Finding patterns in random data
by Nelson Weiderman, Ph.D.
Since the advent of the personal computer, a great deal of time has been devoted to finding tradeable
patterns in stock and commodity data. The idea is that patterns found in historical data are likely to repeat
themselves, indicating opportunities for successful trading (see "Trading close-to-close patterns" in this
issue). Unfortunately, not much attention has been given to the statistical significance of historical
patterns and whether they are different from patterns observed in random data.
The purpose of this article is to point out how random data may appear to be non-random. Once a few
basic principles of probability are understood, it may be possible for the non-mathematically inclined to
apply a few tests to determine whether a pattern is unusual or not.
To demonstrate how random data may appear to be non-random, I simulated two coin tossing
experiments on a computer model. The patterns that emerge in the results show that patterns do occur in
random data and they look very similar to patterns that have been observed as tradeable patterns in
Suppose a coin is tossed four times. There are 16 possible outcomes: HHHH, HTTH, TTHT, etc. If we
have n trials (repetitions), then we would expect that each of the 16 patterns would occur approximately
n/16 times since all of the occurrences are equally likely. This is elementary and understood even by
people who have little understanding of statistics.