«Internal and External Attributions by Managers in Earnings Conference Calls by Zhenhua Chen Business Administration Duke University Date:_ Approved: ...»
12.1 Attributions in Causal Reasoning Statements One disadvantage of measuring IvsE using the whole call transcript is that the entire conference call is not dedicated to explaining firm performance. I rely on the assumption that pronoun usage during managers’ attributions is correlated with pronoun usage during the whole conference call. However, pronoun usage in sentences that are not related to attributions adds noise to the measure. One possible way to test the effect of this measurement error is to extract the sentences in which a manager clearly makes attributions and only analyze these attributions statements. This adds to the power to my tests by mitigating measurement error in the attribution measure.
I extract causal reasoning statements (CRS) from earnings conference calls and re-measure the attribution variable IvsE using only these statements. I define causal reasoning statements as all the sentences that contain at least one causal word (e.g., due to, because, lead to, etc.) and at least one word related to performance (e.g., earnings, cash flow, growth, etc.). I delete filler words such as “I believe” that are not related to attributions but may add noise to the attribution measure. The list of causal, performance related and filler words are presented in Appendix B.
After extracting CRS from the earnings conference calls, I repeat the empirical analysis discussed in prior sections using attributions from causal reasoning statements.
Table 12 presents the results of testing the main hypothesis. I find that the results are similar to prior results reported in table 5. The coefficient on ROA is positive and statistically significant, suggesting that managers are more likely to refer to internal factors for better performance. The sign of coefficients all control variables are consistent with prediction except for LENGTH.
This table presents the ordinary least squares regression between management attributions in CRS and firm performance. All the variables are defined in Appendix A. Tstatistics are shown in parentheses with standard errors clustered at both firm and quarter level.
***/**/* means significance at 0.01, 0.05 and 0.10 level, respectively.
12.2 Alternative Method of Identifying Self-serving Attributions The current attribution measure, IvsE, relies on a frequency count of different types of personal pronouns. One disadvantage of frequency count is that it does not consider the context of the sentence. For example, “The operating success was because of my leadership” conveys a different leadership style from “The poor performance reflected our marketing strategy.” This kind of difference will not be captured by a frequency count of personal pronouns resulting in weak statistical power of my tests.
To mitigate this concern, I adopt a different method of identifying managers’ self-serving attributions. For each sentence with a causation word, I define the sentence as a “self-serving” sentence if the sentence has a first-person pronoun and a positive word, or a third-person pronoun and a negative word. Similarly, I define a “leadership” sentence as a sentence with a first-person pronoun and a negative word, or a thirdperson pronoun and a positive word. If a sentence has both first-person and thirdperson pronouns (or both positive and negative words), I use the one with higher frequency.
After classifying each sentence as leadership or self-serving, I construct two measures of self-serving attributions at the conference call level. The first measure, NET_SELFSERVING, is the number of self-serving sentences minus the number of leadership sentences. The second measure, PCT_SELFSERVING, is the percentage of the number of self-serving attribution sentences to the number of all sentences with at least one causation word and a personal pronoun.
To test the main hypothesis of whether self-serving attributions or leadership notions are more descriptive of managerial behavior, I compare the amount of selfserving sentences to the leadership sentences. The descriptive statistics of the two attribution measures are presented in Table 13. The mean and median of NET_SELFSERVING are 1.17 and 1.00, respectively. Untabulated results show that both the mean and the median are significantly different from zero (p 0.001). This suggests that CEOs are self-serving in attribution, consistent with the results in prior sections. The mean and median of PCT_SELFSERVING are 0.62 and 0.67, respectively, indicating that on average 62% of CEOs’ causal sentences are self-serving. Untabulated results show that both the mean and the median are significantly different from 0.5 (p 0.001). The results based on presentation and Q&A sections are also presented in Table 13, and the empirical inferences remain the same.
Table 13: Descriptive Statistics of Self-serving Attribution Sentences
This table presents the descriptive statistics of self-serving attribution sentences, as defined in section 12.2. All the variables are defined in Appendix A.
12.3 Using IvsWE as an Alternative Attributions Measure A manager may take the credit when performance is good. However, when performance is poor, a manager may talk more about the company as a group setting, hiding his or her personal factor in the crowd. Therefore, another form of self-serving attributions is to use singular first-person pronouns (i.e., words like I, me, my, etc.) for positive outcomes and plural first-person pronouns (i.e., words like we, us, our, etc.) for negative outcomes. To investigate this issue empirically, I define a new variable, IvsWE, is defined as ln((1+I)/(1+WE)), which measures the ratio of singular first-person pronouns to plural first-person pronouns.
Results of re-estimating equation (1) with IvsWE are presented in Table 14 Panel A. The coefficients on ROA are negative and statistically significant, suggesting that better (worse) performance results in more frequent reference to plural (singular) firstperson pronouns. This is opposite of the predictions of self-serving attribution bias theory conditional on IvsWE being an alternative proxy for self-serving attribution.
However, empirically, IvsE and IvsWE are negatively correlated (-0.29). This negative association between IvsE and IvsWE is likely due to the fact that CEOs tend to use plural first-person pronouns as shown in Table 2. Thus variation in IvsE is driven heavily by first-person plural pronouns in the numerator of the ratio. When first person plural pronouns are removed from the numerator of IvsE and placed in the denominator of IvsWE, a negative association results.
As an alternative examination, similar to section 12.2, I define a sentence with at least one causation word as “self-serving” if it contains a singular first-person pronoun and a positive word, or a plural first-person pronoun and a negative word. I also define a sentence with at least one causation word as “leadership” if it contains a singular firstperson pronoun and a negative word, or a plural first-person pronoun and a positive word. I aggregate these sentences into conference call level measure.
Results presented in Panel B of Table 14 reveal a mean and the median of NET_SELFSERVING are -1.28 and -1.00, respectively. The mean and the median of PCT_SELFSERVING are 0.37 and 0.33, respectively. These results are consistent with Panel A, suggesting that managers use more plural first-person pronouns as firm performance improves.
Panel A presents the ordinary least squares regression between IvsWE and firm performance. Panel B presents the descriptive statistics of managers’ self-serving attribution sentences. All the variables are defined in Appendix A. T-statistics are shown in parentheses with standard errors clustered at both firm and quarter level. ***/**/* means significance at 0.01, 0.05 and 0.10 level, respectively.
12.4 Additional Robustness Checks In this section I conduct additional robustness checks. I include different ROA definition (use net income instead of operating earnings) in equation (1) and find similar results (not tabulated).
To observe the attribution behavior for a specific CEO, I examine the attribution pattern across positive and negative ROA within a CEO’s tenure in the firm. I require a CEO to appear in at least 16 earnings conference calls from 2002-2007. I have the required time series data for 360 CEOs. On average, CEOs use more first-person pronouns when firms generate profits and more third-person pronouns when firms suffer losses. The mean value of IvsE difference between profits and losses is 0.017, and paired t-statistics suggests that the difference is statistically different from 0 (p=0.04). Of these CEOs, 189 CEOs are more likely to refer to internal factors for profits and external factors for losses. 171 CEOs are more likely to refer to internal factors for losses and external factors for profits.
13. Conclusions Using a large sample of earnings conference calls, I document that managers’ attributions tend to be self-serving when they communicate with investors and financial analysts. I also find that the market reacts negatively to managers’ internal attribution. I argue that the negative market reaction is rational because I am able to document a negative relation between internal attributions and subsequent earnings performance.
My results are robust to alternative methods of identifying CEO attributions.
The paper improves the power and generalizability of prior studies that rely on content analysis. My paper adopts a pronoun-based attribution measure that reduces the cost and bias induced by human coding. However, pronoun-based attribution measure may suffer from its own set of measurement error, especially with respect to external attributions. More refined attribution measures may decrease the classification errors in external attributions and provide additional insights on the determinants and consequences of external attributions.
Although I find that self serving attributions are punished by investors through a negative market reaction, I do not find evidence that CEOs gain other types of economic benefits (e.g. greater compensation) from self-serving attributions. Therefore this paper does not support the motivational explanation for self-serving attributions. Rather, it appears that CEOs on average suffer from self-serving attribution bias.
Appendix A: Variable Description
Singular first-person pronouns relative to plural first-person pronouns IvsWE calculated as ln((1+I)/(1+WE)) Appendix B: Managers’ Attributions from Causal Reasoning Statements I extract causal reasoning statements (CRS) from earnings conference calls from CEO talks in earnings conference calls. To make sure the sentence is about a CEO explain firm performance, I require that a sentence has to include at least one causal word and at least one performance related words to be classified as a CRS.
The causal words and performance related words are presented at the end of this appendix. The causal words and performance related words are from human reading of earnings conference calls. Although the author attempts to be thorough, the lists may not include all the causal and performance words. I do not include “so” in the causal words because “so” may be used in many circumstances but do not necessarily have causal inferences (for example, “or so” and “and so on”.) Using this algorithm to search for CRS lead to both type I and type II errors. It is possible some causal reasoning sentences are not captured by this searching. It is also likely some flagged CRS are not causal explanation of company performance. However, I expect this algorithm to improve the sharpness of the attribution measures.
To calculate the attribution measure (IvsE) in CRS, I also delete filler words that have first-person pronouns (such as “I think”) in the CRS. This reduces the noise in pronoun based attribution. Fillers words like “you know” do not affect the measure and thus are not treated.
Causal Words due to, due in part to, due mainly to, due primarily to, reason, reasons, cuz, because,because of, thus, therefore, as a result, a result of, resulting from, resulted in, function of, driven by, driven primarily by, drove, driver, drivers, contributed to, contribute to, contributes to, contributor to, contributors to, contributions to, contributing to, affect, affects, affected, affecting, attribute, attributes, attributed, attributable to, cause, causes, caused, causing, impacted, impacting, factors, factor, reflected, reflect, reflects, reflecting, reflective of, reflection of, aided by, lead to, led to, leads to Performance Related Words income, incomes, loss, losses, earning, earnings, earnings per share, EPS, revenue, revenues, sale, sales, expense, expenses, cost, costs, profit, profits, SG&A, general and administrative, G&A, R&D, research and development, EBITDA, EBIT, gross margin, margin, deferred tax, tax rate, inventories, inventory, cash flow, cash flows, accounts payable, backlog, result, results, improvement, improvements, improving, increase, increased, decrease, decreased, decline, declined, performance, performed, success, successful, momentum, accomplishments, growth, grow, grows, grew Filler Words I believe, I also believe, we believe, we also believe, I think, I also think, we think, we also think, I guess, I also guess, We guess, We also guess, I mean, we see, I see, we expect, I expect, let me References Baginski, S. P., J. M. Hassell, and W. A. Hillison. 2000. Voluntary causal disclosure: tendencies and capital market reaction. Review of Quantitative Finance and Accounting 15:371-389.
Baginski, S. P., J. M. Hassell, and M. D. Kimbrough. 2004. Why do managers explain their earnings forecasts? Journal of Accounting Research 42 (1):1-29.
Banker, R. D., and S. M. Datar. 1989. Sensitivity, precision, and linear aggregation of signals for performance evaluation. Journal of Accounting Research 27 (1):21-39.
Barton, J., and M. Mercer. 2005. To blame or not to blame: Analysts' reactions to external explanations for poor financial performance. Journal of Accounting and Economics 39:509-533.
Bartov, E., and D. A. Cohen. 2009. The “Numbers Game” in the pre-and postSarbanes-Oxley eras. Journal of Accounting, Auditing & Finance 24 (4):505-534.
Beckman, L. 1970. Effects of students' performance on teachers' and observers' attributions of causality. Journal of Educational Psychology 61 (1):76-82.