Anger, Fear, and Investor’s Information Search Behavior

Abstract

This study investigates the differential effect of anger and fear on investors information search behavior. Based on theories from psychology, I predict that angry investors will seek out less additional information and exert a lower depth of thought than fearful investors after a negative earnings surprise. Additionally, I predict that these differences will be moderated in investors that exhibit higher levels of emotion management ability. Using an experiment, I find that neither anger nor fear had any effect on the number of additional information sources investors access. However, angry investors processed information less deeply than fearful investors. This is evident by significant differences in the amount of time spent reading additional information and the ability to recall details about the information. Finally, high emotion management ability reduces differences in depth of thought for both angry and fearful investors. The results of this study have implications for investors and researchers.

Keywords:

• Emotions
• Emotional intelligence
• Information search
• Investor psychology

Acknowledgement

This paper is based in part on my dissertation completed at Queen’s University. I gratefully acknowledge the invaluable insights and support of my dissertation committee: Pamela Murphy (Chair), Daniel Thornton, Christopher Miners, and Lee Fabrigar. This paper has also benefitted greatly from the comments of Leslie Berger, Mike Durney, Devon Erickson, Jessen Hobson, Rachel Martin, Steve Salterio, James Smith, Angie Takahashi, Michael Welker, and James Zhang. I also appreciate comments from workshop participants at Queen’s University, the 2018 ABO Midyear Meeting, and the 2019 BYU Accounting Research Symposium.

Notes

1 This third prediction is motivated in part by discussion within the industry on the importance of emotional intelligence to financial decision-making. Examples of recent article headlines include “Emotional Intelligence, Not IQ, Is Most Important For Investors” (Duggan, 2016) and “Why Financial Intelligence Is So Emotional: The Emotional Intelligence Aspect of Financial Services” (Miller, 2018).

2 Affect is an umbrella term used to describe differences in valence, i.e., positive versus negative feelings, of both moods and emotions (Lerner et al. 2015).

3 An essential aspect of ATF is the idea that cognitive appraisals and emotions have a recursive relationship, with the possibility that either can be the primary event in a causal chain, ultimately leading to behavior (Han et al. 2007). That is, activation of the responsibility appraisal can lead investors to feel angry, and thus, they will behave in a way consistent with anger (Lazarus 1982). However, it can also be the case that investors feel angry first, and that feeling activates the responsibility appraisal, leading them to behave in a way consistent with anger (Zajonc 1984). While prior experiments have typically manipulated participants' emotions first, and then argued the associated cognitive appraisals lead to a particular outcome (e.g., Lerner and Keltner 2001), I manipulate the cognitive appraisals first to activate the discrete emotions, which are subsequently theorized to influence investor behavior.

4 Three other appraisals are used to differentiate between emotions. The other three appraisals are pleasantness (valence), attentional activity, and anticipated effort. Pleasantness is “the degree to which one feels pleasure versus displeasure,” attentional activity is “the degree to which something draw’s one’s attention versus repels one’s attention,” and anticipated effort is “the degree to which physical or mental exertion seems to be needed versus not needed.” According to Lerner et al. (2015), these other three appraisals show little to no variation between anger and fear. Thus, they are not predictive of when an individual will experience either emotion. Thus I only focus my attention on the cognitive appraisal dimensions that have predictive power. Additionally, it is essential to note that the responsibility appraisal in prior studies is typically associated with differences in self-versus-other attributions. However, in my study participants are perceiving the responsibility of a company versus some other factor. Hence, in my study, the “self” is the participant's perceptions of the firm's responsibility.

5 Within the context of earnings announcements, management often attributes negative earnings performance to external factors and shirks responsibility, consistent with the self-serving attribution bias (Baginski, Hassell, and Hillison 2000; Chance et al. 2015; Kimbrough and Wang 2014). However, management is not the only source of attribution information. Analysts and the media are also critical financial intermediaries that can provide causal attributions for negative earnings announcements (Healy and Palepu 2001) and likely to be more internally focused.

6 The four branches of EI consist of: (1) perceiving emotions, (2) using emotions to facilitate thought, (3) understanding emotions, and (4) managing emotions.

7 The study consisted of 197 participants. However, consistent with other experiments (Hodge, Kennedy, and Maines 2004), some participants did not access any of the additional sources of information available. Because an essential dependent variable measures the participant's ability to recall information they read, participants need to read at least one piece of additional information. For robustness, I ran a second analysis that used the full 197 participants but did not include their recall quiz score. Inferences concerning the number of sources accessed and time spent reading are unchanged. Thus, for ease of writing, all analysis reported in the paper is with the 128 participants.

8 Having a task from which participants earn money was included based on some participants in the pilot study explaining their behavior in terms consistent with the “house money effect.” Thus, changing the experiment from a pure endowment of money to the appearance of effort leading to payment was done to reduce this problem.

9 Theta’s background materials and financial statements were adapted with permission from Chen et al. (2016).

10 To avoid potential order effects, I randomized the order that participants answered the emotion questionnaire and made their updated allocation decisions. Analysis reveals no significant differences in measurement or structural coefficients based on display order.

11 Eutsler and Lang (2015) provide empirical evidence that a fully labeled 7-point scale provides the greatest benefit to researchers. Compared to 5-, 9-, and 11-point scales, a 7-point scale results in (1) the highest between-subjects variance, (2) the lowest central tendency bias (i.e., selecting the middle option), and (3) the highest power.

12 Other characteristics of the attributions could drive observed differences in investor behavior in response to internal versus external attributions, instead of emotions as hypothesized. For example, the implausibility of management’s attributions has detrimental effects on analyst judgments (Barton and Mercer 2005). To test for this, I ask participants how plausible they believe the explanation for the negative earnings surprise was. I find no difference in participants response to internal versus external attributions (internal = 5.43; external = 5.16; p = 0.16). Differences in attributions can lead to perceived differences in the persistence of the negative earnings (Chen et al. 2016). For example, when external events cause earnings surprises, participants may perceive that the negative earnings will persist longer because management has no control. Likewise, internal attributions may be seen as reducing earnings persistence because management is more likely to fix the problem. To see if the attribution manipulation is generating a difference in participant's views of the persistence of the negative earnings, I measure how “likely is it that the lower earnings will persist into the future?” I find no difference in participants response to internal versus external attributions (internal = 4.52; external = 4.52; p = 0.99).

13 In STATA 15, the default estimate for standard errors uses the observed information matrix. For robustness, I also tried estimating the standard error using the expected information matrix as well as using robust standard errors. These alternative estimates did not affect inferences.

14 For robustness, I also used the continuous variable “indicate the extent to which the explanation for Theta's lower earnings was attributed to conditions internal or external to the company.” Results are not affected by the use of this manipulation check variable.

15 For robustness, I included certainty as its own variable in the SEM model. It had no significant paths with any of the other variables and thus is not included in the analysis.

16 The factors related to each latent variable initially came from ex-ante expectations and were verified through exploratory factor analysis. I retained only factors that loaded at 0.65 or higher. The results of this analysis revealed a single factor for the cognitive appraisals (eigenvalue = 3.06; Cronbach’s alpha = 0.93) and two factors for emotions: fear (eigenvalue= 7.55; Cronbach’s alpha = 0.94) and anger (eigenvalue= 2.49; Cronbach’s alpha = 0.93).

17 Dropping # OF SOURCES ACCESSED from the SEM model does not affect the results between the two emotions and the processing variables in the model. Thus, it appears that the effect of emotions on information processing is direct and not mediated by the amount of information that is acquired.