Searching for Patterns in Financial Information: Part 2

Recently, I discussed the similarities between entropy theory, information theory, and efficient markets theory. Essentially, my equity portfolio strategy is a pattern-seeking machine for use in financial markets.

Because all three phenomena make use of distributions, the concept of uncertainty applies to all of them.

In entropy, uncertainty is seen as the non-uniform distribution of, say, gas molecules in a container. Due to randomly differing energy levels, a gas tends not to be evenly distributed. Physicists came to the conclusion that the use of probabilistic distributions was necessary to accurately to make predictions of the behavior of gases in a container.

Similarly, Shannon's information theory deals with uncertainty. Until one can find sufficient symbol redundancy to ascertain "information" amidst a series of symbols, uncertainty is high. Finding the signal amidst the noise uses probabilistic distributions to ascertain likelihoods of information content, and what that content is.

In the same way, our equity portfolio selection and management process seeks out patterns amidst the noise of financial data. Taken on its face, this torrent of financial information, both market (technical), and company-specific (fundamental), is full of uncertainty. Especially over short periods of time, such as hours or days, the data may seem to contain little in the way of useful patterns. Indeed, in today's markets, with hedge funds accounting for up to 50% of the trading volume on many days, it would seem that rapid-fire trading and asset turnover preclude the search by most investors for patterns beyond a few days in duration.

On the contrary, we carefully sift through a variety of longer-term data which proprietary research has shown contains redundancy. That is, we have found that, when certain predictor data are of a certain pattern, then dependent variables, such as total return, have a high probability of also fitting a certain pattern. This redundancy, while not perfect, seems to provide for a significant amount of probability that the "tails" of distributions of certain data will contain desired information.

While much portfolio management and equity analysis seems to focus on point-estimates of prices and earnings, we prefer to make use of probabilistic distributions of key data in order to increase the chances that the equities we choose will continue to behave in a virtuous pattern, pursuant to our research findings.

As I discussed in my earlier post, the fact that tickers represent real companies, full of real people producing real products and services, increases our confidence that the patterns we seek, and find, are reliable and profitable. Borrowing from the probabilistic approaches to pattern discovery and uncertainty reduction in entropy and information theories, we believe we have developed something as powerful and functional in the management of equity portfolios in finance.