Modeling Financial Analysts’ Decision Making via the Pragmatics and Semantics of Earnings Calls

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 493–503 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 493 Modeling financial analysts’ decision making via the pragmatics and semantics of earnings calls Katherine A. Keith College of Information and Computer Sciences University of Massachusetts Amherst kkeith@cs.umass.edu Amanda Stent Bloomberg LP astent@bloomberg.net Abstract Every fiscal quarter, companies hold earnings calls in which company executives respond to questions from analysts. After these calls, analysts often change their price target rec- ommendations, which are used in equity re- search reports to help investors make deci- sions. In this paper, we examine analysts’ de- cision making behavior as it pertains to the language content of earnings calls. We iden- tify a set of 20 pragmatic features of ana- lysts’ questions which we correlate with ana- lysts’ pre-call investor recommendations. We also analyze the degree to which semantic and pragmatic features from an earnings call com- plement market data in predicting analysts’ post-call changes in price targets. Our re- sults show that earnings calls are moderately predictive of analysts’ decisions even though these decisions are influenced by a number of other factors including private communication with company executives and market condi- tions. A breakdown of model errors indicates disparate performance on calls from different market sectors. 1 Introduction Financial analysts are key sell-side players in fi- nance who are employed to analyze, interpret, and disseminate financial information (Brown et al., 2015). For the firms they cover, financial analysts regularly release recommendations to buy, hold, or sell the company’s stock, and stock price tar- gets. Financial analysts’ forecasts are of value to investors (Givoly and Lakonishok, 1980) and may be better surrogates for market expectations than forecasts generated by time-series models (Fried and Givoly, 1982). Analysts’ decisions are influenced by market conditions and private communications1, so it is 1Brown et al. (2015) find over half of the 365 analysts impossible to exactly reconstruct their decision making process. However, signals of analysts’ decision making may be obtained by analyzing earnings calls—quarterly live conference calls in which company executives present prepared re- marks (the presentation section) and then selected financial analysts ask questions (the question- answer section). Previous work has shown that earnings calls disclose more information than company filings alone (Frankel et al., 1999) and influence investor sentiment in the short term (Bowen et al., 2002). However, recently com- pany executives and investors have questioned their value (Koller and Darr, 2017; Melloy, 2018). Earnings calls are extremely complex, naturally-occurring examples of discourse that are interesting to study from the perspective of computational linguistics (see Figure 1). In this work, we examine analysts’ decision making in the context of earnings calls in two ways: • Correlating analysts’ question pragmatics with their pre-call judgements: With domain experts, we select a set of 20 pragmatic and discourse features which we extract from the questions of earnings calls. Then we correlate these with analysts’ pre-call judgments and find bullish analysts tend to …

Content truncated to 500 words