Stocks & Commodities V16:1 (33-37): Smoothing Techniques For More Accurate Signals by Tim Tillson

INDICATORS

Smoothing Techniques For

More Accurate Signals
More sophisticated smoothing techniques can be used to We use moving averages (low-pass filters) in technical
determine market trend. Better trend recognition can lead to analysis to remove the random noise from a time series, to
more accurate trading signals. Here’s how. discern the underlying trend or to determine prices at which
we will take action. A perfect moving average:
by Tim Tillson
fter studying his first stock chart, 1 Would be smooth, not sensitive to random noise in the
the novice technician is most underlying time series. That is, its derivative would not
likely to learn next about moving spuriously alternate between positive and negative
averages. This is a reasonable values.

A progression. First, moving aver-

ages are easy to understand. The
simple moving average is just the
average of a given number, which
2 Would not lag behind the time series from which it is
computed. Lag, of course, produces late buy or sell
signals that kill profits.

we will refer to as n, of previous The only way a perfect moving average could be computed
closing prices, recalculated each is to have knowledge of the future, and if we had that, we
day at the close. And second, technicians use moving aver- would buy one lottery ticket a week rather than trade! That
ages because the moving average offers a smoother visual aside, we can still improve on conventional simple, weighted,
image of the market trend. In effect, the moving average or exponential moving averages.
removes the noise around the trend.
This concept of eliminating noise from the trend is similar TWO BENCHMARKS
to what engineers strive for in their application of digital We will examine two benchmark moving averages based on
filters. As R.W. Hamming observed: linear regression analysis. In both cases, a linear regression
line of length n is fitted to price data. The first benchmark
Digital filtering includes the process of smoothing, predicting,
moving average (MA) is called ILRS, which stands for inte-
differentiating, integrating, separation of signals, and removal of
noise from a signal. Thus many people who do such things are
gral of linear regression slope. In this moving average, the
actually using digital filters without realizing that they are; being slope of a linear regression line is simply integrated as it is
unacquainted with the theory, they neither understand what they fitted in a moving window of length n across the data. The
have done nor the possibilities of what they might have done. derivative of ILRS is the linear regression slope. ILRS is not the
same as a simple moving average (SMA) of length n, which is
This quote applies to the vast majority of indicators in actually the midpoint of the linear regression line as it moves
technical analysis. Moving averages, be they simple, weighted, across the data.
or exponential, are low-pass filters; low-frequency compo- We can measure the lag of moving averages with respect
nents in the signal pass through with little attenuation or to a linear trend by computing how they behave when the
reduction, while high frequencies are severely reduced. Os- input is a line with unit slope. Both SMA(n) and ILRS(n) have
cillator-type indicators, such as moving average conver- lag of n/2, but ILRS is much smoother than SMA.
gence/divergence (MACD), momentum and relative strength Our second benchmark is the end point moving average
index, are another type of digital filter referred to as a (EPMA). It is the endpoint of the linear regression line of
differentiator. length n as it is fitted across the data. EPMA hugs the data more
Early in the steps of calculating any of these indicators, the closely than a simple or exponential moving average of the
difference between today’s price and some price a number of same length. The price we pay for this? EPMA is much noisier
days ago or the difference between two moving averages are than ILRS, and it also overshoots the data when linear trends
measured — hence the term. STOCKS & COMMODITIES Con- are present, as can be seen in Figure 1.
tributing Editor Tushar Chande has observed that many However, EPMA has a lag of zero with respect to linear
popular oscillators are highly correlated, which makes sense, input! This makes sense because a linear regression line will
because they are measuring the rate of change of the under- fit linear input perfectly, and the endpoint of the LR line will
lying time series. be on the input line.

Copyright (c) Technical Analysis Inc. 1

Stocks & Commodities V16:1 (33-37): Smoothing Techniques For More Accurate Signals by Tim Tillson

These two moving averages frame the tradeoffs that we EPMA. Can we build something that is comparable, but
ANDREW VANDERKARR

are facing. On one extreme we have ILRS , which is very smoother? Figure 1 shows ILRS, EPMA and IE/2.
smooth but has considerable phase lag. At the other ex-
treme, EPMA has zero phase lag, but it is too noisy and FILTER TECHNIQUES
overshoots. We would like to construct a better moving Any student of filter theory would be able to tell you that
average that is nearly as smooth as ILRS , but runs closer to the smoothness of a filter could be improved by running it
where EPMA lies, without the overshoot. through itself multiple times, at the cost of increasing
A easy way to do this is to split the difference — that is, phase lag.
use (ILRS(n)+EPMA(n))/2. This will give us a moving There is a complementary technique (called twicing by J.W.
average (call it IE/2) that runs in between the two, has Tukey, author of Exploratory Data Analysis) that can be used to
phase lag of n/4 but still inherits considerable noise from improve phase lag. If L stands for the operation of running data

Copyright (c) Technical Analysis Inc. 2

Stocks & Commodities V16:1 (33-37): Smoothing Techniques For More Accurate Signals by Tim Tillson

through a low-pass filter, then twicing can be described by:

EPMA
L’ = L(time series) + L(time series - L(time series))

METASTOCK FOR WINDOWS (EQUIS INTERNATIONAL)

We add a moving average of the difference between the input
and the moving average to the moving average. This is ILRS
IE/2
algebraically equivalent to:
2L-L(L)
This is the double exponential moving average (DEMA),
popularized by STOCKS & COMMODITIES contributor Patrick
Mulloy. DEMA has some phase lag (although it exponentially
approaches zero) and is somewhat noisy, comparable to IE/ FIGURE 1: HEWLETT-PACKARD. The EPMA(15), IE/2(15) and ILRS(15) are red,
2. We will use these two techniques to construct our better blue and green, respectively in MetaStock. Note how the EPMA hugs the data more
moving average after we explore the first one. closely than a simple or exponential moving average of the same length. The price
we pay for this is that it is much noisier than ILRS, and it also overshoots the data
when linear trends are present.

FIXING OVERSHOOT
An n-day EMA has smoothing The problem with this approach is that multiple runs
constant alpha=2/(n+1) and a though these filters increase their tendency to overshoot† the
lag of (n-1)/2. Thus, EMA(3) data, giving an unusable result. This is because the amplitude
has lag 1, and EMA(11) has response of DEMA and EPMA is greater than 1 at certain
lag 5. Figure 2 shows that if I frequencies, giving a gain of much greater than 1 at these
am willing to incur five days of frequencies when run though themselves multiple times.
lag, I get a smoother moving average if I run EMA(3) through Figure 3 shows DEMA(7) and EPMA(7) run through them-
itself five times than if I just take EMA(11) once. This suggests selves three times. EPMA3 has serious overshoot, and DEMA3
that if EPMA and DEMA have zero or low lag, why not run fast similarly has problems.
versions (for instance, DEMA(3)) through themselves a num- The solution to the overshoot problem is to recall what we
ber of times to achieve a smooth result? are doing with twicing:

METASTOCK IMPLEMENTATIONS

MetaStock 6.5 code for ILRS: Running the filter though itself three times is equivalent to
cubing f:
{input number of lookback periods, default is 11}
periods:=Input(“Periods? “,2,63,11); –a 3 x 6+ 3a 2+3a 3 x 5+ –6a 2–3a–3a 3 x 4+ 1+3a+a 3+3a 2 x 3

{determine how many points are in the time series}

Thus, the MetaStock 6.5 code for T3 is:
size:=LastValue(Cum(1));
periods:=Input(“Periods? “,1,63,5);
{determine the constant of integration by taking the simple
a:=Input(“Hot? “,0,2,.7);
moving average of the first periods points in the time
e1:=Mov(P,periods,E);
series}
e2:=Mov(e1,periods,E);
start:=LastValue(Ref(Mov(P,periods,S),periods-size));
e3:=Mov(e2,periods,E);
e4:=Mov(e3,periods,E);
{value is the integral of linear regression slope plus the
e5:=Mov(e4,periods,E);
constant of integration}
e6:=Mov(e5,periods,E);
Cum(LinRegSlope(P,periods))+start;
c1:=-a*a*a;
c2:=3*a*a+3*a*a*a;
If x stands for the action of running a time series through
c3:=-6*a*a-3*a-3*a*a*a;
an EMA, f is our formula for generalized DEMA with the
c4:=1+3*a+a*a*a+3*a*a;
variable “a” standing for our volume factor:
c1*e6+c2*e5+c3*e4+c4*e3;

f : = 1 + a x – ax 2 —T.T.

Copyright (c) Technical Analysis Inc. 3

Stocks & Commodities V16:1 (33-37): Smoothing Techniques For More Accurate Signals by Tim Tillson

DEMA(7)3
3
EPMA(7)
EMA(3)5

EMA(11)

FIGURE 2: HEWLETT-PACKARD. In MetaStock, the EMA(11) is red versus the FIGURE 3: HEWLETT-PACKARD. In MetaStock, the DEMA(7)3 is red and EPMA(7)3
EMA(3)5, which is green. EMA(11) and EMA(3)5 both have five days of lag, but is blue. When DEMA(7) and EPMA(7) are run through themselves three times,
EMA(3)5 is smoother. DEMA3 has serious overshoot, and EPMA3 similarly has problems.

DEMA(n) = EMA(n) + EMA(time series - EMA(n)) This is algebraically the same as:

The second term is adding, in effect, a smooth version of the EMA(n) * 1.7-EMA(EMA(n)) * 0.7
derivative to the EMA to achieve DEMA. The derivative term
determines how hot the moving average’s response to linear I have chosen 0.7 as my volume factor. The general formula
trends will be. We need to simply turn down the volume to (which I refer to as “generalized DEMA”) is:
achieve our basic building block:
GD(n,v) = EMA(n) * (1+v)-EMA(EMA(n)) * v
EMA(n) + EMA(time series - EMA(n)) * 0.7

PERFECT MOVING AVERAGES AND OSCILLATORS

10.00
If we had knowledge of the future, perfect moving averages
and oscillators could be constructed. Here’s how it is done
computationally:
5.00
• A perfect moving average can be constructed by
adding an exponential moving average that moves
backward in time to one that moves forward in time,
0.00
then dividing by two. The phase lead of the backward
EMA cancels the phase lag of the forward EMA, pro-
ducing a moving average that is both smooth and in
phase (but only for historical data, not the data on the -5.00
right-hand side of the chart, where we want to trade).
• A perfect oscillator can be had by subtracting the -600.00 -500.00 -400.00 -300.00 -200.00 -100.00 0.00
forward EMA from the backward EMA, or by taking the
derivative of the perfect moving average. These calcula- SIDEBAR FIGURE 1: PERFECT(20). Tillson plotted the two-sided derivative as
tions are analogous to the two-sided derivative in calcu- a dotted line, while the 20-day linear regression line slope is the black line.
lus, which uses the future f(x+h) term in the definition:

Lim The limitations of technical analysis become very apparent

h->0 f(x+h)-f(x-h) when one compares left-sided oscillators with perfect oscil-
2h lators on historical data! Using about 600 days’ worth of
HWP daily closes, I plotted the two-sided derivative PER-
In the real world, oscillators are constrained to be like the FECT(20) as a dotted line, while the 20-day linear regres-
noisier left-sided derivative: sion line slope is the black line (sidebar Figure 1). PER-
Lim FECT(20) has about 10 days of phase lead over LRS, about
h->0 f(x)-f(x-h) what one would expect. —T.T.
h

Copyright (c) Technical Analysis Inc. 4

Stocks & Commodities V16:1 (33-37): Smoothing Techniques For More Accurate Signals by Tim Tillson

The system used MA Optimal parameter Best APR

ILRS(15) was very simple. A ILRS 7 19.02
moving average was SMA 8 20.20
computed using each EMA 8 28.17
EPMA(15)
of the five above. A DEMA 7 28.81
derivative was taken T3 5 (volume = 0.7) 33.59
T3(6) (one-period rate of FIGURE 5: PARAMETERS AND PERFORMANCE.
change function). A Here’s a summary of the best results that each MA was
able to achieve. Note that the optimal parameters did
long position was en- not vary widely.
tered at bottoms and
EI/2(15) closed at tops, of the
derivative. No shorts were taken. For example, the code for
FIGURE 4: HEWLETT-PACKARD. The EPMA(15) is blue, the T3(6) is red, the “enter long” for an EMA was:
IE/2(15) is yellow and the ILRS(15) is green in MetaStock.
res:=Mov(C,opt1,E);
where v ranges between zero and 1. When v = 0, GD is just an d1:=ROC(res,1,points);
EMA, and when v = 1, GD is DEMA. In between, GD is a less d1 > Ref(d1,-1) AND Ref(d1,-1) < Ref(d1,-2)
aggressive version of DEMA. By using a value for v less than
1, we cure the multiple DEMA overshoot problem but at the I let opt1 vary from two to 10. This is a fast system, which
cost of accepting some additional phase delay. Now we can executes a lot of trades. Figure 5 summarizes the best results
run GD through itself multiple times to define a new, smoother that each MA was able to achieve; note that the optimal
moving average (T3) that does not overshoot the data: parameters did not vary widely.
No single set of backtesting results is conclusive. But these
T3(n) = GD(GD(GD(n))) numbers confirm that T3 has merit — it not only looks good
to the eye on the chart, but it can also be a powerful building
Figure 4 shows T3(6) plotted with EPMA(15), ILRS(15) and
block in other indicators and trading systems.
IE/2(15). T3 is very similar to IE/2 and DEMA, but smoother
than both, which was our goal.
Tim Tillson is a software project manager at Hewlett-Packard,
In filter theory terminology, T3 is a six-pole nonlinear
with degrees in mathematics and computer science. He has
Kalman† filter. Kalman filters are ones that use the error —
privately traded options and equities for 15 years.
in this case, (time series - EMA(n)) — to correct themselves.
In the realm of technical analysis, these are called adaptive
moving averages; they track the time series more aggres-
RELATED READING
Hamming, R.W. [1989]. Digital Filters, 3d edition, Prentice-Hall.
sively when it is making large moves.
Mulloy, Patrick G. [1994]. “Smoothing data with less lag,”
Technical Analysis of STOCKS & COMMODITIES, Volume 12:
TRADING RESULTS
February.
I used MetaStock 6.5 to compare five moving averages (SMA,
_____ [1994]. “Smoothing data with faster moving averages,”
ILRS, EMA, DEMA and T3) on the NASDAQ index (NDX) from
Technical Analysis of STOCKS & COMMODITIES, Volume 12:
July 19, 1993, to June 30, 1997, almost four years of data. I
January.
set the interest rate at 4% annualized, and a trading cost of
Tukey, J.W. [1976]. Exploratory Data Analysis, Addison-Wesley:
0.1% for entry and exit. This is realistic, since I can trade up
Reading, MA.
to 1,000 shares through Fidelity Web Express for $14.95, and
a typical trade might be 300 shares of a $50 stock. †See Traders’ Glossary for definition S&C

Copyright (c) Technical Analysis Inc. 5