«Internal and External Attributions by Managers in Earnings Conference Calls by Zhenhua Chen Business Administration Duke University Date:_ Approved: ...»
Newman et al. (2003) find that truth tellers use more singular first-person and thirdperson pronouns than liars. Libby and Rennekamp (2010) find that self-serving attribution bias is related to manager overconfidence. If my attribution measure, IvsE, captures some other construct instead of attribution, it may lack power in predicting internal or external attributions.
Finally, pronoun usage may mistakenly indicate internal or external attributions for benign reasons. For instance, speakers may use fillers such as “I think” and “you know” that have no attribution qualities but would be nonetheless be included as part of the IvsE attribution measure. This may be true in a conference call setting where managers use fillers to “buy time” to react to analysts’ questions. Another example is that a manager may refer to company employees as “they”, which is an internal attribution, but my classification scheme will misclassify it as an external attribution.
Managers, in deflecting praise to others in the organization may refer to “our team,” which would be classified as an internal attribution with the IvsE metric. Such misclassifications and improper identification of internal and external attribution introduce noise into the attribution measure.
5.1 Manual Coding of Management Attributions during Earnings Conference Calls As the first step in the validation process, I manually classify attributions for a subsample of conference call transcripts and then correlate the self-constructed attribution measure with the automated attribution measure, IvsE. I choose 80 earnings conference call transcripts (from 67 unique firms), 40 of which are chosen from the two extreme tails of the ROA distributions, with the remaining 40 observations chosen randomly from the rest of the sample. I choose 40 firms with extreme ROA because I anticipate that managers of firms with extreme positive and negative performance face higher demand for explanations and hence are more likely to provide attributions. I also include 40 random observations in order to observe attributions for a broad cross-section of firms.
For each conference call transcript, I first identify attributions made during the call. To be identified as an attribution, it must be the case that a manager points to specific factors to explain some performance outcome, such as earnings, revenue, or other operational performance. There must also be causal reasoning between the factor and firm performance. I read the entire transcript to search for attributions, and follow framework of Baginski et al. (2004)1 to classify an attribution as internal or external.
Internal attributions refer to factors inside the organization, such as changes in business See Section 3.2 of Baginski et al. (2004).
strategies and management efforts. External attributions refer to events outside the organization, such as increasing competition and regulatory changes. There are many forms of attributions and sometimes an attribution points to both internal and external factors. It is also possible that it is difficult to distinguish whether an attribution is internal or external.
Miscalculation introduces noise and reduces the power of the construct validity test. To mitigate this concern, a graduate student coder and the author independently read the earnings conference call transcripts based on the coding guideline specified above. Coding conflicts were identified, discussed and resolved before conducting the validity test.
Table 3 presents the descriptive statistics for the subsample of earnings conference calls used for construct validity testing. Panel A shows that firms in this sample are smaller and less profitable than the whole population of conference call firms. The mean of ROA is -0.06, compared to 0.01 in table 1. The average firm size is much smaller. The mean of assets is $1.16 billion, compared to $4.9 billion reported in table 1. The mean of the book-to-market ratio is 0.16 compared to 0.48 in table 1, suggesting that there are more growth firms in the two tails of ROA distribution. The table also shows that 33% of the firms report special items not equal to zero, and 58% of them report R&D expenses.
This table reports descriptive statistics of my manual coding sample of 80 quarterly earnings conference calls. Panel A shows firm characteristics. Panel B presents the characteristics of attributions. All continuous variables are winsorized at the 5th and 95th percentiles; ROA is winsorized to the range between -0.5 and 0.5.
Panel B shows that managers provide attributions in 64 of the 80 earnings conference calls analyzed. That is, 80% of my sample firms offer attributions, which is close to the percentage observed in prior research (Baginski et al. 2004). Specifically, Baginski et al. (2004) find that 72.4% of their sample firms provide attributions. In my sample, 28 firms provide attributions in the 40 observations drawn from the extreme tails of ROA distribution, while 36 firms provide attributions in the randomly chosen 40 firms. This differential proportion of attributions between the extreme tails subsample and the random subsample is contrary to expectations. I expect that firms facing extreme profits or losses will face higher demand for explanations and therefore, have more attributions. I examine the 12 firms that do not have attributions in the extreme ROA sample and find that 4 of them are in the positive ROA group and 8 of them are in the negative ROA group. Further investigation suggests that firms with large losses spend a considerable amount of time during earnings conference calls discussing survival and restructuring plans instead of explaining prior quarter performance. I also find that firms are more likely to make internal attributions: 48 observations have internal attributions while only 26 observations have external attributions.
The second table in Panel B provides information at the attribution level. Of the 110 attributions identified, 67 are internal, 33 are external, 5 are both internal and external, and the remaining 5 I am unable to unambiguously label as internal or external. In addition, I document that 95% of the attributions are offered during the presentation sections. This is consistent with the notion that managers understand the demand for attributions from investors and provide explanations in the presentation session in anticipation to satisfy the information demand. I also find that firms are more likely to offer favorable attributions than unfavorable attributions (57 vs. 40), which is consistent with prior theory and empirical evidence that firms tend to disclose favorable news (Dye 2001; Wasley and Wu 2006).
There is variation in managers’ attribution behavior, consistent with both selfserving attribution and leadership theories. For example, Herve Caen, CEO of Interplay Entertainment, makes the following internal attribution for good performance during the firm’s Q4 2003 conference call: “Our operating income for fiscal year 2003 was 1.4 million compared to an operating loss of 12.4 million in 2002. The results show a dramatic improvement in operations, resulting from a successful turnaround, which I led since taking over destiny of Interplay almost three years ago.” In contrast, the following example finds a manager attributing good performance to external factors: “We had a favorable tax rate decrease in the quarter from 36.50 a year ago to 33.7. If you remember we had given an outlook expecting our average effective rate to be about 35.5 but because of favorable state legislation that was enacted at the end of, June 30, really, and some other settlements we actually had a benefit of about $0.025 per share as a result of that reduced tax rate.” (Healthcare Realty Trust Q4 2006 earnings conference call) Managers also exhibit different patterns for attributing poor performance. The
following example reveals a CEO blaming external factors for poor performance:
“Revenues (are) slightly down because of the decline in the U.S. dollar.”2 (Technip Q1 2005 earnings conference call), while the following quote from Ambassadors Groups shows a manager attributing internal factors for bad performance: “We also believe that part of the decline (of revenue) is driven by an unexpected decline in the performance of one of our named databases or sources of names.” (Ambassadors Group Inc. Q3 2007 earnings conference call) Content in the parenthesis is added by the author.
5.2 Internal and External Attributions and Personal Pronouns To further examine the relation between personal pronouns and attribution characteristics, I calculate IvsE for each attribution that I can identify as internal or external (100 out of the total 110 attributions). I then examine the association between the automated pronoun based attribution measure, IvsE, and the attributions manually coded internal or external.
Panel A of Table 4 presents the univariate analysis results. The mean of IvsE for internal attributions is 1.70, and the mean of IvsE for external attributions is 0.89. The two groups are significantly different with a t-statistic of 4.70. To assess explanatory power, estimate a simple logistic model with IvsE as the only independent variable. The dependent variable equals 1 if an attribution was manually coded as internal and 0 if an attribution was manually coded as external. The results are presented in Panel B. The sensitivity of the model is 88.06%, as evidenced by the correct classification of 59 out of the 67 internal attributions. The specificity of the model is 45.45%, as evidenced by the model classifying 15 attributions correctly out of the 33 external attributions.3 Collectively, the model correctly classifies 74% of the attributions. The area under the Receiver Operating Characteristic (ROC) curve for my model is 0.740, which suggests This suggests that the pronoun based measure does a poor job capturing external attributions. It is likely due to the fact that managers refer to specific firms and economic factors for external attributions. A possible way to sharpen the measure is to append a dictionary that includes certain economic and environmental words.
that the automated IvsE measure provides acceptable discrimination of internal and external attributions. 4 The results show that the pronoun based measure is better at capturing internal attributions than external attributions (i.e., the measurement error is greater for external attributions than for internal attributions). This is likely due to the fact that managers refer to specific firm and macroeconomic factors for external attributions and these
conclude from this section that the IvsE variable reasonably captures internal and external attributions.
Hosmer and Lemeshow (2000) indicate that areas under ROC curves equal to 0.50 suggest the model is no better than chance, models with areas between 0.70 and 0.80 provide acceptable discrimination and models with areas between 0.80 and 0.90 provide excellent discrimination.
5 An alternative method is to incorporate these macroeconomic factors (for example: exchange or interest rates, tax policies, regulatory actions, etc.) into the attribution measure. In this paper, my objective is to construct a parsimonious measure without subjective modification.
Table 4: Personal Pronouns and CEO Attributions
This table presents the prediction of attribution types using personal pronouns. Panel A shows results from the univariate analysis; Panel B shows results from a logistic model with IvsE as the independent variable.
6. Firm Performance and Managers’ Attributions during Earnings Conference Calls In this section I use my measure to investigate how CEOs attribute firm performance. The main research question is to examine whether self-serving bias or leadership is more descriptive of managers’ attributions. I regress IvsE on firm performance and control for other variables that affect managers’ attributions by
estimating the following ordinary least squares model:
Both the measurement of each independent variable and the expected association with IvsE are discussed below.
1. Firm performance. I use ROA as the measure of firm performance. Prior research on earnings conference calls also uses performance measures such as stock returns (Matsumoto et al. 2011) and unexpected earnings (Frankel et al. 2010). Stock returns also incorporate information that is not conveyed by earnings, and prior research finds that only a small portion of stock returns can be explained by earnings (Lev 1989). Unexpected earnings are affected by market expectations, and managers may conduct expectation management to walk down market expectations in order to beat the earnings target (Bartov and Cohen 2009; Richardson et al. 2004). ROA is not impacted by market movements or earnings expectations. Moreover, I find in my manual coding of attributions in section 4 that most explanations provided by managers are about earnings or earnings components rather than stock returns or earnings surprises. Therefore, I use ROA to proxy for firm performance in this paper.
Under attribution theory, I expect to be positive (i.e., better performance leads to more internal attributions relative to external attributions). If the leadership effect dominates in the sample, I expect to be negative.
2. Proprietary cost. Verrecchia (1983) argues that a manager may not disclose valuable information to investors because this information may be used by other parties to reduce the firm’s competitive advantage. That is, if a firm has higher proprietary cost, the manager is less likely to talk about internal factors that affect firm performance. Hence, in equation (1) I include an indicator variable, DRND, which is equal to 1 if a firm has R&D expenses during the past three years. Proprietary cost theory predicts the coefficient on DRND to be negative.
3. Special items. Baginski et al. (2004) find that firms mentioning one-time charges in their management forecasts are more likely to provide internal attributions and less likely to provide external attributions. Therefore, I include an indicator variable, DSI, which takes the value of one if a firm reports a non-zero value for special items during the fiscal quarter, and zero otherwise. I expect the coefficient of DSI to be positive.