By Simons, Howard L.
Futures (Cedar Falls, IA) , Vol. 27, No. 9
There remain frontiers to conquer in the inexact science of pattern recognition. But the blurring of closes and opens as the industry enters an era of 24-hour electronic trading may render such work obsolete if it's done before its time.
Past performance does not predict future results." Future archaeologists will puzzle over this incantation, which they'll discover in many apparently unread - but nevertheless expensively printed - financial documents and doubtless will conclude it was the misguided mantra of some bizarre sect. Otherwise, why would it be ignored so often in daily life? And for good reason: If past performance does not predict future results, then just what does?
Human behavior common across cultures and instruments, such as eagerness to take small profits and distaste for small losses, drives markets. It does not matter whether you are an American trading soybeans, an Italian trading Eurolira or a Japanese trading silk cocoons, the behavior of traders should be the same in all markets.
One corollary to this postulate is thai trading patterns are independent of the level of technology used in a market: It should not matter whether we are using clay tablets, rice paper, an abacus or the Internet, the resulting footprints of a market are the same. We can read a cotton price chart from the Civil War, stock price charts from the 1920s and grain price charts from the early 1970s on the same basis. A second corollary is fractal scalability - the comparability of five-minute bar charts to daily bar charts to weekly bar charts, for example.
Pattern recognition, for better or worse, is part of our socialization as traders. We look at charts and see pennants, double tops, spikes, flags, and we learn to associate these patterns with an underlying economic story. One modification of a standard bar chart to include aspects of intraday structure yields Japanese candlesticks. But as useful as candlesticks are, they are not a mutually exclusive and collectively exhaustive classification system, one that provides a unique label for all days.
The key to developing such a classifier is normalization of the major identifiers of a day's structure - its open, high, low, close and midpoint - by locating them on a stochastic distribution of the day's range. Because there is no specific notation for this concept, a standard relational notation will be used. For example, if one-half the range of the day's stochastic distribution fell within the range between that day's open and close, that day would be designated as "O [greater than] = C." In "Distinguishing days" (above), the first classification would be for a day where the open/close range, the open/midpoint range and the midpoint/close range all exceeded one-half the range of the day's stochastic distribution.
Candlestick examples of each of these eight intraday structures are shown in "Lighting the way" (below). The black bodies represent closes less than the open, the white bodies represent closes greater than the open, and the bars are extensions higher and lower than the maximum and minimum, respectively, of the open and close.
The utility of such an intraday classification lies in its information regarding the relative anxieties of buyers and sellers. For example, a day structure such as No. 2 demonstrates the willingness of buyers to step in and stop a price slide; the low of this day will be recognized as an important support point. A day structure such as No. 3, however, indicates overhead resistance.
In isolation, these observations may or may not have any sort of predictive capability for interday price change or for the next day's intraday structure. The intraday structures need to be placed in the context of price trend; any trader instinctively recognizes the significance between a spike bottom occurring after days of decline and one occurring as the reaction to a news-related development within an uptrend or consolidation. …