数据科学与应用AI：从财报电话会议预测未来公司行动

Predicting future corporateactions from earnings-calllanguage Data Science & Applied AI Vincenzo Pota(iii)+44 (0)20 3134 2097vincenzo.pota@barclays.comBarclays, UK According to our analysis, management's earnings callcomments on potential corporate actionsliftthe odds ofpayout announcements, but only some signals translate intomarket outperformance. We test whether payout language on earnings calls predicts corporate actions and subsequentreturns. Across STOXX 600 and S&P 500 companies from 1 January 2005 to 1 June 2026, forward-looking mentionsliftthe probability of follow-on announcements across all five payoutcategories: •Dividend raises1.9x,dividend cuts3.8x,special dividends9.6x,buyback increases2.0xandbuyback decreases1.6x. •Special dividends are the strongest event signal. A mention raises the probability of anannouncement from 1.2% to 11.6%, the largestliftin our analysis. •Buyback mentions carry the cleanest positive return signal. Buyback-increase mentionsreturn +0.62% over 63 days, versus a -0.15% no-mention baseline. •Special-dividend comments can be a short-term warning if delivery does not follow. Onaverage, firms that mention and deliver a special dividend are roughly flat over 63 days, at-0.30%, while firms that mention but do not subsequently announce one return -1.47%. Are payout discussions leading indicators ofannouncements? Earnings calls are full of forward-looking payout language. Management teams talk aboutdividend capacity, possible increases, repurchase authorisations and excess cash. The questionis whether those comments are useful to investors, or whether they are simply part of the usualcorporate script. Using data from the past 20 years, we test this directly. Barclays Capital Inc. and/or one of itsaffiliatesdoes and seeks to do business with companiescovered in its research reports. As a result, investors should be aware that the firm may have aconflict of interest that couldaffectthe objectivity of this report. Investors should consider thisreport as only a single factor in making their investment decision. This research report has been prepared in whole or in part by equity research analysts basedoutside the US who are not registered/qualified as research analysts with FINRA. (iii)This author is a member of the EMEA Equity Research department who may publish equityand debt research Please see analyst certifications and important disclosures beginning on page 6.Completed: 15-Jul-26, 20:36 GMTReleased: 16-Jul-26, 03:00 GMTRestricted - External For each earnings call, we identify forward-looking mentions of dividend increases, dividendcuts, buyback increases, buyback decreases and special dividends. We then ask whether therelevant corporate action is announced before the next earnings call, and compare the resultwith firms that made no such mention. The aim is simple: measure whether managementlanguage 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 announcementsacross all five categories, although the strength and interpretation of the signal vary materially. •Dividend-increase mentionsliftthe probability of a raise from9.0% to 17.1%, or1.9x.Dividend-cut mentions are much rarer, but more informative: the probability of a cut risesfrom2.6% to 10.0%, a3.8xlift.We think this asymmetry is intuitive: management teamsmay speak freely about potential increases, but tend to be more cautious about discussingcuts. •Special dividends are the cleanest signal. A mention raises the probability of anannouncement from1.2% to 11.6%, equivalent to a9.6xlift.In other words, special-dividend talk remains unusual, but when it appears, it is highly informative. •Buybacks aredifferent.A buyback-increase mentionliftsthe probability of an upsizedauthorisation from2.8% to 5.6%, or2.0x. A buyback-decrease mentionliftsthe probabilityof a downsize from2.8% to 4.4%, or1.6x. The direction is less clean than for dividends:buyback language appears to indicate that the repurchase programme is active or underreview, rather than providing a precise directional signal. The absolute conversion rates are still modest. Evenaftera mention, most firms do notannounce the matched action before the next call. That is partly mechanical: payout decisionsfollow board calendars, annual dividend schedules and capital-allocation windows. The signalis therefore in thelift,not in the raw probability level. Same-day announcements are excluded,so the results measure genuine anticipation rather than confirmation of press releases issuedalongside the call. Predicting an event does not always translate intoalpha We next test whether the signals matter for returns. The strategy buys at the close of the firsttrading dayafterthe call and holds for 5, 21 or 63 trading days. We compute returns in excess ofthe equal weighted universe on that day. W