Common Analysts: Method for Defining Peer Firms

Abstract

We develop a method for defining groups of peer firms on the basis of joint analyst coverage. Besides industry boundaries, analysts’ coverage choices reflect other aspects of firm relatedness such as business model. We find that the analyst-based method produces substantially more homogeneous groups of firms compared to common industry classifications, and has a number of other desirable properties. The paper has two broader implications. First, it demonstrates the advantages of a self-organizing approach to classification, as opposed to a hierarchical system. Second, it illustrates a new positive information production externality generated by the institution of security market analysis.

Social learning and analyst behavior

“…In other words, if two analysts follow the same firm, typically, about two-thirds of the firms in the rest of their portfolio are different. Importantly, the coefficient values show that there is significant 13 Previous studies show that, besides industry, the composition of analysts' firm portfolios can also reflect other aspects of firm similarity such as geographic proximity (O'Brien and Tan 2015, Jennings, Lee, and Matsumoto 2017), supply chain relationships (Guan, Wong, andZhang, 2010, Luo andNagarajan 2015), and business model (Kaustia and Rantala 2020). Brown, Call, Clement, and Sharp (2015) report that 48% of analysts agree to the notion that "The similarity of the company with other companies you follow" is very important when considering whether to cover a particular firm.…”

“…Additionally, analyst rankings that identify top analysts within the same industry sector provide an additional incentive to study what other analysts are saying in their reports. Performance in these rankings has a significant impact on analyst compensation(Stickel 1992, Michaely and Womack 1999, Hong, Kubik, and Solomon 2000.3 Most analysts specialize in covering related firms in a particular industry or industry group(Michaely and Womack 1999, Mikhail, Walther, and Willis 2004, Boni and Womack, 2006, Merkley, Michaely, and Pacelli, 2017, Kaustia and Rantala 2020. Industry-specific knowledge is also an important input into analysts' earnings forecasts(Piotroski and Roulstone 2004, Kadan, Madureira, Wang, and Zach 2012, Bradley, Gokkaya, and Liu, 2017.Electronic copy available at: https://ssrn.com/abstract=3487103…”

Using Textual Analysis to Diversify Portfolios

“…In their application to predicting stock return correlations, Ibriyamova et al (2017) run a 'horse race' among several similarity measures, alongside the one provided by semantic fingerprinting, and show that the latter has a significantly higher predictive power than the Hoberg & Philips measure (Hoberg & Phillips, 2016). At the same time, the similarity measure based on semantic fingerprinting compares well with other candidates for identifying peer companies, based on similar patterns in "analyst coverage" (Kaustia & Rantala, 2021) or "Internet searches" (Lee, Ma, & Wang, 2015). The latter two methods proved similarly successful in predicting stock return correlations but are constrained by design to fit a smaller set of research questions.…”

Company2Vec — German Company Embeddings Based on Corporate Websites

“…The presented algorithm identifies co-searches of multiple firms by a single person under the assumption that these firms must show a positive relation to each other. More directly, Kaustia&Rantala [16] combine estimations of multiple financial analysts in order to model the relatedness between firms.…”