Pattern Recognition in Price Chart: A survey

B.G. Malkiel [1973],A random walk down Wall Street, Norton & Company, 2012 Edition

This book is dedicated to long-term investing, to simple strategies that avoid complex derivatives structuring that feed only banks and brokers (according to him). All the arguments run along the concept of random walk on prices, namely, the unpredictability of prices in the short-run, in other words, the absence of any significant autocorrelations in the returns.

This books is meant to be read fast. This is not a boring and detailed academic report. Forget any bibliography reference, the authors relies solely on his experience in investing to advise his readers. No careful, balanced opinions and views. No: this book goes straight to the point.

Very critical on 2 points: Complexity in finance, and Technical Analysis.
The recent editions of this book even include some comment on Lo et al. [2000] and subsequent academicworks on price predictability, which he judges badly.

Overall, from the very first pages of this book, the feeling is overwhelming: the author is arrogant, with a lot of self-esteem

Distinction between investing and speculating –

The Firm Foundation Theory

Each financial shall be grounded on an intrinsic value, like a stock’s value depends on the future stream of dividends.

The Castle-in-the-air Theory

Irrational, frenzy bubble. He spent a lot of time describing speculative crazes. >> Madness of the crowd.

Switch to short-term performance reports modified portfolio managers’ incentives.

“There is no evidence that anyone can generate excess returns by making consistently correct bets against the collective wisdom of the market”.

Opposition between technical versus fundamental analysis – Which factors explain market movements. Are they 90% psychological or conversely 90% fundamentals-driven?

 

Why chartism could work ?

1. Self-fulfilling prophecies
2. Unequal Access to fundamentals information (eventually private, e.g. insider information)
3. Initial under-reaction to initial news release

 

Support Levels: second chance to buy at a good level.
“A support area that holds on successive declines becomes stronger and stronger”.
“Another bullish signals is flashed when a stock finally breaks through a resistance area. In the lexicon of chartists, the former resistance area becomes a support area, and the stock should have no trouble gaining further ground.”

Why chartism could fail?

Chartists would buy only after a trend has been established. Some will eventually try to front-run their own signals, and buy early, with the risk of buying too early and see the signal not being triggered.

His biggest argument against chartism validity is that price could jump to the new equilibrium level instead of smoothly trending to it.

Omrane and Van Oppens [2008] – Summary

Technical analysis can be subdivided into 3 independent research fields according to Béchu et Bertrand [1999]: Chart Analysis and the identification of geometrical patterns, Quantitative Analysis through the use of moving average on prices and related oscillators, and Behavioral Analysis which tries to extract the overall sentiment of market.

Quant Analysis has led to a huge development in the last 2 decades while progressively the two other fields received less interest. One can speculate that it is the ability to easily automatize trading rules based on moving averages, and therefore backtest any of these methods studying past prices dynamics in order to infer future prices movements.

This paper focuses on geometrical patterns – let’s call it charting (or chartist method) – and investigate whether these patterns could lead to some predictability and profitability.

Geometrical patterns :

Overall, a chart pattern is defined by a sequence of local maxima and minima.
The extrema identification is done in 2 steps :
1) Smoothing the price dynamics through a kernel estimator and,
2) Detection of the first detection change of sign.

Which patterns?

Head-and-Shoulders and Inverted Head-of-Shoulders.
Double Top, Double Bottom
Triple Top, Triple Bottom
Rectangle Top, Rectangle Bottom
Broadening Top, Broadening Bottom
Triangle Top, Triangle Bottom

The definition of these methods is highly parametric.
Price direction Predictability is time-dependent. their set-up is analogous to an American Binary Option: Hit the target between a time window.

Interesting point of their method: comparison between real financial series and Monte-Carlo simulated ones (Bootstrapping).
The null hypothesis of the predictability test implies that the percentage of positive predictions with simulated series is not different from those obtained with real series.

Overall conclusion:
Highly parametric methods based on a single (EURUSD) currency pair, in 1992.
Results are high dependent on the parametrization. What can we infer about the strength and robustness of the method.
Absolutely no clue about why these charting methods could be working in real life.

 

Trendlines: Outdated or robust-to-jump tools?

 

Most market participants could find trendlines outdated – after all many sophisticated tools exist now. 
 
Remember that Lo et al. [2000] use trendlines & segments in their PR algos, not because they are useful, because they are still used by the chartists community.
 
Let’s try to review why we selected these tools could be useful for us:
 
1) They are Visual : Combining trendlines allow to express complex ideas in simple CHARTING Forms.
 
2) They are easy to compute. In real-life trading, this can be a plus
 
3) They are non-sensitive to jumps. This point is absolutely crucial for us. Current jumps impact future conditional volatility (or variance). 
 
 
Let’s think a bit about 3), and especially about Bollinger Bands, which allow to measure an envelope around a local trend. The standard deviation of prices enters the definition of this envelope. However, jumps impact significantly any volatility measures.
We now know that jumps are different from the standard volatility process. Jumps can be interpreted, while Volatility can’t. This is our basic assumption. 
Therefore, as soon as we integrate sophisticated jump tests in our algos, we can’t use non-robust to jumps volatility measures. This is true for volatility measures on prices, as well as on returns. 
 
Dealing with volatility on prices, Trendlines offer therefore an interesting alternative. 

 

References:

T.N. Bulkowski [2005], Encyclopedia of chart patterns, Wiley Trading, Second Edition (2005)

P.J. Kaufman [2005], Trading systems and methods, Wiley Trading, Fifth Edition (2013)

A. Lo, H. Mamayski and J. Wang [2000]: “Foundations of technical analysis, computational algorithms, statistical inference and empirical implementation”, The Journal of Finance

W.B. Omrane and H.V. Oppens [2008], “The performance analysis of chart patterns: Monte Carlo simulation and evidence from the euro/dollar foreign exchange market”, in L/ Bauwens, W. Polhmeier and David Veredas, High Frequency Financial Econometrics, Physica-Verlag

T. Béchu and E. Bertrand [1999], L’analyse technique : pratiques et méthodes, Economica

W. Brock, J. Lakonishok and B. LeBaron [1992],”Simple technical trading rules and the stochastic properties of stock returns”, The Journal of Finance

C.J. Neely and P.A Weller [1997], “Is technical analysis in the foreign exchange market profitable, a generic programming approach”, Journal of Financial Quantitative Analysis.

C. Osler [2000],”Support for resistance: technical analysis and intraday exchange rates”, FED of NY Working Paper

K. Chang and C.L. Osler [1998], “Is more always better? Head and shoulders and filter rules in foreign exchange markets”, in E. Acar and S. Satchell, Advanced Trading Rules, Butterworth-Heinemann Finance, Second Edition (2002)

E. Knuth [2011], Trading between the lines – Pattern Recognition and visualization of markets, Bloomberg Press