Predicting future corporate actions from earnings-call language

研报英文原文证据摘录

Predicting future corporate actions from earnings-call language

Barclays | Data Science & Applied AI

For each earnings call, we identify forward-looking mentions of dividend increases, dividend

cuts, buyback increases, buyback decreases and special dividends. We then ask whether the

relevant corporate action is announced before the next earnings call, and compare the result

with firms that made no such mention. The aim is simple: measure whether management

language is a leading indicator of payout action, and whether that signal has return content.

Mentions raise the probability of subsequent announcements

Results are shown in Figure 1: mentions raise the probability of subsequent announcements

across all five categories, although the strength and interpretation of the signal vary materially.

FIGURE 1. How Often Payout Mentions Convert into Announcements

Mention on Call Predicted event n mentions P(event | mention) Base rate Lift [95% CI]

Dividend increase Dividend raise 8,915 17.10% 9.00% 1.90 [1.80-2.00]

Dividend decrease Dividend cut 780 10.00% 2.60% 3.80 [3.06-4.71]

Special dividend Special div. announced 1,077 11.60% 1.20% 9.59 [8.03-11.46]

Buyback increase Authorization upsized 9,521 5.60% 2.80% 2.02 [1.83-2.22]

Buyback decrease Authorization downsized 1,400 4.40% 2.80% 1.58 [1.23-2.02]

Base rate is the probability of the event before the next earnings call when no relevant mention is detected. Lift is defined as the event probability after a mention divided by

the no-mention base rate. Confidence intervals are estimated using the Katz log method and treat company calls as independent observations.

Source: Barclays Research, LSEG Data & Analytics, Bloomberg, S&P Global

• Dividend-increase mentions lift the probability of a raise from 9.0% to 17.1%, or 1.9x.