The Availability Heuristic and Investors' Reaction to Company-Specific Events

Contemporary research documents various psychological aspects of economic decision making. The main goal of our study is to analyze the role of the availability heuristic (CitationTversky and Kahneman [1973, Citation1974]) in financial markets. The availability heuristic refers to people's tendency to determine the likelihood of an event according to the easiness of recalling similar instances and, thus, to overweight current information as opposed to processing all relevant information. We define and test two aspects of the availability heuristic, which we dub outcome and risk-availability. The former deals with the availability of positive and negative investment outcomes and the latter with the availability of financial risk. We test the availability effect on investors' reactions to analyst recommendation revisions. Employing daily market returns as a proxy for outcome availability, we find that positive stock price reactions to recommendation upgrades are stronger when accompanied by positive stock market index returns, and negative stock price reactions to recommendation downgrades are stronger when accompanied by negative stock market index returns. The magnitude of the outcome availability effect is negatively correlated with firms' market capitalization, and positively correlated with stock beta, as well as with historical return volatility. Regarding risk availability, we find that on days of substantial stock market moves, abnormal stock price reactions to upgrades are weaker, and abnormal stock price reactions to downgrades are stronger. Both availability effects remain significant even after controlling for additional company-specific and event-specific factors, including market capitalization, stock beta, historical volatility of stock returns, cumulative excess stock returns over one month preceding the recommendation revision, rating category before the revision, and number of categories changed in the revision. Keywords: Analyst recommendation revisionsAvailability heuristicBehavioral financeOutcome availabilityRisk availability Notes ∗p < 0.10 ∗∗ p < 0.05 ∗∗∗p < 0.01. ∗∗∗p < 0.01. ∗p < 0.10 ∗∗ p < 0.05 ∗∗∗p < 0.01. ∗∗ p < 0.05 ∗∗∗p < 0.01. ∗p < 0.10 ∗∗ p < 0.05 ∗∗∗p < 0.01. ∗p < 0.10 ∗∗ p < 0.05 ∗∗∗p < 0.01. a Dummy variable from5i (from1i) is included in the regression analysis only for upgrades (downgrades). ∗p < 0.10; ∗∗ p < 0.05; ∗∗∗p < 0.01. ∗p < 0.10; ∗∗ p < 0.05; ∗∗∗p < 0.01. ∗p < 0.10; ∗∗p < 0.05; ∗∗∗p < 0.01. a Dummy variable from5 (from1) is included in the regression analysis only for upgrades (downgrades). ∗p < 0.10; ∗∗p < 0.05; ∗∗∗p < 0.01. ∗p < 0.10; ∗∗ p < 0.05; ∗∗∗p < 0.01. ∗p < 0.10; ∗∗ p < 0.05; ∗∗∗p < 0.01. a Dummy variable from5 (from1) is included in the regression analysis only for upgrades (downgrades). ∗p < 0.10; ∗∗p < 0.05; ∗∗∗p < 0.01 ∗p < 0.10; ∗∗ p < 0.05; ∗∗∗p < 0.01. a Dummy variable from5 (from1) is included in the regression analysis only for upgrades (downgrades). 1. That is, we presume that either the factors increasing the availability of gain and loss outcomes are also expressed in positive and negative contemporaneous stock market returns, respectively, or that the market tendency itself increases the availability of the respective outcomes. Both schemes would result in the same prediction with respect to the effect of availability as proxied by contemporaneous market returns. 2. For each recommendation revision, firm's market capitalization is registered as the market value of the firm's stocks outstanding at the end of the fiscal year preceding the revision. 3. Daily historical standard deviations are calculated from the daily stock returns over the year preceding the recommendation revision. 4. The category 1 corresponds to Strong Buy; 2, Buy; 3, Hold; 4, Sell; 5, Strong Sell. 5. We also perform the analysis with Market Adjusted Returns (MAR), that is, return differences from the market index and obtain similar results. These results are available on request from the authors. 6. There were no days with zero market return. 7. Conditioning event-driven ARs on the sign of the previous day's MR does not yield significant differences between the groups (the results are available on request from the authors). Thus, only the contemporaneous direction of change in the value of stock market index affects the availability of company-specific news of the respective direction. 8. Though the potential problem of clustering in event dates within our sample looks mild (see Table 3), we have performed a robustness check for the results reported in Table 5, Table 6, Table 7, Table 8, based on the alternative method of variance estimation as described by Brown and Warner [1985]. We estimated the standard deviations based on portfolio returns. The results we have obtained (available on request from the authors) are qualitatively similar. 9. We have also performed the tests separately for each of the sectors reported in Table 1, as well as for each initial rating category. Our findings, available upon request from the authors, apply to the partitioned analysis as well. 10. The results for day -1 (not reported here and available on request from the authors) are qualitatively similar. 11. Alternatively, for day 0, we have run regression (3) with actual levels of market returns split in two categories, using two variables to account for positive and negative market movements, instead of MR_dum. Coefficient estimates for the negative market returns are positive and highly significant, as they comprise two effects acting in the same direction: the outcome availability effect and the risk availability effect (see description in the sequel). Coefficient estimates for the positive market returns are non-significant, as in this case, the two effects act in the opposite directions. The results are available on request from the authors. 12. Though the potential problem of clustering in event dates within our sample looks mild (see Table 3), we have performed a robustness check for the results reported in Table 9, Table 12, Table 13, Table 15, and Table 17. Because of our sample's characteristics, we could not employ the multistage regression described in Fama and MacBeth [1973]. Thus, for each trading day in our sample, we have randomly chosen one upgrade and one downgrade revision (maximum of two revisions per day, if available, and overall, 510 upgrades and 511 downgrades). The results we have obtained (available on request from the authors) are qualitatively similar. 13. We have also repeated the regression analysis reported in subsection 4.4 on the cumulative event-days ARs. The effect of contemporaneous MRs on stock price reactions to recommendation revisions remains highly significant. The regression results are available on request from the authors. 14. CitationKarpoff [1987] surveys a large amount of empirical studies documenting that the correlation between trading volume and absolute value of stock price change is positive, for example, CitationWesterfield [1977], CitationWood et al. [1985], CitationJain and Joh [1986]. 15. Alternatively, for day 0, we have run regression (3) with actual levels of market returns split in two categories, using two variables to account for positive and negative market movements, instead of MR_dum. Coefficient estimates for the negative market returns are highly significant with their signs corresponding to those of MR_dum in Table 12. These estimates comprise two effects acting in the same direction: the outcome availability effect and the risk availability effect (see description in the sequel). Coefficient estimates for the positive market returns are non-significant, as in this case, the two effects act in the opposite directions. The results are available on request from the authors.