On the nonlinear relation between product market competition and earnings quality

Abstract

The literature documents conflicting results regarding the influence of product market competition on earnings quality. We extend this stream of literature by incorporating competition’s effect on both the opportunities and the incentives to manage earnings. The combination of both effects results in a nonlinear relation between product market competition and earnings quality. At low competition levels, additional information associated with one more rival helps reveal earnings irregularity and deter earnings management to a larger extent than its effect on the incentives to manage earnings, suggesting a positive relation between competition and earnings quality. At high competition levels, the latter effect dominates the former. We thus predict a positive (negative) relation between competition and earnings quality at low (high) competition levels. Consistent with our hypothesis, we document an inverted U-shaped relation between earnings quality and product market competition.

Keywords:

• Earnings quality
• Competition
• Non-linearity
• Earnings restatement
• Earnings management

JEL Classification:

• M41

Acknowledgements

We thank the editor, two anonymous reviewers, Taesik Ahn, Bok Baik, Bjorn Jorgensen, Qianqiu Liu, Willie Reddic (discussant), Ghon Rhee, Jaeyong Shin, David Wang, and workshop participants at Seoul National University, the University of Connecticut, the University of Hawaii at Manoa, and 2014 Northeast AAA meeting (Best paper award winner) for their helpful suggestions and comments.

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability

All data used in this study are publicly available from the sources identified in the text.

Notes

1 We define earnings quality as the ability of earnings to measure a firm’s financial performance (e.g. Statement of Accounting Concepts No. 1, Dechow and Dichev 2002; Dechow et al. 2010). We measure it in multiple ways developed in the literature, including the extent to which working capital accruals map into cash flows (Francis et al. 2005) and earnings restatement frequency (Hennes et al. 2014).

2 Our study does not completely rule out other alternative explanations for the conflicting results in the literature. For example, the different results could be due to selections of different sample firms, periods, or proxies for earnings quality and competition. Thus, when feasible, we hold a constant sample for different proxies for earnings quality and competition. We also replicate two studies and still document the nonlinear relation in their empirical settings (see Section 4).

3 Hart’s model assumes that management does not value income above a certain level. Once management utility strictly increases with income, Scharfstein (1988) shows that competition actually increases management slack.

4 Our hypothesis development and the model derivation in the Appendix make two implicit assumptions. First, at the lowest level of competition, managers still have incentives to manage earnings due to, e.g. incentives to hide inefficiency or meet thresholds specified in various contracts with stakeholders, such as compensation contracts or debt contracts. Second, fierce competition does not eliminate opportunities to manipulate earnings. Thus, our empirical analyses serve as joint tests of these assumptions and our hypothesis.

5 We also measure competition for year t−5 and year t−3 such that competition (partially) lags the accounting quality proxy. Untabulated results based on these measures are qualitatively similar to those in Table 3. For firms with multiple segments, we also recompute COMP_SALE as the weighted average of competition in all the industries in which the firms operate (Markarian and Santaló 2014). The weight for each industry is sales in the industry as a percentage of the firm’s total sales based on Compustat Segment data. Untabulated results based on this measure are qualitatively similar to those in Table 3.

6 At least five non-missing observations are required for STD_SALES, STD_CFO and NEG_EARN.

7 We also relax the requirement of a constant sample and use the maximum number of observations for the main test and the robustness tests. Our main results and inferences are mostly robust to this change.

8 When we instead use the number of firms within each industry to proxy for competition, the graph exhibits an inverted U-shape (not reported) similar to Panel A of Figure 2.

9 Without the squared term of competition in equation (1), the estimated coefficient on competition is significantly negative, consistent with univariate correlation.

10 To assess the generalizability of the nonlinear relation we document, we replicate Balakrishnan and Cohen (2014) and Karuna et al. (2012) and further allow for nonlinearity in their settings. We select these two papers due to their opposite results and their selection of a wide range of firm years and industries. We confirm their respective results when not considering nonlinearity. More importantly, we document nonlinearity between competition and earnings quality based on their samples. Results are available upon request.

11 We choose to examine each of the proxies separately for two reasons. First, using individual measures facilitates comparison of results in this paper with those in other studies. Second, we can gauge the robustness of the results to alternative measures of earnings quality and thus mitigate concerns about measurement error driving the results. For the same reasons, in the next subsection, we employ individual competition proxies. In section 5.3, we further construct composite measures of earnings quality and competition and document robust results based on the composite measures.

12 For the estimation of both AQ_DD and AQ_BLW for each firm-year, we require at least eight years’ data.

13 This dataset was obtained from Andy Leone’s website at https://sbaleone.bus.miami.edu. It could also be collected from the GAO dataset (www.gao.gov) after eliminating multiple announcements of the same restatement and additional restatements by the same firm and after correcting data errors.

14 In untabulated tests, where we examine the estimated innate portion of earnings quality, the coefficient on COMP_SALE (COMP_SALE² ) is −0.002 (0.001) with a t-value of −1.24 (1.14). Taken together, these results imply that earnings quality at management’s discretion drives the results in Table 3.

15 In untabulated results, we also document an inverted U-shaped relation between competition and earnings smoothness, which is measured as the ratio of standard deviation of assets-scaled earnings before extraordinary items to the standard deviation of assets-scaled cash flow from operations (Leuz, Nanda, and Wysocki 2003).

16 Firm years with missing R&D expenses in the Compustat are set to zero. If none of the firms in an industry report R&D expenses, we exclude the industry from the analysis. Thus, the sample for this analysis is reduced from 52,332 to 47,786 firm-year observations.

17 We also use the natural logarithm of the number of firms within each industry and find that the results (untabulated) are consistent with those in Panel C of Table 5.

18 We also include COMP_SALE, COMP_R&D, COMP_ROA, and COMP_MKTSIZE and their squared terms simultaneously in model (1) to analyze the relations between earnings quality and different types of competition. The coefficients on each competition measure and its squared term have similar magnitude to those in Tables 3 and 5, suggesting that competition among existing rivals and competition from potential entrants are distinct from each other.

19 The tariff data is available from Schott’s webpage at http://faculty.som.yale.edu/peterschott/sub_international.htm and from the NBER website at http://www.nber.org/. Tariff is computed as duties collected by U.S. Customs divided by the Free-on-Board value of imports. The original U.S. import data in Schott (2008) are provided by Feenstra (1996), Feenstra et al. (2002), and the U.S. Customs Service.

20 We retrieve the data from http://webuser.bus.umich.edu/feng/.

21 Li et al. (2013) find that COMP_LLM reflects aspects of competition distinct from other measures. In untabulated results, we find that in general, COMP_LLM is significantly positively correlated with the other competition measures except an insignificant correlation with COMP_MKTSIZE and a negative correlation with COMP_R&D. The absolute magnitude of all correlation coefficients is at or below 0.111.

22 In untabulated results, we use the Herfindahl-Hirschman Index provided by the U.S. Census Bureau for only the manufacturing sector and do not find a significant relation between competition and earnings quality (see also Ali, Klasa, and Yeung 2009). We find that the competition measure based on the census data is more highly correlated with competition from potential entrants (correlation coefficient is above 0.26) than competition among existing rivals (correlation coefficient is below 0.10), and has low standard deviation (0.05 versus 0.14 for COMP_SALE) due possibly to it being collected once every five years. The former property suggests a weak relation with earnings quality, as documented in Table 5, and the latter suggests the lower power of this measure.

23 One caveat of the composite competition is that we assign equal weight to each individual measure when their economic effects may vary. Nonetheless, we believe that these results provide a reasonable estimate of the economic significance of our findings.

24 We derive model (5) from model (1). Taking the derivative of earnings quality on competition from regression (1) yields dAQ_FLOS = β₁·dCOMP_SALE + 2β₂·COMP_SALE·dCOMP_SALE. We then replace the earnings quality variable with its industry mean and competition variable with the tariff rate.

25 For consistency with our main tests, we also measure the control variables at firm-year levels and document results consistent with Table 7. As described in Section 5.4, our main results are also robust to measuring control variables at industry-year level or firm-year level.

26 Note that in (a), we assume under-performing firms have stronger incentives to manage earnings, likely upward (see Markarian and Santaló 2014). However, our hypothesis of a nonlinear relation does not depend on this assumption and holds for both upward and downward earnings management. In Section 4.5, we further explore the interaction of relative performance with competition. In addition, we assume a linear relation between incentives and competition in (a). To the extent that the effect of competition on performance and liquidity risk is larger when competition is fiercer, the relation can be convex (e.g. I + COMP² /P). Imposing this nonlinear relation leads to a similar inference.