows the
result. The channel we constructed on April 9, 1990, defined prices for another 12 months and allowed us
to immediately recognize a continuing uptrend. By using Figure 17, we could predict the end of the
downtrend weeks in advance by observing the coming channel intersection at L.
PREDICTING CHANGES IN TREND
The prediction of changes in price trends is the final and most intriguing application of regression
channel analysis. This may seem far-fetched until we realize that we are applying the principle of channel
intersection to channels of different time spans. To answer the question of where we go from here, it is

Article Text

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

FIGURE 14:

FIGURE 15:

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

FIGURE 16:

FIGURE 17:

Stocks & Commodities V. 9:10 (403-408): Trading The Regression Channel by Gilbert Raff

necessary to ask where we have been. Add to this the quantitation possible with the regression channel,
and prediction seems to become quite possible.
The regression channel, which we can legitimately term a new method of trend analysis, allows notable
precision in the quantitation of trend direction, strength, range, initiation and termination. Its use and the
concepts of prediction behavior and channel intersection allow us to validate the channels, trade them and
even predict trend change months in advance.
Gilbert Raff is president of Torsede Investments, Inc. He has traded equities and commodities since
1981. He trained in math and computer science at MIT.

Figures

Copyright (c) Technical Analysis Inc.