Disclosure Transparency and Disagreement Among Economic Agents: The Case of Goodwill Impairment

Abstract

We examine whether more transparent disclosure about goodwill impairment tests conveys useful information to sell-side analysts about the parameters used in the complex and often opaque impairment testing process. Drawing on a sample of European companies from 2006 to 2014, we construct a unique dataset on the transparency of goodwill impairment disclosure and develop two analyst disagreement measures by extracting analysts’ opinions about firms’ impairment decisions in brokers’ reports. We show that the level of disclosure transparency is negatively associated with both disagreement among analysts, a proxy for information uncertainty, and disagreement between analysts and managers, a proxy for information asymmetry. Further, we find that discount-rate-related disclosure transparency is associated with both types of analyst disagreement, while cash-flow-related disclosure transparency is associated with disagreement between analysts and managers only. Our paper speaks to the usefulness of goodwill impairment test disclosures to analysts, while also highlighting that opportunistic and boilerplate disclosure by some firms hampers the ability to resolve information asymmetry and information uncertainty.

Keywords:

• Corporate Disclosure
• Sell-side Analysts
• Goodwill Impairment
• Disagreement

JEL codes:

• M40
• M41

Acknowledgements

We are grateful for helpful comments and suggestions from Paul André, Leonidas Doukakis (discussant), Andrei Filip, Katharina Hombach, Peter Joos, Michel Magnan, Ramond Olivier, Luc Paugam, and especially Florin Vasvari (the Editor) and two anonymous reviewers. This paper has also benefited from conference discussions at AAA 2017, CIG 2017, EAA 2017, and BAFA 2018. We thank Camille Raynaud for her excellent research assistance. All errors in the paper are our own.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental Data and Research Materials

Supplemental data for this article can be accessed on the Taylor & Francis website, 10.1080/09638180.2019.1677259.

Notes

1 In line with Paugam and Ramond (2015) and Lobo et al. (2017), we use disclosure transparency to refer to both disclosure quality and disclosure quantity.

2 In 2018 alone, European nonfinancial companies covered by Thomson Reuters Eikon reported a total of €23.6 billion of goodwill impairment. This amount represents an increase of 6% over the amount recorded in 2017.

3 Opportunistic use of goodwill impairment is evidenced by a stream of literature showing that the decision not to impair goodwill is associated with agency-theory-based motives (e.g., Li et al., 2011; Ramanna & Watts, 2012).

4 We consider goodwill impairment to be material if its amount exceeds €10 million or 1% of beginning total assets (Jarva, 2009; Knauer & Wöhrmann, 2016).

5 Discount rates are a single piece of information that can be inferred from other internal and external disclosures, while cash flow projections have multiple components, such as near-term cash flows, mid-term cash flows, and terminal values, that are not externally verifiable.

6 The recoverable amount of goodwill for a CGU is defined as the higher of the fair value of goodwill less costs of disposal and its value in use (VIU). The recoverable amount of goodwill is mostly determined based on its VIU (Petersen & Plenborg, 2010), which is usually calculated using the discounted cash flow method (IAS 36).

7 Although IAS 36 and the Statement of Financial Accounting Standards 142 are relatively similar, firms based in Europe and the United States (U.S.) exhibit different patterns of goodwill impairment recognition (André, Filip, & Paugam, 2016). Specifically, relative to U.S. firms, European firms book more untimely goodwill write-offs. Consequently, IFRS offers an interesting setting to measure the disclosure transparency of goodwill impairment tests.

8 The sample in André et al. (2018) covers only one year, whereas the sample in Paugam and Ramond (2015) is limited to a single country.

9 Opportunism may manifest in the selection of inappropriately lower or higher discount rates, the number of forecasting periods to discount future cash flows, the current level of cash flows, or the terminal value.

10 We examine non-zero goodwill firms with no impairments as a control group as part of the robustness tests in Section 6.

11 Analysts’ reports are retrieved from Thomson Reuters InvesText, analyst data from I/B/E/S, goodwill impairment data and all other firm and market data from Thomson Reuters Eikon.

12 Morningstar, Inc. is an investment research firm that compiles and analyzes funds and stocks.

13 Dividing the disclosure index by the number of items implies that all items are considered equally important.

14 Given that it is difficult to identify the exact date on which goodwill impairment is announced for all observations in our sample, we choose to focus on analysts’ reports issued within a period after the annual earnings announcement date for fiscal years in which goodwill is impaired. To ascertain that earnings announcement dates are likely the first dates when goodwill impairment news is made public, for a subset of firms with non-zero goodwill impairments during our sample period, we manually downloaded their goodwill impairment announcements on Factiva for a 12-month period from three months after the last fiscal year end to three months after the current fiscal year end. Out of the 61 firm-years in the search, we find that 23% had related announcements during interim reporting periods. We did not find any observations announcing goodwill impairment other than during quarterly or annual earnings announcements. We excluded observations that have non-zero goodwill impairments on Thomson Reuters Eikon during interim reporting periods from our sample.

15 Appendix B provides examples of analysts’ discussions of goodwill impairment losses in their research reports.

16 In most instances, all opinions expressed in the same research report are either in agreement or disagreement with the managers. If the Python coding process determines that analysts have expressed divergent opinions in the same report, we read the corresponding report to determine the direction of the opinions expressed.

17 We do not include non-directional opinions. In Section 5, we incorporate non-directional opinions into the measure as a robustness test.

18 Our measure of disagreement among analysts is positively correlated with analysts’ earnings forecast dispersion (Pearson correlation coefficient = 0.166; p-value = 0.048), and our measure of disagreement between analysts and managers is positively correlated with analysts’ earnings forecast error (Pearson correlation coefficient = 0.127, p-value = 0.094), thus providing some comfort that the two measures are valid proxies for analyst disagreements.

19 Since DISAGREEMENT _AA is bounded between zero and 0.707, we also take the natural logarithm of the variable and re-estimate all regressions. The results remain qualitatively unchanged.

20 This regression specification implicitly assumes that analysts’ ex-ante beliefs are held constant. We attempted to determine analysts’ ex-ante beliefs by analyzing their research reports issued within three months before the annual earnings announcement dates for all sample firms. We found that only a very small number of reports explicitly express directional opinions on goodwill impairment, and the sample size would be reduced to 38 observations if we required directional opinions both before and after earnings announcement dates. We deem it not meaningful to conduct analyses using this small sample.

21 Following Kothari et al. (2009), we also use the natural logarithm of market value of equity as a measure for SIZE in all analyses. Our results remain qualitatively similar.

22 Our primary results remain qualitatively unchanged when we cluster the standard errors at the firm or analyst levels in both estimations. We choose not to cluster the standard errors because many clusters contain only one observation.

23 An alternative explanation to the negative association is that if managers intentionally disclose a substantial amount of misleading information about the impairment testing process and parameters, in the extreme case, analysts may all disagree with the firm, resulting in lower disagreement among analysts. While this is a possible explanation, past literature generally uses disclosure quantity as a proxy for disclosure quality (e.g., Francis, Nanda, & Olsson, 2008; Hail, 2002), and there is some evidence that the two constructs are positively correlated (André, Filip, & Moldovan, 2016). We thank an anonymous reviewer for highlighting this alternative explanation to us.

24 It is not always possible to verify all information about discount rates because firms have multiple CGUs, whose discount rate parameters are only observable to a certain extent.

25 By complexity, we refer to multiple layers of information, such as the cash flow stream and the terminal value, and the lack of verifiability of the information.

26 Countries classified as having a low enforcement regime are Austria, Belgium, the Czech Republic, Finland, Germany, Hungary, Ireland, Netherlands, Poland, Spain, and Sweden; countries classified as having a high enforcement regime are Denmark, France, Italy, Norway, Switzerland, and the United Kingdom. We use the 2005 total enforcement score to partition 2006–2008 sample observations, and the 2008 total enforcement score to partition observations from 2009 onwards.

27 We also partition the sample into high and low enforcement quality by common vs. civil law countries as in Knauer and Wöhrmann (2016) and three clusters of countries as in Leuz, Nanda, and Wysocki (2003). The coefficients on the interaction term(s) remain statistically insignificant in both specifications.

28 It is not feasible to include non-directional opinions in the metric on disagreement between analysts and managers. We provide an explanation in the discussion of Table 7.

29 The covariates in the model include stock return, lagged stock return, an income-smoothing indicator, a big-bath indicator, financial leverage, a Big 4 indicator, free-float percentage, institutional ownership, analyst coverage, goodwill amount, number of segments, number of consecutive years with goodwill impairment losses before the current year, return on assets, size, market-to-book ratio, and monthly stock return volatility. We do not include CEO compensation and CEO turnover in the model because we do not have access to the data. To maximize the number of matches, we do not include country, industry, and year as matching criteria.

30 According to Shipman, Swanquist, and Whited (2016), the caliper distances range from 0.00005 to 0.23 in 29 out of 86 accounting research studies that employ propensity score matching in their methodology and report the information on caliper distance. Although we use a large caliper distance of 0.2 in the propensity matching process, only 120 observations out of the starting 154 impairments could be matched with observations in the no impairment group. The matched sample is further reduced to 76 pairs after eliminating observations with fewer than two opinions issued by different analysts and non-missing values for the control variables.

31 The International Accounting Standards Board has a project on its agenda related to the accounting for goodwill and goodwill impairment. See http://www.ifrs.org/projects/work-plan/goodwill-and-impairment/.

Additional information

Funding

The authors acknowledge the financial support of Autorité des Normes Comptables.