Can Investors Adjust for Managerial Bias?

Abstract

Financial information from firms often contains biased information. In this study, we posit and experimentally test the idea that investors will have difficulty in unraveling known biases in management’s earnings forecasts but will be most likely to fully adjust when the information about bias is in quantitative, EPS form and the investor’s judgment is compatible with that bias information (also in quantitative, EPS form). Results from three experiments suggest that indeed quantification and compatibility are beneficial for unraveling managerial bias, but even under these conditions not all investors are able to unravel. We also show that this result is robust to several moderator variables that capture factors that are commonly found in the management earnings forecast setting. Our study has implications for firm managers, regulators, and investors.

Keywords:

• Forecasts
• Bias
• Unraveling

Acknowledgement

We thank participants at workshops in the following universities—Alberta, Amsterdam, Erasmus, Georgia Tech, Kentucky, Lehigh, Lethbridge, Iowa, Iowa State, Massachusetts, Melbourne, Saskatchewan, Texas, and Virginia—for their insightful comments and suggestions on previous drafts of this paper. We also thank our anonymous reviewer, Aysa Dordzhieva, and Hyun Hwang for their helpful comments.

Data availability

Notes

1 A number of studies in the analyst domain similarly reveal that investors have difficulty unraveling bias in analyst forecasts (Herrmann and Thomas 2005; Hilary and Hsu 2013; So 2013).

2 All experiments reported herein were conducted at different times and were conducted with different participants. Moreover, all experiments reported herein are the first experiment in a set of two otherwise unrelated experiments, with only the first experiment in each set pertaining to the current paper. Because the experiments reported herein always were the first study that the participant completed, there are no concerns for carryover effects for the results reported in this paper.

3 Human subjects’ approval was obtained for all experiments reported herein.

4 Quantitative bias and quantitative responses create quantitative compatibility, while qualitative bias and qualitative responses lead to qualitative compatibility. Incompatibility is created when quantitative (qualitative) bias is paired with a qualitative (quantitative) response.

5 To be clear, all of these comparisons are between-participant tests. The across-bias-comparisons are essentially conducting interaction tests by comparing the difference in mean judgments for optimistic and pessimistic conditions across the bias conditions (e.g., qualitative high bias versus quantitative high bias), while the within-bias comparisons are testing for differences in mean judgments for optimistic and pessimistic conditions for a given bias condition (e.g., quantitative high bias).

6 We measure full unraveling as those who adjust their EPS estimates to $1.26 or higher (lower) in the pessimistic (optimistic) high bias conditions (i.e., they adjusted their EPS estimates by five cents or more). We acknowledge that those participants who overcorrected by making a bias correction of more than five cents could be excluded from these tests. Although robustness tests reveal that this small number of participants does not change any inferences, we nevertheless include them in the tabulation as doing so allows the reader to readily determine how many participants undercorrected for bias (i.e., one minus the reported frequency).

7 We also omitted the consensus forecast from the experiment three materials. Recall that we included this $1.26 consensus forecast in the materials to not only draw attention to the direction of bias in management’s forecast, but also to provide an independent forecast that would be confirmed if participants were to fully unravel management’s biased forecast (of either $1.31 or $1.21) by the historical bias of five cents. Although one might argue that the inclusion of this consensus forecast would bias our results toward full unraveling, it is possible that it was not viewed as credible given its issuance one month prior to the CEO’s forecast. Thus, we eliminated this forecast from the experiment three materials.

Additional information

Funding

Funding for this paper was provided by the Harrington Fellowship Program at The University of Texas, the Chartered Professional Accountants Education Foundation of Alberta, and the Deloitte Foundation.