Вы находитесь на странице: 1из 4

The Information Content of Technical

Trading Rules:
Evidence from US Stock Markets

Yong Ye
School of Economics,
Huazhong University of Science and Technology,
Wuhan, 430074, China
tigerey@hotmail.com
AbstractThis paper examines the predictive power
of momentum indicators, a kind of technical trading
rules measuring short-term momentum, on three
popular US stock market indices, the Dow Jones
Industrial Average, Standard & Poors 500
Composite Index, and NASDAQ Composite Index.
Generally, the main findings indicate that returns
conditional these trading rules are significantly
different from unconditional returns. Null models,
such as random walk, first-order autocorrelation,
and GARCH-M, cannot explain the excess profit
made by these rules.
Key words-Technical analysis, Momentum Indicators,
Return Predictability
I. INTRODUCTION
Since the 1990s, technical analysis has
become a more active research area in finance.
Many new studies argue that it may provide
incremental information and hence contain some
predictive power.[1][2][3]. This study intends to
analyze the predictability and profitability of a
kind of important technical trading rules, the
momentum indicators.
So-called momentum refers to the rate at
which prices change over certain time period. On
this definition, a large body of literature
documents that momentum strategy can provide
stock return predictability on a relatively long
horizon[4-7].
Technical analysts also believe that
momentum conveys information about price
direction as well as strength and have developed a
lot of momentum indicators to detect investment
opportunities. Although they have different
expressions, these indicators share the similar idea
with the fundamental momentum, and are utilized
popularly by technicians and investors for the
purpose of predicting the price trend. In this paper
I will introduce three most widely used momentum
indicators: Stochastics (KDJ), Relative Strength
Index (RSI), and Moving Average Convergence-
Divergence (MACD), and examine their statistical
efficiency for capturing investment opportunity on
three popular market indices, the Dow Jones
Industrial Average(DJIA), Standard & Poors 500
Composite Index, and NASDAQ Composite Index.
Standard t-test for comparing the mean difference
of two samples and Kolmogorov-Smirnov two-
sample test will be used to test the difference
between unconditional returns and returns
conditional on trading rules, and bootstrap analysis
will be applied to generate the test statistics of
these rules and thus evaluate their profitability.
II. THE RULES
A. Stochastics
Stochastics is an approach to measure price
movement velocity. Two lines are used to express
its idea: %K, which measures the relative position
of the current close in a certain time range, and
%D, a three-day moving average of the %K line.
When %K line crosses above or below %D line,
buy or sell signals were generated to imply the
future trend of price movement. Sometimes we use
J-value to present the divergence of %K and %D,
so this rule is also called KDJ indicator[8].
The formula for calculating stochastics is as
follows:

100 % u

n n
n t
t
L H
L C
K
(1)
where
C
t
is current closing price
H
n
is the highest price of the last n days
L
n
is the lowest price of the last n days.
3 / ) % % (% %
1 2 t t t t
K K K D

(2)
If we choose n equals 5, 10, and 15 in
equation (1) and apply a 50-day average of %D as
the long-period moving average, then the rules
without a band can be denoted as (5, 50, 0), (10, 50,
0) and (15, 50, 0) respectively.
B. Relative Strength Index
The formula for calculating relative strength
index is:
3!
___________________________________
978-1-61284-109-0/11/$26.00 2011 IEEE


RS
RSI


1
100
100 (3)

DOWN closes s period given of Average
UP closes s period given of Average
RS
'
'

(4)
Similar with the stochastics oscillator, relative
strength index can be used as a trend indicator as
well. The most popular period for calculating RSI
is 14-day. This study will also consider 9-day and
20-day as other choices for robust test. When 50-
day is used for the long-term period moving
average, the rules without a band can be expressed
as: (9, 50, 0), (14, 50, 0), and (20, 50, 0).
C. Moving Average Convergence-Divergence
Moving Average Convergence-Divergence
applies two lines to present market equilibrium,
overbuy and oversell conditions, and signal line
crossovers. One is the difference of two
exponential moving averages (EMAs) of price
with different time period, and the other is an
exponential moving average of the first line.
Denote EMA(n)
t
a n-day exponential moving
average at time t, then

1
) (
1
1
1
2
) (

t t t
n EMA
n
n
C
n
n EMA (5)
where
C
t
is the current price
EMA(n)
t-1
is the n-day EMA at previous time
t-1. The initial value is a n-day simple moving
average of price.
According to definition, MACD is the
difference of two EMAs of price series, so

t
(6)
Denote DEM the signal line, the exponential
moving average of MACD, then
t t
period short EMA
period long EMA MACD
) (
) (



1
1
1
1
2

t t t
DEM
n
n
MACD
n
DEM
(7)
similarly, the initial value of DEM
t-1
can be
substituted by the MACD value of previous day.
Normally, technical analysts apply the
difference of 12-day and 26-day EMA to generate
the momentum oscillator MACD, and a 9-day
EMA of this oscillator to act as the signal line. So
the rule without a band can be expressed as (12, 26,
9, 0). This study will also examine rules (9, 15, 9,
0) and (9, 26, 26, 0).
III. THE DATA
Historical series of Standard & Poors 500
index and NASDAQ composite index were
retrieved from Center for Research in Securities
Prices (CRSP) database. Because of the data
availability, the time span for DJIA and S&P 500
is from July 1, 1962 to January 31, 2003 (totally
10236 and 10234 observations respectively), and
the NASDAQ data is from February, 5, 1973 to
January 31, 2003 (7574observations).
IV. EMPIRICAL RESULTS
A. T-Test Results
For each period, identified signals will form
corresponding conditional return distributions. To
test the predictability of rules, the standard method
is to compare the test statistics, mean return and
variance, of conditional returns with that of
unconditional returns. Standard t-test and
Kolmogorov-Smirnov test can be applied for this
purpose, one is parametric and the other is non-
parametric.
Table I shows the statistical results of
conditional buy and sell returns and the results of
t-test for all the three indices for each period. The
reported mean buy and sell returns are quite
distinguishable. Almost all the buy returns are
positive and almost all the sell returns are negative,
although only less than half tests reject the null
hypothesis that the returns equal the unconditional
returns at 5% significance level. This implies that
these rules are able to predict the price trend to
certain extent. The test for the difference of the
mean of buy-sell returns gives even stronger
evidence. All the buy-sell returns are positive and
the t-tests for the difference of buy-sell returns
from zero are highly significant.
TABLE I. STANDARD T-TEST RESULTS
Rules Buy Sell Buy-Sell
KDJ (5,50,0) 0.00119 -0.00036 0.00155
(1.61) (-1.54) 13.677
(10,50,0) 0.00072 -0.00011 0.00083
(0.75) (-0.81) 6.19
(15,50,0) 0.00103 -0.00123 0.00226
(1.04) (-3.10) 20.99
RSI (9,50,0) 0.00159 -0.00084 0.00243
(2.09) (-2.76) 22.80
(14,50,0) 0.00238 -0.00067 0.00305
(3.29) (-2.10) 29.17
(20,50,0) 0.00168 -0.00119 0.00287
(2.11) (-3.06) 27.32
MACD (9,15,9,0) 0.00036 -0.00094 0.0013
(0.22) (-1.68) 11.05
(12,26,9,0) 0.00170 -0.00059 0.0023
(2.02) (-1.25) 21.39
(9,26,26,0) 0.00143 -0.00068 0.00211
(1.17) (-1.10) 19.49
3!8
Avg. 0.00134 -0.00073 0.00207
B. Kolmogorov-Smirnov Test
Note that the summary statistics of
unconditional returns indicates the characteristic of
leptokurtic, auto-correlated, and conditionally
heteroskedastic, we further use the Kolmogorov-
Smirnov two-sample test to examine the difference
between conditional and unconditional return
distributions, and the test results have been
reported in Table II. For Stochastics, its more
efficient when applied on S&P 500 series and
NASDAQ series. For RSI, most rules are
significant at least at 5% level. For MACD.
Almost all the rules for all the three indices are
insignificant. It seems that MACD indicator is
useless for predicting the market trends.
C. Bootstrap Tests
The bootstrap methodology can be employed
to evaluate the profitability of those momentum
indicatorss[9]. To generate artificial series, we
need to specify a model describing the stock price
TABLE II. KOLMOGOROV-SMIRNOV TEST RESULTS
Rules Buy Sell
KDJ (5,50,0) 0.1092 0.047
(0.0003) (0.1507)
(10,50,0) 0.0727 0.0521
(0.1539) (0.2291)
(15,50,0) 0.0548 0.0923
(0.5855) (0.0006)
RSI (9,50,0) 0.08 0.0901
(0.0508) (0.0010)
(14,50,0) 0.1139 0.0655
(0.0043) (0.0642)
(20,50,0) 0.1055 0.0878
(0.0232) (0.0137)
MACD (9,15,9,0) 0.0511 0.0767
(0.6813) (0.2226)
(12,26,9,0) 0.1107 0.0713
(0.0472) (0.4124)
(9,26,26,0) 0.066 0.071
(0.6362) (0.5239)
behavior and get the residuals of estimation. We
employ three classical models, random walk,
autocorrelation at lag 1, and GARCH-M, to fit the
raw series.
The first model is a random walk which can
be expressed as

t t t
r r H
1
(8)
where r
t
is the return on day t and
t
is
independent, identically distributed. This set-up is
consistent with the Efficient Market Hypothesis
and is still the most representative model for stock
market behavior so far. With an initial number, its
simple to generate simulated series by scrambling
the raw returns with replacement.
Considering the significance of
autocorrelation of the raw indices series, the
second model is for the simulation is an AR (1),

t t t
r r H E D
1


(9)
where and can be estimated from the raw
series using OLS. Then the residuals are chosen
with replacement and used along with estimated
parameters to generate 500 artificial series.
The final model describing the market is a
GARCH-M model:
t t t t
h r r H D D D
2 1 1 0
(10)
(11)
1 2
2
1 1 0

t t t
h h E H E E
) 1 , 0 ( ~
2 / 1
N z z h
t t t t
H
In this model, the error term is not IID. Its
assumed to be conditionally normally distributed
on variance, which is a linear function of the
square of last periods errors and of the conditional
variance of last period. This model specifies the
market by describing the phenomenon of volatility
dependence. That is, high (or low) volatility
periods are likely to be followed by same kind of
periods. Similarly, to generate simulations,
residuals are scrambled with replacement and
added to the fitted series with estimated parameters.
For the random walk model, among the three
indicators, it seems that RSI is more reliable than
the other two. All the three rules of RSI give
significant test results for buy, sell, and buy-sell
returns. However, all the insignificant results are
focus on KDJ and MACD indicators. For AR(1)
model, among the three indicators, results further
confirm the advantage of RSI. Almost all the tests
for buy, sell, and buy-sell returns with RSI are
highly significant except that RSI ( 14,50,0) for
sell is marginally significant. For the KDJ and
MACD respectively, there is only one rule for buy
and sell is significant. However, this has been
improved when considering tests on buy-sell
returns with these indicators. For GARCH-M
model, all RSI rules are still significant for buy,
sell, and buy-sell returns. Except for buy-sell
returns, results tend to accept the null hypothesis
that KDJ and MACD rules cannot make an excess
profit.
V. CONCLUSION
In summary, the results show that momentum
indicators have the predicative power for future
stock trends. Almost all the mean conditional
returns following a sell signal is negative. Almost
all the average difference of buy and sell returns is
significantly close to zero. The conditional buy,
3!9
sell, and the difference of buy and sell returns from
original index series cannot be explained by null
models such as random walk, first order
autocorrelation, and GARCH-M. There is no
significant difference between the conditional
standard deviation from raw series and that from
bootstrapping series for most rules over all the
three indices, indicating that risk premium cannot
be accounted for the excess profit from trading
rules.
Particularly, all the tests strongly support the
efficiency of relative strength index and
stochastics, but only very weak evidence confirms
the efficiency of moving average convergence-
divergence.
So the results suggest that indices series may
not fully reveal all the public information about the
market. Technical trading rules in this study may
capture the nonlinear dependencies in the series
and thus make excess profit.
ACKNOWLEDGEMENTS
This paper is sponsored by the National
Natural Science Foundation of China ( Project No.
70971051). The author would also like to thank the
valuable helps and comments of Prof. Xiaozu
Wang, School of Management, Fudan University,
and Prof. Qiming Tang, School of Economics,
Huazhong University of Science and Technology,
and the two anonymous referees.
REFERENCES
[1] Brock, W., Josef Lakonishok, and Blake LeBaron,
Simple Technical Trading Rules and the Stochastic
Properties of Stock Returns, Journal of Finance, vol.47,
Dec.,1992.
[2] Lo, Andrew, W., Harry Mamaysky and Jiang Wang ,
Foundations of Technical Analysis : Computational
Algorithms , Statistical Inference , and Empirical
Implementation, Journal of Finance, vol.55, August,
2000.
[3] Blume, Lawrence, David Easley, and Maureen
OHara , Market Statistics and Technical Analysis:
The Role of Volume, Journal of Finance, vol.49,
pp153-181,1994.
[4] Chen, Hsiu-Lang, Narasimhan Jegadeesh, and Russ
Wermers, The Value of Active Mutual Fund
Management: An Examination of the Stockholdings and
Trades of Fund Managers, Journal of Financial and
Quantitative Analysis, vol.35, pp343-368, 2000.
[5] Jegadeesh, Narasimhan, and Sheridan Titman, Returns
to Buying Winners and Selling Losers: Implications for
Stock Market Efficiency, Journal of Finance, vol.48,
pp65-91, 1993.
[6] Lehman, B.N., Fads, Martingales, and Market
Efficiency, The Quarterly Journal of Economics,
vol.105, pp1-28, 1990.
[7] Rouwenhorst, K.G., International Momentum
Strategies, Journal of Finance, vol.53, pp267-284,
1998.
[8] Edwards and Magee, Technical Analysis of Stock
Trends , sixth edition , John Magee Inc., Boston,
1996.
[9] Efron, B., Bootstrap Methods : Another Look at the
Jackknife, The Annals of Statistics vol.7, pp1-26,
1979
TABLE III. BOOTSTRAP TEST RESULTS
Random Walk AR(1) GARCH-M
Rules Buy-Sell p-value Buy-Sell p-value Buy-Sell p-value
KDJ (5,50,0) 0.00003 (0.002) 0.00002 (0.012) -0.00003 (0.012)
(10,50,0) 0.00003 (0.142) 0.00002 (0.144) 0.00000 (0.150)
(15,50,0) 0.00004 (0.002) 0.00006 (0.006) -0.00002 (0.002)
RSI (9,50,0) -0.00002 (0.000) -0.00001 (0.000) 0.00004 (0.000)
(14,50,0) 0.00006 (0.000) 0.00008 (0.000) -0.00001 (0.000)
(20,50,0) 0.00001 (0.000) -0.00001 (0.002) 0.00005 (0.000)
MACD (9,15,9,0) -0.00005 (0.078) -0.00003 (0.088) -0.00001 (0.098)
(12,26,9,0) -0.00004 (0.016) 0.00001 (0.012) 0.00005 (0.030)
(9,26,26,0) 0.00005 (0.038) -0.00005 (0.036) 0.00007 (0.038)
Average 0.00001 (0.000) 0.00001 (0.000) 0.00001 (0.000)
3?0

Вам также может понравиться