Market reaction to second-hand news: inside the attention-grabbing hypothesis

Abstract

This article investigates whether the market reaction to second-hand information is due to price pressure or information dissemination. We use the perspective of attention grabbing to analyse the market reaction to the dissemination of analysts’ recommendations published in print media. This perspective is able to explain the asymmetric market reaction to ‘buy’ and ‘sell’ advice, which is difficult to rationalize within the price pressure hypothesis. We base our empirical analysis on the content of a weekly column in the most important Italian financial newspaper, which publishes past information and analysts’ recommendations on listed companies. Our findings show asymmetric price and volume reactions on the publication day. Contrary to previous evidence, we document a positive relationship between the number of analysts quoted in the column and the price (volume) increase associated with positive recommendations. Because the weekly columns seem to simply attract investors’ attention, with no additional new information, observing a reaction positively related to the column’s salience (proxied by the number of quoted analysts) is natural. In addition, we find that the market reaction is higher when the order size is lower, i.e., when individual investors’ trades constitute a higher fraction of the total trading activity in the market.

Keywords:

• attention grabbing
• analyst recommendations
• anomalous market reaction
• individual investors
• event study

JEL Classification:

• G14
• D82

Notes

¹ The empirical studies listed in Tables A1 and A2 in the appendix analyse printed columns based on analysts’ explicit or implicit recommendations. Other studies consider the publication of columnist recommendations or rumours (Lee, 1986; Ferreira and Smith, 1999;  Kiymaz, 2002; Muradoğlu and Yazici, 2002; Kerl and Walter, 2007) or TV broadcast analysts’ recommendations (Pari, 1987; Ferreira and Smith, 2003). 

² Kraus and Stoll (1972) and Scholes (1972) illustrate both the PPH and the IDH. In particular, they indicate the PPH as an alternative to the IDH.

³ We underscore again that this is not what the original attention-grabbing phenomenon by Barber and Odean (2008) claims, but it is our hypothesis on what we expect to find in terms of abnormal returns.

⁴ According to the Federation of European Stock Exchanges, individual investors play a greater role in Italy than in other countries. For example, in 2003 (2007), Italian household owned 26.6% (26.6%) of the (Italian) market capitalization. The same statistics for other countries are 14.9% (12.8%) for the United Kingdom, 14.1% (13.3%) for Germany and 8.5% (6.7%) for France.

⁵ The number of analysts quoted in the column is not equal to the total number of analysts covering the company.

⁶ In the case of momentum stocks, a pre-event estimate of the market model generates inflated alpha values that bias the post-event normal return upward, leading to negative abnormal returns in the days after publication.

⁷ Other papers related to the attention-grabbing consequences on prices focus on other types of events. Seasholes and Wu (2007) find that unsophisticated investors are net buyers of stocks that hit their daily upper price limits the day before. Fehle et al. (2005) study the attention-grabbing potential of TV commercials during Super Bowl broadcasts.

⁸ From a theoretical point of view, the AGH is not the only hypothesis to predict a nonsymmetric price reaction to good and bad news, there is also the so-called ‘feedback effect’ proposed by Edmans et al. (2011). Their model assumes that outsiders possess private information about the company’s fundamental value, which is unknown to insiders. For uninformed outsiders, the ‘feedback effect’ creates an incentive to sell the firm’s stock since trading without information is profitable only with sell orders (Goldstein and Guembel, 2008). However, in our case, insiders surely know the second-hand information; thus, we discard this alternative explanation.

⁹ We are grateful to an anonymous reviewer for bringing this point to our attention.

¹⁰ As noted, we are not aware of the selection rule used by the journalist to indicate the ‘stock of the week’. As an anonymous reviewer noted, it is hard to believe that ‘each week a stock is randomly selected by’ the journalist ‘in a pool of quoted Italian companies’. For example, a selection mechanism based on the simple rule ‘pick the good/bad’ is more plausible. Thus, in our analysis, there is a risk of nonrandom data collection, which may bias our inference. Unfortunately, this risk characterizes most of the studies using the event-study methodology. All of our robustness checks (most of which suggested by the reviewer him/herself) described in the following sections (e.g., using the Monday before the publication day as the event day, excluding observations with confounding events, using the four-factor model for calculating normal returns, including past performance as a regressor in the regression analysis, etc.) suggest that even if we cannot completely rule out the risk of nonrandom data collection with the journalist’s selection, we can deem the effect of his data collection procedure weak enough as not to invalidate our results.

¹¹ Up to the merger with MTA, we include all the companies listed on the Expandi in the Standard group (eight observations).

¹² To get the tone of analysts’ recommendations, we calculated the mean, median and modal scores. The results are unaffected by the choice of central tendency measure.

¹³ As a robustness check, we repeated all our analysis separating negative and neutral recommendations. The results are not qualitatively different from those reported in the following.

¹⁴ We also repeat the analyses using the turnover ratio instead of volumes. However, the results remain unaffected.

¹⁵ We perform a time-series empirical test for a potential ‘Monday effect’. We estimate a four-factor model with an additional Monday dummy variable for each firm in our sample in the considered period. Since the Monday dummy is not statistically significant for any of the firms, we reject the hypothesis of ‘Monday effect’ in our data. As a further robustness check, we repeated our event-study (and regression) analysis considering the Monday before (and not after) the publication of the column as an event day and find no significant ARs or AVs.

¹⁶ Article 114 of the Italian Law on Financial Markets prescribes that listed companies, with no delay, have to make inside information publicly available when it has a precise nature and is likely to influence stock prices in a significant manner. Articles 184 and 187-bis of the same law prosecute insider trading.

¹⁷ We did not consider potential confounding effects in the period before Friday because we did not find any abnormal returns in those days.

¹⁸ Even the presence of arbitrageurs does not necessarily imply the disappearance of the phenomenon at the event day. As a matter of fact, to realize a profit, arbitrageurs should short sell the ‘Stock of the Week’ only after the price rise has already occurred.

¹⁹ Ex-ante, we may consider the number of analysts as a proxy for either the salience of the column or the information set available to the market. If the column conveyed some real news, then more information already available on the market, i.e., more analysts following, would result in smaller ARs/AVs. Thus, we should expect a negative coefficient associated with the ‘No. Analysts’ variable in our regression. Instead, because the column seems to attract investors’ attention, with no additional new information, expecting a reaction positively related to the number of quoted analysts is natural.

²⁰ Before commenting on the results, we observe that our specification tests do not reject our models at all conventional levels of significance. We use the Reset general specification test and the Breush–Pagan heteroscedasticity test.

²¹ This variable can be regarded as a proxy for a wide analyst coverage as consensus earnings forecasts are computed when many analysts cover the firm. Therefore, we could expect a negative sign for this variable since wide coverage means larger information set already available to investors.

²² Because several companies in the sample are associated with more than one observation – that is, they are ‘the stock of the week’ more than once – we test for the presence of any firm-specific effect on abnormal returns or abnormal volumes. The result of two LR-tests indicates that this effect is indeed not relevant for abnormal returns, while it is important for abnormal volumes. The sign and significance of the coefficients in the regression of abnormal volumes remain, however, unchanged by the inclusion of these firm-specific dummies with respect to the more parsimonious model presented in the table.

²³ Initially, we included both industry and year dummies in our regression analysis as additional controls. However, both set of dummies were not significant in explaining abnormal returns and volume, while reducing sensibly the number of degree of freedoms. Thus, we eventually decided to exclude them from the reported models.

²⁴ As an anonymous reviewer noted, a mix of effects may contribute to the estimated coefficients. So, in our case, ‘both PPH and AGH effects of some investors, together with no-changes due to IDH effects’, can influence estimated coefficients. Even if our preliminary analysis on abnormal returns suggests that the predominant effect in our application is the AGH, we cannot completely rule out the risk of having misattributed estimated coefficients to AGH and suggest caution in interpreting results.