Are trades based on complex technical patterns, such as head-and-shoulders, rational speculations or noise? In other words, do such patterns reliably indicate opportunities to capture excess returns? In her July 1998 paper entitled “Identifying Noise Traders: The Head-And-Shoulders Pattern in U.S. Equities”, Carol Osler investigates whether head-and-shoulders trading is significant and whether it is profitable. In their August 2000 paper entitled “Foundations of Technical Analysis: Computational Algorithms, Statistical Inference, and Empirical Implementation”, Andrew Lo, Harry Mamaysky and Jiang Wang apply advanced empirical methods (compare with fingerprint identification or face recognition) to evaluate technical analysis patterns such as head-and-shoulders and double-bottoms. These papers conclude that:
In “Identifying Noise Traders: The Head-And-Shoulders Pattern in U.S. Equities”, Carol Osler uses an algorithm model to find head-and-shoulders patterns in daily closing price and volume data for a sample of 100 randomly selected stocks during July 1962 through December 1993. She determines that:
- Head-and-shoulders patterns occur roughly once a year for a typical stock.
- Head-and-shoulders pattern traders temporarily boost volume by an average peak level of about 11% on the day price crosses the pattern “neckline”. Unusual trading activity is positive and statistically significant on the neckline-crossing day and the two subsequent days.
- Head-and-shoulders trading is generally not profitable. The mean profit per round-trip trade on positions held for an average of 10 business days (based on stop loss and trend violation exits), excluding transaction costs, is -0.24%.
- The price effects of head-and-shoulders trading disappear slowly but completely in about two weeks.
In summary, the author determines that head-and-shoulders pattern trading exists but is on average unprofitable, mistakenly interpreting randomness as information.
In “Foundations of Technical Analysis: Computational Algorithms, Statistical Inference, and Empirical Implementation”, the authors define geometric properties for various technical analysis patterns and apply automatic pattern recognition techniques to daily price and volume data for a large number of U.S. stocks from 1962 to 1996. Focusing on head-and-shoulders, inverse head-and-shoulders, broadening tops and bottoms, triangle tops and bottoms, rectangle tops and bottoms, and double tops and bottoms, they find that:
- The most common patterns are double tops and bottoms and head-and-shoulders, both occurring on average roughly once a year per stock. Patterns are roughly evenly distributed through time. For some patterns, frequencies of occurrence differ markedly from those generated by a random walk simulation.
- Patterns occur more frequently for large capitalization stocks than small, and for NYSE/AMEX stocks than for NASDAQ stocks.
- For NYSE/AMEX (NASDAQ) stocks, seven (ten) of the ten patterns considered indicate significantly abnormal subsequent returns.
- For a few patterns, volume trends affect the significance of abnormal returns.
In summary, the authors find that several technical indicators do provide incremental information, especially for NASDAQ stocks, and may have some practical value for investors/traders. They stop short, however, of concluding that technical analysis can produce excess trading profits.
In overall summary, even if pattern-based technical analysis can yield abnormal returns (with transaction costs), success may be critically sensitive to precise pattern definitions and pattern recognition capability.
