Discovering company revenue relations from news: A network approach

Ma, Z., Sheng, O.R.L. & Pant, G., 2009. Discovering company revenue relations from news: A network approach. Decission Support Systems, 47(4), pp.408—414.

This article also draws upon network metrics to solve a binary classification problem (see my previous post), but applies the binary classifier to company revenue relations (higher revenue true/false) rather than to competitor relations (is a competitor true/false). The authors are able to achieve a precision close to 80% for predicting a company's relative revenue (i.e., revenue rank).

Introduction and Related Work

1. the company revenue relation (CRR) is defined as a simple value that indicates whether the target company is "more powerful" (has a higher review) than the source company.
2. Graph-based attributes often reflect certain properties of nodes.
• outdegree - centralitiy
• indegree - prestige, authority (which suggests that a company with a higher indegree (i.e. higher media coverage) within a business sector might be more powerful than a company with a lower one).
3. In contrast to forecast models, this approach does not rely on any prior information on the company's revenues.
4. The relation between news and and stock ratings as well as exchange-rate movements have been thoroughly studied by research such as the one performed by Engel and Ng (1993), Conrad et al. (2002), and Domiguez and Panthaki (2006).

Method

The authors compare the following revenue relations between companies using their CRR measure:

1. Sectoral revenue rank
2. Normalized revenue rank \[ \text{rank}(n_i) = \frac{\text{revenue rank}(n_i)}{ |\text{sector}(n_i)| } \]
3. Revenue share \[ \text{revenue share}(n_j) = \frac{\text{revenue}(n_i)} {\sum_{n_j \in\text{sector}(n_i)} \text{revenue}(n_j)} \]

where $$sector(n_i)$$ refers to all companies in the sector to which company $$n_i$$ belongs. All three measures delivered approximately the same performance.