In their May 2005 draft paper entitled “The Market Impact of Corporate News Stories”, Werner Antweiler and Murray Frank apply computational linguistics to 245,429 Wall Street Journal news stories published during 1973 to 2001 to examine how, and how quickly, stock prices fully reflect 43 different kinds of news. They find that:
- News stories are mostly unpredictable.
- Consistent with the assumption that leaks occur, there is often an abnormal return the day before news is published. On average, pre-event and post-event abnormal returns have opposite signs.
- A day or two after an event, returns start drifting in the opposite direction from the initial jump. In other words, the initial reaction to news is usually an overreaction. “Big” news is especially likely to exhibit reversal of the initial jump. The dominant pattern is that a news story in the Wall Street Journal induces a positive abnormal return and a positive abnormal trading volume, with these abnormal shocks then reversing over then next several weeks. Obviously “bad” news price movements follow the inverse pattern. Return momentum is statistically significant for many days after publication.
- Small firms exhibit greater pre-publication leakage. Large firms exhibit longer reversals (greater momentum).
- NASDAQ stocks show stronger reactions than NYSE stocks. AMEX stocks exhibit large cumulative returns.
- News stories typically have bigger and more prolonged impacts during recessions than during expansions. During expansions, adjustments normally take two or three weeks. During recessions, adjustments are more prolonged but the endpoints unclear.
- The patterns found may be a consequence of the time it takes institutional traders to unwind large portfolio positions.
In summary, the stock market shows significant predictability (rather than perfect efficiency) in its reactions to news.
