Wednesday, February 10, 2021

Know your taxonomy and solve problems in finance

Taxonomy is critical component for fields like botany and zoology. The same should be said about investments, yet not enough time has been spent on the issue categorization. There are reasons for this lack of focus, but this should change as more data science work is applied to investments. No one says that categorization is easy but engaging in the process of finding groups with similarities will help with building any portfolio and generating diversification. In particular, unsupervised learning through tools like cluster analysis will help develop better thinking in this area. 

Investment management has generally taken a simple approach to taxonomy. For example, indices use size as a mechanism for characterization. There are large cap, mid-cap, and small cap indices. However, these size-dependent indices are often inefficient. The advances in defining other factors create different taxonomies. We now know that size may be a poor way of looking at categorizing stocks.

Another taxonomy is based on industry groups, yet cluster analysis and even simple correlation analysis shows that industry groups may not be a good way to bundle stocks. Firms, unlike plants or animals, can change their business group and many companies have characteristics of multiple industry groups. The same problem can be seen with country groups. The composition of one country index may be very different from another. Firm characteristics like are dynamic and my not be tied to risk.

Even asset class taxonomy may be fuzzy. How do you classify convertibles? What are the categories for fixed income? When does investment grade end and high yield begin if you look beyond ratings categories? A close look at commodities shows that many are not highly correlated.

A taxonomy based on factors beyond size are by their very construction stochastic. Stocks will fall in and out of a value or momentum category, and there are many factors that may have some excess returns for a period of time only to see them disappear. This does not even address the issue of stocks that may fall into multiple factor categories. 

At this point many will shake their head in frustration and state that any taxonomy in finance is flawed and without relevance. The concept of taxonomy is fluid and dynamic in finance, yet this offers an investment opportunity. 

Clustering can be used to find commonality across equities that are not seen through naming conventions. By using clustering as tool for groups, there can be more focused opportunity management. The power of cluster analysis is that there can be a greater depth of understanding than found with simple correlation analysis which is usually calculated is blunt linear calculation. Unsupervised learning techniques can offer a better way of categorizing and forming perhaps a better asset taxonomy. 

No comments: