Stocks & Commodities V. 10:3 (108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande, Ph.D.

Adapting Moving Averages To Market Volatility

by Tushar S. Chande, Ph.D.

If a market is active, it has volatility: that cannot be avoided. And because the market is continuously
changing, an indicator that attempts to predict market activity must itself adapt and change. How?
Tushar Chande presents a dynamicnot staticindicators: a variable-length moving average, which
adapts to the volatility in question by exponentially smoothing data based on standard deviation.

echnicians can be trend followers or contrarians. Trend followers use price-based indicators, such as

moving averages, while contrarians prefer oscillators such as overbought-oversold indicators. But the
market never does quite the same thing twice, and so no indicator works all the time. The market is
dynamic, adjusting rapidly to information: a continuous tug of war between greed and fear, fact and
fiction. Technical indicators, on the other hand, are static, mechanically applying the same formula to the
relevant data. What is needed is a combination, dynamic indicators that will automatically adapt to the
changing nature of markets, a new class of dynamic indicators that combine exponential moving averages
with other technical indicators to adapt automatically to changing price behavior. What is needed is an
exponential moving average with a continuously variable smoothing index that adjusts rapidly to changes
in price behavior. The smoothing index can be tied to any market variable. It is the continuous, not
discrete, changes in the smoothing index that increases the sensitivity of these moving averages to
changes in price behavior. These new dynamic exponential averages can be referred to a variable index
dynamic average (VIDYA ).
Let us first examine exponential moving averages and how they can be modified to obtain VIDYA , and in
turn compare VIDYA with conventional indicators to illustrate its dynamism. Then we will combine
dynamic averages to derive other indicators and then illustrate their effectiveness.
BUT FIRST, THE BACKGROUND
Exponential moving averages give greater weight to more recent data. An exponential moving average

Article Text

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 10:3 (108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande, Ph.D.

may be defined as:

(1) Ed = t Cd + (1-t) Ed-1
where
Ed is the new value of the moving average
Ed-1 is the previous value
Cd is the new data value
t is the smoothing constant of the average
The smoothing constant t of the average may be referred to as the smoothing index of the moving
average, so it no longer has to be visualized as a numerical constant. Implicitly, t must be less than 1 for
the term (1-t) to be positive. As the smoothing index t increases, the new value has a greater proportion
of the most recent data and the exponential average moves more rapidly. Conversely, as t decreases, more
weight is given to the previous value, giving a heavily smoothed average that changes quite slowly. Thus,
a larger value of t makes the exponential moving average more sensitive to new data; a smaller index t
makes it less sensitive.
George Arrington in the June 1991 S TOCKS & COMMODITIES discussed a variable-length simple moving
average in which the number of days changed by discrete integers. The length is increased or decreased
by an integral number of days based on the magnitude of price changes. He argued that this approach
does not work well with exponential moving averages. A closer look at the smoothing constant t suggests
that it could be a continuous variable (index), thus allowing the use of fractional time periods. For
example, we can write t as
(2) t = k V
where k is a numerical constant such as 0.15 and V is a dimensionless market-related variable, such as a
ratio of the standard deviation of the market's closing prices over two different periods.
The smoothing constant is simply a mechanism for incrementing the old value of the moving average to a
new value, and a fractional value of t less than 1 prevents instability. Varying the smoothing index
corresponds to taking larger or smaller portions of the latest data to update the moving average.
Any indicator may be used to connect the index t to the nature of the market's price changes. For
example, the midpoint oscillator, %M, may be used as a measure of the market and inserted as V in
equation 2. By design, the moving average will move faster as prices approach an overbought or oversold
condition. Alternately, market momentum indicators (say, the 26-week price rate of change) may be used
so that changes occur more quickly as prices change rapidly, slowing down as prices stabilize. This
ability to quicken or slow down gives these variable-index exponential averages their dynamism.
DEFINING VIDYA AND RAVI
Specifically, I constructed a variable-index dynamic moving average (V IDYA ) connecting the smoothing
index to the market's volatility as follows:

Article Text

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 10:3 (108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande, Ph.D.

VIDYAd = k n Cd + 1 k n VIDYAd -1
ref
ref

Here, the subscripts d and d-1 denote the new and old time period, C is the closing price at the end of
period d, (sigma) is the standard deviation of the market's prices over the past n periods, is a
reference standard deviation of the market over some period of time longer than n, and k is a numerical
constant. The reference standard deviation could also be an arbitrary value to obtain the desired degree of
smoothing.
From an investor's viewpoint, a "long" VIDYA can be defined with k=0.078, corresponding roughly to a
25-week exponential moving average. A 13-week standard deviation is used to adapt to market volatility.
A reference standard deviation of 6 is used, which represents a 10-year average for weekly Standard &
Poor's 500 data. More precisely, the long V IDYA is given by
(4) VIDYA Ld = 0.078 13 / 6 Cd + (1 - 0.078 13/6) VIDYA Ld-1
A "short" dynamic average is also defined with k = 0.15, roughly equal to a 12-week exponential moving
average. The standard deviation of closing prices is calculated over 10 weeks. The value of the reference
standard deviation is set at 4.
(5) VIDYA Sd = 0.15 10/4 Cd + (1-0.15 10/4) VIDYA Sd-1
To clarify the dynamism of these averages, I tabulated the 13-week standard deviation of the S&P 500
weekly close during three market periods. Also shown is the effective smoothing index using equation 4.
Then I estimated the effective length of the equivalent simple moving average using the well-known
formula for the smoothing constant of an exponential moving average (2/(n+1)), where n is the length of
the equivalent simple moving average.

The dynamic range of VIDYA L is about a factor of 10 (i.e.,30.75 3.18), since it adjusts all the way from a
rather long to a very short moving average based on market volatility. As market volatility increases, the
effective length decreases. Even greater dynamic range is possible, as long as the factor (1-t) is positive.
Clearly, V IDYA L is superbly responsive to the market. VIDYA S also exhibits a similar dynamism.
Now, we define the rapid adaptive variance indicator (R AVI ), defined as
(6) RAVI d = VIDYA Sd - VIDYA Ld
where the long and short dynamic averages are as defined in equations 4 and 5.
Article Text

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 10:3 (108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande, Ph.D.

OTHER TRADING STRATEGIES

All trading strategies based on moving averages can be implemented using VIDYA . For example, RAVI is
a two-average crossover indicator. A buy signal is generated when R AVI turns positive from a negative
value. Similarly, a sell signal occurs when R AVI turns negative from a positive value.

I used weekly S&P 500 data to illustrate the smoothing

characteristics using a VIDYA L and VIDYA S, representing heavy and
light smoothing, respectively.
RAVI can be combined with its moving average to simulate a moving average convergence/divergence
(Macd) strategy. Due to its sensitivity, this more responsive or TurboM ACD can be defined as follows:
(7) TMACDd = RAVI d - (0.2 RAVI d + 0.8 RAVI d-1)
where the second term represents the trigger line, an exponential moving average with a smoothing
constant of 0.2. A buy signal is produced when TurboM ACD changes from negative to positive values,
and a sell signal occurs when it goes from positive to negative values.
BACK TESTING RESULTS
The results of back testing can be divided into two parts: first, the smoothing characteristics of V IDYA ,
and second, long trades using RAVI and TMACD. Now let us use moving average crossovers and MACD for
comparison.
I used weekly S&P 500 data to illustrate the smoothing characteristics using a VIDYA L and VIDYA S,
representing heavy and light smoothing, respectively. Figure 1 covers a market period from January 1990
to July 1991 and compares VIDYA L to the equivalent exponential moving average with a smoothing
constant of 0.078 . Note how the variable index dynamic average adjusts rapidly to the drop in August
1990. In contrast, an exponential moving average barely responds to the rapid price changes. VIDYA L
takes small steps when the market trades in a narrow range and takes large ones when the market makes
big moves in any direction. Observe how VIDYA L responded near the market bottom in
September-October 1990, while rising rapidly and leveling out in May 1991 when the market entered a
trading range.
The lightly smoothed VIDYA S follows the market even more closely, as shown in Figure 2. It also flattens
as the market trading range narrows, as in June-July 1990. Clearly, it can be used as a trigger line in
market trading systems. Even though the equivalent exponential moving average with a smoothing
constant of 0.15 responds quickly to price changes, VIDYA S responds with even greater agility.
An even better picture of VIDYA 's responsiveness to price changes can be seen in Figure 3, which
compares the smoothing behavior of VIDYA L with a short exponential moving average with a smoothing
constant of 0.15. The long dynamic average is more sensitive to rapid changes than even the short
exponential average and tracks price changes more powerfully.

he responsiveness of VIDYA and its vulnerability to instabilities was severely tested during the severe

Article Text

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 10:3 (108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande, Ph.D.

FIGURE 1: The S&P 500 weekly closes (A) are plotted along with both the long variable index dynamic
moving average (B) and with the equivalent exponential moving average (C). Note how quickly the
indexed moving average responds to the decline by the S&P 500 in August 1990.

FIGURE 2: The S&P 500 weekly closes (A) are plotted along with both the short variable index dynamic
moving average (B) and with the equivalent exponential moving average (C). The short version of the
indexed moving average follows the market closely.

Stocks & Commodities V. 10:3 (108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande, Ph.D.

FIGURE 3: The S&P 500 weekly closes (A) are plotted this time with the long variable index dynamic
moving average (B) and with the short exponential moving average (C). The long version of the indexed
moving average still follows the market more close/y than a short exponential moving average.

FIGURE 4: The S&P 500 weekly closes (A) are plotted this time with the long (B) and short (C) variable
index dynamic moving average and with the equivalent long exponential moving average (D). The
narrowing of the difference between the two indexed moving averages indicates a possible trend
reversal

Stocks & Commodities V. 10:3 (108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande, Ph.D.

FIGURE 5: Using the difference between the short (A) and long (B) versions of the varriable index
dynamic moving average produces the rapid adaptive variance indicator, creating a crossover method
for trading signals.

Stocks & Commodities V. 10:3 (108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande, Ph.D.

market correction of October 1987 (Figure 4). The narrowing of the variance between the two dynamic
averages signaled a possible market top in September, confirmed in early November as the two averages
crossed over. The lightly smoothed V IDYA S tracks the market tightly and the heavily smoothed VIDYA L
responds rapidly as well. In comparison, the equivalent long exponential average glances over the
extreme price changes. Note the slight instability caused by the market volatility when the short average
dipped below the market low in early December.
The quantitative results are now easier to interpret, as we have a good feel for the smoothing behavior of
these dynamic averages. Now look at RAVI . A buy is signaled when RAVI turns positive from negative
territory and a sell occurs when RAVI goes negative from positive ground, which is the same as the short
VIDYA crosses over or under the long VIDYA (Figure 5).
The results of long trades using RAVI over a test period of January 1980 to July 1991 are in Figure 6.The
total point gain is approximately 207 points (with one open trade) using this strategy. If trade 9 were
closed on July 12, 1991, the gain would be approximately 265 S&P 500 points. For comparison, the same
moving average crossover strategy using the equivalent exponential moving averages with smoothing
constants of 0.15 and 0.078 (same as the k values for VIDYA S and VIDYA L) produced a gain of just 122
S&P 500 points with five long trades (see Figure 7). If the currently open trade for the exponential
averages were closed on July 12, 1991, the gain would be 159 S&P 500 points.
The MACD indicator is composed of the difference between two exponential moving averages with
smoothing constants of 0.15 and 0.077. A moving average of the difference with a smoothing constant of
0.2 is used as a trigger line. I tested the profitability of MACD over the same test period using MetaStock
software. The MACD indicator produced 20 trades and a gain of 205 S&P 500 points (Figure 8).
Currently, there are no open trades with M ACD, since it issued a sell signal on May 17, 1991. Trades with
TurboM ACD over the same period are summarized in Figure 9. In total, there were 25 long trades, with a
gain of approximately 218 points in all.

comparison of each of the four methods discussed is found in Figure 10. Clearly, R AVI beat both the

simple crossover of exponential averages and MACD over the past 11 years. RAVI produced more trades
than did the simple crossover model but less than half as many as did the MACD. A look at the detailed
trades shows that TMACD signaled key turning points two to three weeks before MACD. Its greater
sensitivity produced more trades, which is a key limitation of the MACD approach from an investor's
perspective. RAVI can also be used as an overbought-oversold indicator, since its values peak at over
+10 or -10.
TO SUMMARIZE
Overall, the back testing results show that VIDYA tracks the market (as measured by the S&P 500 index)
better than exponential moving averages do using a fixed smoothing constant. Trading strategies based
on VIDYA seem to perform better than those based on exponential moving averages, as summarized in
Figure 5.

Article Text

Copyright (c) Technical Analysis Inc.

Stocks & Commodities V. 10:3 (92108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande
FIGURE 6

FIGURE 8

R A VI - LONG POSITIONS ON
5/30/80 to 7/12/91
D A TE

TRADE

LY

S&P500

5/30/80
9/11/81
9/30/82
3/9/84
8/17/84

BUY
SELL
BUY
SELL
BUY

111.24
121.61
120.97
154.35
164.14

9/26/86
10/31/86
10/16/87
12/18/87

SELL
BUY
SELL
BUY

232.33
243.98
282.70
249.16

12/25/87
1/1/88
1/8/88
1/15/88

SELL
BUY
SELL
BUY

252.02
247.08
243.40
252.06

1/26/90
5/11/90
8/10/90
11/30/90

SELL
BUY
SELL
BUY

325.80
352.00
335.52
322.22

7/12/90

SELL **

380.25

Total long trades:

9
Profitable longs: 7 (78%)
Biggest gain:

GAIN/LOSS
TO TAL
POINTS
POINTS

S&P 500, MACD

(9-unit moving average of indicator)
6/27/80 to 7/12/91
Total long trades :
20
Profitable longs : 11 (55.0%)
Total buy stops :
0

Total short trades : 0

Profitable shorts : 0 (0.0%)
Total sell stops
: 0

10.37

10.37

33.38

43.75

Biggest gain
:
Successive gains :

70.760
1

Biggest loss
: -15.230
Successive losses : 1

68.19

11.94

Total gain/loss
: 204.560
Total gain/loss ($) : 1,344.96

Average gain/loss : 10.228

Total gain/loss (%) : 134.50

38.72

150.66

2.86

153.52

-3.68

149.84

73.74

223.58

-16.48

207.10

58.03

265.13

73.74

** Trade closed for completeness; no sell signal from this model on

7/12/90

The results of long positions only are presented. The sell signal on July
12, 1991, is the open profit on that date.

FIGURE 7

PRICE OSCILLATOR (LONG POSITIONS ONLY)

S&P 500, Formula A
6/27/80 to 7/12/91
Total long trades :
5
Profitable longs : 3 (60.0%)
Total buy stops :
0

Total short trades : 0

Profitable shorts : 0 (0.0%)
Total sell stops
: 0

Biggest gain
:
Successive gains :

84.080
1

Biggest loss
: -40.490
Successive losses : 1

Total gain/loss
: 122.170
Total gain/loss ($) : 1,004.77

Average gain/loss : 24.434

Total gain/loss (%) : 100.48

Using only long positions in a crossover trading strategy based on the equivalent short and long exponential moving averages produced a gain of 122.17
points in the S&P 500. If the current open trade had been closed on July 12,
1991, the total gain would have been 159 points.

Testing the standard moving average convergence/divergence indicator for

long positions resulted in the above. The MACD uses the difference between a
26- and 12-day exponential moving average and a nine-day exponential moving
average of the MACD line for the signal line. Crossovers of the MACD line and
the signal line generate the buy and sell signals.

Stocks & Commodities V. 10:3 (92108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande
FIGURE 9

FIGURE 10

TURBO MACD - LONG POSITIONS ONLY

5/30/80-7/12/91
D A TE

TRADE

S&P500

5/30/80
1/16/81
11/27/81
1/15/82

BUY
SELL
BUY
SELL

111.24
134.77
125.09
116.33

4/16/82
6/18/82
8/20/82
12/10/82

BUY
SELL
BUY
SELL

116.81
107.28
113.02
139.57

1/14/83
6/17/83
3/8/84
11/30/84

BUY
SELL
BUY
SELL

146.65
169.13
162.35
163.58

1/18/85
3/22/85
5/17/85
8/2/85

BUY
SELL
BUY
SELL

171.32
179.04
187.42
191.48

11/8/85
1/17/86
2/21/86
4/11/86

BUY
SELL
BUY
SELL

193.72
208.43
224.62
235.97

6/27/86
7/11/86
8/29/86
9/12/86

BUY
SELL
BUY
SELL

249.6
242.22
252.93
230.67

10/31/86
3/13/87
6/19/87
9/4/87

BUY
SELL
BUY
SELL

243.98
289.89
306.97
316.7

11/20/87
12/4/87
12/11/87
4/15/88

BUY
SELL
BUY
SELL

242
223.92
235.32
259.77

6/10/88
7/22/88
9/30/88
11/18/88

BUY
SELL
BUY
SELL

271.26
263.5
271.91
266.47

12/23/88
3/3/89
4/14/89
6/16/89

BUY
SELL
BUY
SELL

277.87
291.18
301.36
321.35

7/28/89
9/15/89
12/8/89
1/12/90

BUY
SELL
BUY
SELL

342.15
345.06
348.69
339.93

3/16/90
4/27/90
5/4/90
7/13/90

BUY
SELL
BUY
SELL

341.91
329.11
338.39
367.31

10/5/90
3/15/91

BUY
SELL

311.5
373.59

EXP MA *
CROSSOVER

GAIN/LOSS
TO TAL
POINTS
POINTS
23.53

23.53

-8.76

14.77

-9.53

5.24

26.55

31.79

22.48

54.27

1.23

55.5

7.72

63.22

4.06

67.28

14.71

81.99

11.35

93.34

-7.38

85.96

-22.26

63.7

45.91

109.61

9.73

119.34

-18.08

101.26

24.45

125.71

-7.76

117.95

-5.44

112.51

13.31

125.82

19.99

145.81

2.91

148.72

-8.76

139.96

-12.8

127.16

28.92

156.08

62.09

218.17

Total long trades:

25
Profitable longs: 16 (64%)
Biggest gain:
62.09
Testing the moving average crossover method produced these results.

Total long trades:

Profitable longs:
Biggest gain:
Biggest loss:
Average gain:
Total gain:

6
4 (67%)
84.08
-40.49
26.56
159.37

MACD
20
11 (55%)
70.76
-15.23
10.23
204.56

R A VI *
9
7 (78%)
73.74
-16.48
29.45
265.13

TURBO
MACD

25
16 (64%)
62.09
-22.96
8.73
218.17

NOTES:
* Assumes last open trade closed on 7/12/91; no sell signal from these
models at 7/12/91
Comparing the four different methods highlights the strengths of using moving
averages that respond to changes in the market volatility.

Stocks & Commodities V. 10:3 (92108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande

FIGURE 11

The basic formulas for each column are presented in a Lotus spreadsheet. Column
F is the weekly close for the S&P 500, while column G is the exponential moving
average using a 0.078 smoothing constant. Column H is the EMA using a 0.15
smoothing constant, and column I is the Lotus formula for the long VIDYA. STD is
the standard deviation formula in Lotus for the cell range stated within the
parenthesis. Column J is the short version formula. Column K is the difference
between column J and I.

Stocks & Commodities V. 10:3 (108-114): Adapting Moving Averages To Market Volatility by Tushar S. Chande, Ph.D.

In sum, this new class of variable index dynamic moving

averages VIDYA adapts moving averages to the changing
nature of markets.
A single formulation tracks market changes well despite the increased volatility of recent years. VIDYA
serves as a variable-length exponential moving average, taking a greater "bite" out of the most recent data
as market volatility increases. Like all moving averages, VIDYA also lags the market, since it is derived
from past data probably its biggest limitation.
In sum, this new class of variable index dynamic moving averages VIDYA adapts moving averages
to the changing nature of markets. Any dimensionless market variable can be used to link these averages
to the market. The figures provided illustrate well the responsiveness of these averages to market
changes. Common trading strategies for moving averages can be implemented using VIDYA ,
demonstrated with RAVI and TurboM ACD. VIDYA is also well suited for setting stops, as it closely tracks
the market. VIDYA should be a formidable and dynamic addition to the trader's arsenal.
Tushar Chande holds a doctorate in engineering from the University of Illinois and a master's degree in
business administration from the University of Pittsburgh.

REFERENCES
Arrington, George R. [1991] . "Building a variable-length moving average," Technical Analysis of STOCKS
& COMMODITIES, Volume 9: June.
Chande, Tushar S. [1991]. "The midpoint oscillator," Technical Analysis of STOCKS & COMMODITIES:
Volume 9: September.
Pindyck, Robert S., and Daniel L. Rubinfeld [ 1981 ] . Econometric Models and Economic Forecasts,
McGraw-Hill Inc.

Figures

Copyright (c) Technical Analysis Inc.