The supply chain of a company can tell us a lot about its future performance. After all if customers struggle or go out of business, the supplier companies lose business. If the supplier stumbles, then the customer has to source elsewhere, incurring additional cost. Hence we have begun to model these relationships using news analytics.

Why have we started down this path? Well, the statements above are intuitive to the point where we can imagine, without testing, that news about a company’s customers will directly affect the supplier. Take Apple, for example. It uses aluminosilicate glass in its iPad/iPhone screens – manufactured by just a few firms globally. If Apple were to sneeze, those manufacturers would surely get a cold.

So, if one could model these supply chain networks one may be able to understand a key driver of return predictability across linked companies.

We began our research by cooperating with FactSet, the global data supplier, looking at the impact of media attention across the supply chain. Our hypothesis was that abnormal newsflow for one company alters the return predictability across economically-linked companies.

We created “winner” and “loser” portfolios by using 1-month price momentum as an information proxy when grouping our supplier and customer companies. We then created a measure to dynamically decide the “direction of information” (i.e. from customers to suppliers or vice versa) on a per sector basis. Finally, we filtered the winner and loser portfolios by news volume abnormality.

The result? The amount of newsflow magnifies the outperformance or underperformance. That is, suppliers or customers of “winner companies” tend to outperform when abnormal news volume is low around the supplier or customer company, while suppliers or customers of “loser companies” tend to underperform more, when the related company is in the media spotlight.

We were able to generate a Sharpe Ratio of 1.5 with annualized returns of 15.4% using a 1-month investment horizon. The figure below plots the cumulative return profile of our supply-chain model with and without a media attention overlay.

The problem with our study is that the entity relationships, or supply chains, are relatively static. Some of our future work will look at co-referencing in the news. That is, what’s the return correlation between companies most often mentioned in the same news articles? We’re pretty sure there are many ways in which this data could be used to add value, but our first study will look at improving the returns of a short-term reversal strategy using co-referencing data.