Technical analysis as the representation of typical cognitive biases
-
Upload
independent -
Category
Documents
-
view
0 -
download
0
Transcript of Technical analysis as the representation of typical cognitive biases
International Review of Financial Analysis
13 (2004) 217–225
Technical analysis as the representation
of typical cognitive biases
Piotr Zielonka
Warsaw University SGGW, Rakowiecka 26/30, 02-528 Warsaw, Poland
Accepted 13 February 2004
Abstract
The present research provides a justification for the popularity of the technical analysis. It finds
that financial analysts firmly discriminate between two types of technical signals—those based on
typical cognitive biases and ‘‘empty’’ signals that sound like a technical analysis but are without any
connotation with psychological inclinations.
At the same time that they treat them differently, different analysts rate these items very similarly.
These results suggest that the popularity of technical analysis is associated with its relation to the
typical cognitive biases of humans.
D 2004 Elsevier Inc. All rights reserved.
JEL classification: D84; G14
Keywords: Behavioral finance; Psychological biases; Technical analysis
1. Introduction
In spite of a long list of publications showing that market movement is random
(Alexander, 1961; Cootner, 1964; Malkiel, 1996; Samuelson, 1965) or at least very
difficult to predict (Lo & MacKinlay, 1999), considerable effort is devoted to forecasting
future stock prices.
One of the forecasting tools very popular among practitioners is technical analysis.
When the keyword ‘‘technical analysis’’ is put into the Internet seek engine Google,
326,000 URLs are located as opposed to only 49,700 URLs for ‘‘portfolio theory.’’
Technical analysis is the examination of past price movements to forecast future price
movements. As previous research shows, it is more an art than an objective analytical tool,
1057-5219/$ - see front matter D 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.irfa.2004.02.007
E-mail addresses: [email protected], [email protected] (P. Zielonka).
P. Zielonka / Int. Rev. Financ. Analy. 13 (2004) 217–225218
and its efficiency has not been proven (Fama & Blume, 1966; Jensen & Bennington,
1970). Technical analysis is open to interpretation. Many times, two technicians will look
at the same chart and paint two different scenarios or see two different patterns, and both
would be able to come up with logical support to justify their positions. Moreover,
technical analysts’ opinions are often extremely ambiguous, like ‘‘the stock seems
positioned to break through resistance at US$65, but a stop loss should be placed at
US$54.’’
In addition, even if stock prices completely followed a random walk, people would be
able to convince themselves that there are patterns having predictive value. In laboratory
experiments, subjects are reported to have found patterns in purely random sequences of
stock prices (Roberts, 1959; Warneryd, 2001).
An extensive body of research examining the efficiency of technical analysis on the
Polish stock market has shown negative results. No efficient technical analysis tools were
found (Janiec, 2000); neither was there any evidence showing that any replicable technical
scheme works for an extended period of time.
The following fundamental question arises: if there is no evidence that technical
analysis can result in exceptional profits, why is it so popular? Since technical analysis
started at the beginning of the 20th century when the possibilities of testing its
efficiency were very limited, it has become more and more popular, as it offers an
unlimited set of tools and signals and seems to be an interesting method of market
analysis. However, recent studies since the 1950s have shown that its efficiency is
very low.
So why does technical analysis still remain so popular? The first answer may be
related to tradition and education. Investors have used it for over a century, and new
market participants learn it from more advanced market participants. The second answer
is related to some common cognitive biases that affect people’s behavior. It has been
proven that stock prices approximately follow a random walk pattern most of the time,
and psychologists have described a number of ways in which people deal with
randomness. Additionally, market participants may be subject to herd behavior. Perhaps
these psychological mechanisms are the invisible fundamentals of technical analysis.
Below is a list of four common psychological inclinations affecting stock market
participants:
1. The gambler’s fallacy is observed in a casino, where after a long run of red, the roulette
player is inclined to believe that black will be next. In financial markets, the gambler’s
fallacy manifests itself in predicting a trend reversal. If a trend continuation is predicted,
then the behavior is called antiregressive (Wagenaar, 1988).
2. Misperception of regression to the mean disregards the statistical fact that in a long
series of a chance process, extreme values are likely to be followed by less extreme
values (Tversky, 1974). In financial markets, the misperception of regression to the
mean manifests itself in predicting a trend continuation.
3. The anchoring effect shows that most people initially tend to base their estimations or
assessment of information on a first source of reference value (anchor) and
subsequently adapt this to the real value. Empirical research shows that the adjustment
process is regularly cut short and is incomplete. The original value (anchor) is afforded
P. Zielonka / Int. Rev. Financ. Analy. 13 (2004) 217–225 219
too much weight (Goldberg & Nitzsch, 2001; Slovic & Lichtenstein, 1971; Tversky &
Kahneman, 1971). Investors overestimate the importance of a recently quoted or
otherwise significant stock price. This price may then serve as an anchor for further
probability assessments.
4. Herd behavior reflects the immense power of social pressure on individual judgment
(Asch, 1952). Many people accept the perceived authority of others on stock market
valuation and stop trusting their own judgments if the majority of market participants
hold a different opinion.
The present research was designed to determine if technical analysis signals are
associated with some common cognitive biases. A questionnaire was administered to
24 Polish financial analysts, and there were three groups of items within the
questionnaire:
5 regular technical analysis signals representing four common psychological biases;
5 technical-like signals created by the author of the questionnaire, representing the same
four biases: and
5 technical-like signals also created by the author of the questionnaire, not representing
any psychological biases.
Hypothesis 1: A majority of the financial analysts will highly rate signals belonging to the
first and the second groups and will assign no predictive value to the signals from the third
group.
If Hypothesis 1 is supported, then it can be concluded that some technical analysis
signals represent common cognitive biases, and this may be responsible for the great
popularity of technical analysis.
Hypothesis 2: A majority of the financial analysts will agree on the impact of particular
items for future stock prices.
If this hypothesis is supported, then it can be concluded that the analysts are subject to
some psychological biases, inasmuch as they agree on the predictive value of unproven
measures.
2. Method
The study was carried out in Warsaw in January–February 2002. The respondents were
24 financial analysts or dealers employed by banks and Polish capital market institutions.
A convenience sample was used. Each participant was administered a 24-item question-
naire. As noted, there were three groups of items within the questionnaire. Each group
consisted of eight items.
The first group consisted of regular technical analysis signals representing four
common psychological inclinations. Each inclination was represented by two signals.
Usually, one from a pair of signals was a predictor of a stock fall (� ), whereas the other
signal was a predictor of a stock rise (+).
P. Zielonka / Int. Rev. Financ. Analy. 13 (2004) 217–225220
(1) Gambler’s fallacy
– Rising WIG1 index breaks its main trend line (� );
– After a big rise, WIG index creates a ‘‘head and shoulders’’ formation (� ).
Both signals above are believed by technicians to precede a rapid trend reversal.
(2) Misperception of the regression to the mean
– Falling WIG index breaks successive levels of support (� );
– Rising WIG index confirms its main trend line (+).
The two signals above are regarded as signifying a continuation of an existing trend.
(3) Anchoring effect
– Rising WIG index repeatedly bounces from a resistance level—rising WIG fails to
surpass some barrier (� );
– Rising WIG index breaks strong psychological barriers (+).
According to technical analysts, both signals indicate that some stock price (or market
index) levels are especially significant for making predictions.
(4) Herd behavior
– A WIG index fall accompanied with a decreasing popularity of stock market (� );
– Dramatic rise in the value of a buy order (+).
Both signals above are believed to represent the behavior of a large group of investors,
which, according to technicians, have a strong impact on the future movements of
stock prices.
The second group consisted of technical-like signals created by the author of the
questionnaire, representing the same four psychological inclinations (two items for each
inclination).
(1) Gambler’s fallacy
– Rising WIG fails to surpass 4/5 of the last peak (� );
– After the index falls, a big overbalance of demand occurs (+).
(2) Anchoring effect
– Falling WIG index breaks October effect peak (� );
– Rising WIG breaks local peak of January effect (+).
(3) Misperception of the regression to the mean
– The fall of WIG index accompanied by a government crisis (� );
– WIG index’s increase accompanied by a drop of the unemployment rate (� ).
(4) Herd behavior
– Diminishing rate of a WIG fall accompanied by a vanishing trade turnover (+);
– Over 2 months’ rise of WIG accompanied by a high trading volume (+).
The third group consisted of technical-like signals (created by the author of the
questionnaire) that did not represent any psychological inclinations.
1 WIG is the main Polish stock exchange index—a total return index encompassing all shares listed on the
main market, calculated since 1991.
P. Zielonka / Int. Rev. Financ. Analy. 13 (2004) 217–225 221
– Drop of chemical companies’ prices;
– Horizontal, typically sinusoidal, WIG index movement;
– Rising WIG index creates longer and longer horizontal shelves;
– A fan formation support line moves upward;
– Diminishing dynamics of price rise in textile branch;
– An alternate large and small daily trade volume;
– Second moving average convergence/divergence (MACD)2 derivative goes negative;
– WIG index creates horizontal small amplitude sinusoid curve.
The technical analysis signals were assigned to particular psychological inclinations by
the author of the research. In a few cases, it is not clear if a technical signal represents only
one bias. For instance, it is not clear whether the item ‘‘Rising WIG index breaks strong
psychological barriers’’ represents the anchoring effect or herd behavior. However, it should
be emphasized that the problem of proper assignment is not crucial for the results of this
research. What is important is the distinction between the items representing any psycho-
logical inclination (the first and the second group) and the empty items (the third group).
The cover page of the questionnaire stated that the survey was designed to better
understand the opinions of experts on the implementation of technical analysis. This
remark allowed the respondents to feel more like experts, whose opinion was needed for
some further research rather than merely the persons to be examined. The respondents
were assured of confidentiality. Demographic information was collected on the respond-
ents’ age, educational level, sex, and employment status.
The respondents to the questionnaire were asked to assign a score of � 3, � 2, � 1, 0,
+ 1, + 2, or + 3 to each item, depending on how they estimated the impact of the item for
the future (within a few weeks) behavior of stock prices (the WIG index). The analysts
were asked what market movement they would expect if a certain technical signal
occurred, with a higher score indicating a stronger positive impact of the item, resulting
in a stock price rise. Thus, ‘‘� 3’’ meant that the item was expected to evoke a strong
stock price drop, while ‘‘ + 3’’ meant a strong rise. Zero meant that the item was regarded
by the participant as a neutral measure, having no effect on stock price changes.
All 24 respondents filled out the questionnaires. On average, less than 3% of the
questionnaire items remained unscored. No regularity was noticed among unscored items.
Statistical analysis of the scores was done with the use of the StatSoft computer program
Statistica 5.5.
3. Results
To examine which of the 24 items received the highest, lowest, and median marks,
indicating respondents’ opinions on its power to affect stock prices, a cluster analysis was
carried out (Hartigan, 1975) in two stages. A horizontal hierarchical tree plot and a graph
2 A technical indicator that signals overbought and oversold conditions by measuring the intensity of public
sentiment. TheMACD indicator is created by calculating the difference between two exponential moving averages.
A third exponential moving average is plotted on top of the MACD as a trigger line to provide buy and sell signals.
P. Zielonka / Int. Rev. Financ. Analy. 13 (2004) 217–225222
of the amalgamation schedule (Ward’s method amalgamation rule, Euclidean distances)
indicated three main clusters. K-means clustering (constant intervals) divided all 24 items
into the three clusters, as presented in Fig. 1.
Cluster 1 consists of eight signals, which, according to respondents, have little or no
impact on future stock prices. All eight in this cluster were technical-like signals created
by the author of the questionnaire, not representing any psychological inclinations.
Cluster 2 consists of eight items rated as highly important predictors of stock price
declines. All the technical and technical-like questionnaire items representing cognitive
biases associated with a stock price decline are included in this cluster.
Cluster 3 consists of eight signals rated by questionnaire respondents as highly important
predictors of stock price rises. All the technical and technical-like questionnaire items
representing cognitive biases associated with stock price rises are included in this cluster.
These results confirm Hypothesis 1; however, an ANOVA analysis of the mean absolute
values of scores for each of the three types of signals was used to precisely verify this.
The post hoc T-Tukey (HSD) test showed that the respondents discriminated between
the technical-like signals that trigger cognitive biases (TLS-P) and the technical-like
signals which did not represent any biases (TLS-N): P < .001. The respondents also
discriminated between the real technical signals (TA) and technical-like signals that
represented cognitive biases (TA-L biases): P < .001.
Fig. 1. The plot of the means for each of the three clusters. All eight technical-like signals not based on
psychological traps belong to Cluster 1. All the technical and technical-like signals based on psychological traps
representing a stock fall are included into Cluster 2. All the technical and technical-like signals based on
psychological traps representing a stock rise are included into Cluster 3.
Fig. 2. Mean rates of the predictive value (absolute values of the scores) of the three types of signals. TS—
technical signals; TLS-P—technical-like signals based on psychological traps; TLS-N—technical-like signals not
based on psychological traps. ANOVA F=163.78, p<0.0000.
P. Zielonka / Int. Rev. Financ. Analy. 13 (2004) 217–225 223
As Fig. 2 shows, the respondents assigned the highest predictive value to the real
technical signals (TS) and slightly lower values to the technical-like signals that
represented cognitive biases (TLS-P). At the same time, they assigned close to zero
predictive value to technical-like signals that did not represent any cognitive bias (TLS-N).
3.1. Interjudge agreement
The standard deviations of the scores assigned to each questionnaire item were between
0.5 and 1.3, which can be considered relatively small values. The small standard
deviations confirm that the analysts were in general agreement about the predictive value
of different signals of unproven efficiency. The likely explanation for the general
agreement among analysts is that they are subject to some psychological biases typical
for all humans. The results of this research confirm Hypothesis 2. The demographic
features of the respondents did not correlate in any way with their responses.
4. Discussion
In spite of much evidence of its low or zero efficiency for stock market forecasting,
technical analysis remains one of the most popular analytical tools among practi-
tioners. Its popularity may have been derived partly from habitual behavior. However,
P. Zielonka / Int. Rev. Financ. Analy. 13 (2004) 217–225224
habit alone cannot account for the popularity of technical analysis, considering its
complete lack of efficiency. Thus, there must be a more fundamental reason for this
persistence.
The present research provides an explanation for the popularity of technical
analysis. The popularity may well be caused by the fact that technical signals
represent typical cognitive biases such as the gambler’s fallacy. Financial analysts
firmly discriminated between two types of technical signals—those based on typical
cognitive biases and empty signals that sound like technical analysis but are without
any connotation of psychological inclinations. Both the ‘‘official’’ technical signals
and the technical-like signals created by the author and representing cognitive biases
were assigned a high predictive value by the financial analysts. On the other hand,
the empty technical-like signals created by the author and that did not correspond to
any cognitive inclinations were assigned a low predictive value. The official technical
analysis signals were scored significantly higher than technical-like signals represent-
ing cognitive biases. Presumably, this difference can be explained by the fact that
financial analysts are familiar with the official technical signals and unfamiliar with
technical-like signals created by the author. However, even new technical-like signals
which represent cognitive biases were scored fairly highly. Thus, one can create
relatively easily one’s own technical signals which will be perceived by technical
analysts as good predictors of stock market behavior. The only necessary condition is
that these new technical-like signals have to correspond to common cognitive
inclinations.
The believers in technical analysis for almost two centuries have discovered various
patterns on stock charts and have formulated their predictions accordingly. It appears as if
the patterns which have survived correspond to psychological heuristics. In particular,
those patterns which are the most popular among practitioners trigger the most persistent
cognitive inclinations, such as the gambler’s fallacy, anchoring effect, or the misperception
of the regression to the mean. This is in line with many findings (cf., e.g., Stephan & Kiell,
2000) showing that not only lay people but also assorted experts are subject to judgmental
heuristics and biases.
References
Alexander, S. S. (1961). Price movements in speculative markets: Trends or random walk. Industrial Manage-
ment Review, 2(2), 2–26.
Asch, S. (1952). Social psychology. New York: Prentice-Hall.
Cootner, P. H. (1964). The random character of stock market. Cambridge: The MIT Press.
Fama, E., & Blume, M. (1966). Filter rules and stock market trading. Journal of Business, 39, 226–241.
Goldberg, J., & Nitzsch, R. (2001). Behavioral finance. Baffins Lane, Chichester: Wiley.
Hartigan, J. A. (1975). Clustering algorithms. New York: Wiley.
Janiec, M. (2000). Efficiency analysis of technical analysis tools (unpublished manuscript). Cracow: Cracow
University of Economics.
Jensen, M., & Bennington, G. (1970). Random walks and technical theories: Some additional evidence. Journal
of Finance, 2(25), 469–482.
Lo, A. W., & MacKinlay, A. C. (1999). A non-random walk down Wall Street. New Jersey: Princeton University
Press.
P. Zielonka / Int. Rev. Financ. Analy. 13 (2004) 217–225 225
Malkiel, B. G. (1996). A random walk down Wall Street. New York: WW Norton and Company.
Roberts, H. V. (1959). Stock market ‘‘patterns’’ and financial analysis: Methodological suggestions. Journal of
Finance, 1(14), 1–10.
Samuelson, P. A. (1965). Proof that properly anticipated prices fluctuate randomly. Industrial Management
Review, 6, 41–49.
Slovic, P., & Lichtenstein, S. (1971). Comparison of bayesian and regression approaches to the study of infor-
mation processing in judgement. Organizational Behaviour and Human Performance, 6(6), 649–744.
Stephan, E., & Kiell, G. (2000). Decision processes in professional investors: Does expertise moderate judge-
mental biases. IAREP/SABE Proceeding, 416–420.
Tversky, A. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131.
Tversky, A., & Kahneman, D. (1971). Belief in the law of small numbers. Psychology Bulletin, 76(2), 105–110.
Wagenaar, W. (1988). Paradoxes of gambling behaviour. London: Lawrence Erlbaum.
Warneryd, K. -E. (2001). Stock-market psychology. Cheltenham: Edward Elgar.