Are Supply Chain Relationships the Causation for Correlation?

Everyone agrees that understanding correlation is important for risk-aware portfolio management. However, working with correlation is difficult for a variety of reasons, and the trend in quantitative portfolio management is to reduce the dimensionality of correlation and thus eliminate correlation between entities. Viewing correlation from a network perspective has advantages, but, because everything is correlated with everything else, doesn’t help with the dimensionality. Based on FactSet’s Revere datasets of inter-company relationships, a different perspective on correlation is possible.

Correlation is important in risk-aware portfolio management. However, since correlation is between companies, the number of unique correlations grows as (N^2-N)/2. Here, we will focus on the 30 companies in the S&P U.S. Energy sector, which produce 435 unique correlations. 

The current approach to reducing dimensionality is through the use of factor models. Though a variety of approaches exist, the goal is to discover a set of independent factors that are less than the number of correlations, and as a result, drive the remaining idiosyncratic return correlations to zero. The ultimate expression of this method is principal components analysis, a mathematical technique that guarantees independent factors and zero correlations, at the price of potentially unintuitive factors.

The New Approach

A “new” technique is to apply the mathematics on network analysis. Obviously, network analysis has many features, but again, one of its drawbacks in investments is that everything is correlated with everything, so there is no reduction in dimensionality. However, network analysis has several ways of highlighting what is important in a network. A network can be described as a set of: 

• Nodes, which represent companies
• Arcs, which indicates the existence of a relationship between companies and
• Degrees, which are the number of arcs at a specific node

From these features, it is possible to describe optimal paths between nodes in a network, among other statistics.

Using FactSet's Revere Supply Chain Relationships data, we can get visibility into the operating performance of a company via its supply chain by highlighting the networks of a company’s key customers, suppliers, competitors, and strategic partners. Based on the S&P U.S. Energy sector during 2016, there are 30 companies in the sector. The Supply Chain Relationships data noted the below relationship pairs:

• 25 competitors
• 37 customer-suppliers
• 21 partners

Furthermore, there are 251 indirect relationship pairs that were possible by stepping through up to four like-aligned relationships. Finally, there were 102 relationship pairs that were unconnected by a Revere relationship. Of course, as noted earlier, there are a total of 435 potential relationship pairs. Even simply sorting into these groups highlights differences in levels and ranges of correlation: 

 
Based on several network attributes it is possible to estimate the correlation between stock returns through regression. The details are less important than the implications, which are summarized in the table below:

The author asserts (without proof), that because these correlations are based on supply chain relations, they should be more stable in times of market turmoil. Thus, while factor models are valuable in determining relationship of stocks to the market, there exists information in the certain supply chain data that can explain the correlation structure between equity returns of companies. Valuable information for the risk-aware portfolio manager.