All the News That's Fit to Reprint: Do Investors React to Stale Information?
Summary (3 min read)
I. Empirical Data and Methodology
- This study uses financial news stories about publicly traded US firms in the DJ newswire archive from November 1996 to October 2008 to measure these firms' information environments.
- The DJ newswires are the most widely circulated financial news in the United States for institutional investors.
- Before writing their stories, reporters often obtain facts from non-exclusive sources of information that many investors can access directly.
- Prior to that date, the DJ ticker codes exist mainly for larger firms, which appear to have been selected in a non-random way (Tetlock (2010) ).
- Because the DJ ticker codes in each story determine whether a firm is mentioned, the sample period begins in November 1996 and ends in October 2008, which is the last available month.
II. Using Staleness to Predict Firms' Returns after News Events
- This section presents daily Fama-MacBeth (1973) cross-sectional regressions to investigate whether the staleness of news predicts the extent of return reversal after news.
- These regressions control for many other variables that could predict returns after news.
- The initial results conservatively exclude the return on day t + 1 because using adjacent formation and holding periods may induce bid-ask bounce in returns.
- In robustness checks, I report the results for other time horizons to assess the duration of return predictability and the possible impact of bid-ask bounce.
- To control for postearnings announcement drift and return predictability identified in Pritamani and Singal (2001) and Tetlock (2010) , the main specification includes interaction terms between earnings words and day-t returns (Earn it *AbRet it ) and between log newswires and day-t returns (Msg it *AbRet it ).
Stock market control variables in the main analysis include
- Empirical results in studies cited above, along with studies by Banz (1981) , Ang et al. (2006) , and Gervais, Mingelgrin, and Kaniel (2001) , indicate that these variables forecast high-frequency returns.
- To allow for the possibility that reversals differ for small and big firms, the main specification includes an interaction term between firm size and day-t returns (MktCap i,t-1 *AbRet it ).
- Lastly, an exhaustive specification includes five more interaction terms between day-t returns and control variables: abnormal turnover, log words, prior-week abnormal newswires, prior-week return volatility, and prior-week illiquidity.
- As before, all control variables are standardized by trading day and interaction terms are the product of these standardized variables.
A. Cross-Sectional Regressions of Post-News Returns on News Staleness
- To conserve space, the table shows only the most interesting regression coefficients.
- Table 3 shows that the coefficients on AbRet it *stale1 it and AbRet it *stale2 it are negative and both statistically and economically significant.
- The magnitudes of these two coefficients change by a statistically immaterial amount regardless of which control variables are included in the three specifications: no controls, the main set of controls, or all controls.
- The point estimates in columns one through four in Table 4 show that return reversal on days t + 1 and t + 2 is 1.0 to 1.5 basis points per standard deviation of daily returns with modest t-statistics ranging between 0.9 and 1.9.
- The impact of news staleness on reversal is more than twice as large for small firms (-0.082) but remains economically meaningful for big firms (-0.038).
B. Calendar Time Returns after News Events Sorted by Staleness and Initial Market Reaction
- Based on the regressions in the previous section, the inclusion of several control variables does not materially change the estimated impact of stale information on reversal.
- In general, firms in the sample of news events are considerably larger than the representative publicly traded firm.
- I compute the risk-adjusted returns of each portfolio using a standard time series regression of portfolio returns on four return factors: the market (MKT), size (SMB), and book-to-market (HML) factors proposed in Fama and French (1992 and 1993) ), and the UMD factor based on the momentum anomaly.
- As a fraction of the initial difference in the market's reaction to single-word stale and non-stale news, the equal-and value-weighted difference in alphas on days [1, 5] are equal to 6.3% and 19.8%, respectively.
III. Individual and Institutional Trading around Stale News
- This section examines how individual and institutional investors trade around stale news events.
- A second set of tests analyzes whether the return reversal after stale news depends on the relative intensity of trading activity by these groups of investors.
- These two tests help distinguish rational and behavioral explanations for the observed return reversals after stale news.
A. Trading Behavior of Individuals and Institutions after Stale News Events
- The proprietary data on individual and institutional investor trading activity come from an over-the-counter market maker in Nasdaq stocks.
- This market center began as a trading platform for retail broker-dealers to route their orders, but it also attracts some institutional order.
B. Return Reversals after Stale News Sorted by Relative Intensity of Individual Trading Activity
- This subsection explicitly examines how the return reversal after stale news depends on the relative intensity of individual and institutional trading activity, as measured by the IndAct it variable.
- The results in Table 7 suggest that individuals could play this role.
- This analysis is restricted to February 2003 to December 2007 because this is the period in which market center data are available.
- The reader should interpret these results with some caution because many traders, excluding market makers at broker dealers, do not have realtime access to data on the relative trading intensity of individuals and institutions.
- The significantly negative coefficients on AbRet it *stale1 it and AbRet it *stale2 it in columns one and three in Table 8 show that the return reversal after stale news is quite strong in stocks with above-median individual trading activity.
IV. Concluding Thoughts
- This paper presents evidence consistent with the hypothesis that individual investors overreact to stale information about publicly traded firms.
- The return reversal after stale news is distinct from previously documented reversals and momentum, such as the weekly return reversal, volume-induced return reversal, and postearnings announcement drift.
- Yet one could also explore whether there is any positive correlation in the market reactions to successive news events-i.e., and .
- The results here suggest that one role of financial media is to transmit stale news to a subset of investors who unwittingly make prices less efficient in the short run.
- Methodologically, this paper shows that one can use the content of news stories to quantify the staleness of information transmitted by news providers.
Did you find this useful? Give us your feedback
Citations
639 citations
544 citations
310 citations
274 citations
272 citations
References
24,874 citations
18,117 citations
14,517 citations
14,171 citations
10,806 citations
Related Papers (5)
Frequently Asked Questions (7)
Q2. How do you calculate the risk-adjusted returns of each portfolio?
I compute the risk-adjusted returns of each portfolio using a standard time series regression of portfolio returns on four return factors: the market (MKT), size (SMB), and book-to-market (HML) factors proposed in Fama and French (1992 and 1993)), and the UMD factor based on the momentum anomaly.
Q3. Why do rational investors trade intensely on the signal that will soon be re-released?
Because rational investors anticipate that irrational investors will overreact, rational investors trade intensely on the signal that will soon be re-released to irrational investors.
Q4. What is the impact of news staleness on reversal?
The impact of news staleness on reversal is more than twice as large for small firms (-0.082) but remains economically meaningful for big firms (-0.038).
Q5. What is the impact of stale news on return reversals?
The fact that reversal after stale news occurs even within the group of stories that focuses on firms’ earnings indicates that the textual staleness of news is not merely a proxy for news that is irrelevant for valuation.
Q6. What can be done to test the stale information hypothesis?
Researchers can explore other dimensions of information content, such as the evolution of particular news topics over time, using similarity measures analogous to staleness.
Q7. What is the difference between the two-word staleness and the big-firm ?
These columns reveal that single-word staleness is associated with a reduction in return volatility that is over five times stronger in small firms (-0.203) than in big firms (-0.036).