Can Twitter Help Predict Firm-Level Earnings and Stock Returns?

Eli Bartov Lucile Faurel Partha Mohanram
Prior research examines how companies exploit Twitter in communicating with investors, how information in tweets by individuals may be used to predict the stock market as a whole, and how Twitter activity relates to the response to earnings news. In this study, we investigate whether analyzing the aggregate opinion in individual tweets about a company’s prospects can predict its earnings and the stock price reaction to them. Our dataset contains 998,495 tweets (covering 34,040 firm-quarters from 3,662 distinct firms) by individuals in the nine-trading-day period leading to firms’ quarterly earnings announcements in the four-year period, January 1, 2009 to December 31, 2012. Using four alternative measures of aggregate opinion in individual tweets, we find that the aggregate opinion successfully predicts the company’s forthcoming quarterly earnings. We also document a positive association between the aggregate opinion and the abnormal stock price reaction to the quarterly earnings announcement. These findings are more pronounced for firms in weaker information environments (small firms, firms with low analyst following and less press coverage), and robust to specifications that consider a variety of control variables. Overall, these findings highlight the importance for financial market participants to consider the aggregate information on Twitter when assessing the future prospects and value of companies.

Анализ возможности предсказывать отчетность компаний и реакцию на нее с помощью постов твиттера. Собирали посты в даты [-10..-2] до квартальной отчетности, по хэштегам названия компаний или тикеры, чистили затем определяли значение поста следующим способом:

  •  прогоняли через  Байесов классификатор (OPI1);
  • определяли негативную окраску с использованием словаря Loughran and McDonald (OPI2);
  • определяли негативную окраску с использованием словаря Harvard IV-4 (OPI3);
  • использовали комбинацию OPI1-OPI3 (OPI4).

Дальше в общем-то определяли взаимосвязь с сюрпризом на отчетности (SUE) и экстра-доходность за [-1..+1] дни. Статистики везде значимые, с SUE: OPI1 t-statistic = 5.56; OPI2: t-statistic = 2.93, OPI3: t-statistic = 2.71; OPI4: t-statistic = 4.07. Для экстра доходности: OPI1 t-statistic =4.04; OPI2: t-statistic = 7.48, OPI3: t-statistic =6.92; OPI4: t-statistic = 9.09. По OPI4 то есть годовая доходность в принципе может быть получена в размере 10-15% аннуализированно. По OPI1 всего ~5-6%