Can Fuzzy Logic Make Technical Analysis 20/20?

One of the most challenging areas in technical analysis is the automatic detection of technical patterns that would be similarly detected by the eyes of experts. In this study, cognitive uncertainty was incorporated in technical analysis by using a fuzzy logic–based approach. The results show that the algorithm can detect subtle differences in a clearly defined pattern. Significant postpattern abnormal returns were found that varied directly with the fuzziness of a pattern. This approach can be valuable for investors as a way to incorporate human cognition into historical trading statistics so as to form future winning strategies. Technical analysis has been practiced for many decades. The effect of using technical analysis, especially in the area of spotting visual technical patterns, however, is controversial. The reason is, in part, that a subtle difference in technical patterns may be apparent and important to experienced traders but not to average investors. One of the most challenging areas in technical analysis, therefore, is the automatic detection of technical patterns that would be similarly detected by the eyes of experts.We incorporated cognitive uncertainty into technical analysis by using a fuzzy logic–based approach. We first used Gaussian kernels to smooth the time series of adjusted stock prices and identify five extrema—that is, the local stationary maximum and minimum values for the price—that met the definition of eight widely used technical pattern templates, such as the head-and-shoulders pattern. We then “fuzzified” the patterns and assigned a membership value from 0 to 1 to each pattern found in the price data. Membership value represented the degree of resemblance to the clearly defined pattern template. A membership value of 0 meant the chart did not resemble the predefined ideal pattern. A membership value of 1 meant that the chart fully resembled it.We then tested whether the occurrence of technical patterns with a certain membership value would signal certain kinds of future returns on the stocks. We selected a random sample of 1,451 U.S. stocks for the 1962–2000 period. We matched each sample company with a control company on the basis of their market capitalizations and previous year's returns. We then compared the returns of two portfolios—one containing the sample companies and the other containing the corresponding matched companies. We calculated the cumulative abnormal returns of sample companies relative to the control companies up to 120 days after the occurrence of technical patterns.Our results show that stocks with certain technical patterns in their price charts can generate abnormal returns after the occurrence of the patterns for up to 120 days. We found the approach does detect subtle differences in the patterns. Some abnormal returns were generated mainly by stocks with trading prices below $2.00. Such stocks are primarily NASDAQ-listed stocks. For stocks with raw prices of $2.00 or above, the postpattern performances are statistically significantly different among stocks with the same pattern but with different membership values. Some postpattern stock performances were even opposite to the direction the pattern would have led investors to expect. For example, stocks with a head-and-shoulders pattern are expected to underperform. Our results showed, however, that stocks with head-and-shoulders membership no higher than 0.7 significantly outperformed the control companies.Thus, the findings suggest that the fuzzy logic approach can be used to detect subtle differences even within a pattern. The findings can also explain why so much controversy surrounds technical analysis: A pattern can produce entirely different subsequent results. The difference within a pattern detected by our fuzzy logic algorithm may not be apparent to average investors, only to certain experts. Thus, only the experienced technical analysts can avoid using a pattern with a low membership value.The company-matched abnormal returns suggest that forming portfolios with high membership value patterns can be profitable. Using our approach, investors can first examine the historical trading statistics, then use the proposed fuzzy logic–based approach to find visual technical patterns with high membership values, and take a trading position largely in stocks with strong pattern confirmation. They can also set their own parameters and develop their own future winning strategies based on the fuzzy membership values.