The impact of cost stickiness on income smoothing: evidence from employment protection regulations

Abstract

AbstractPrior literature suggests that cost stickiness increases the ex-ante volatility and reduces the predictability of earnings. We examine whether managers intentionally undo such consequences by dampening earnings volatility. Exploiting the staggered adoption of wrongful discharge laws as an exogenous instrument for cost stickiness, we document that cost stickiness increases managers’ income-smoothing activities. This response is more pronounced in firms whose earnings are more sensitive to labour costs than their industry peers are and in firms with stronger information-provision incentives. Additional analyses indicate that income smoothing improves sticky-cost firms’ earnings informativeness and that the identified impact of cost stickiness is primarily driven by labour costs. Our results suggest that labour regulations can influence managers’ financial reporting incentives via cost behaviour.Keywords: cost stickinessincome smoothingemployment protectionearnings informativeness AcknowledgmentsWe are heavily indebted to the associated editor Stefano Cascino and two anonymous reviewers for their constructive and thoughtful guidance. We thank Beatriz García-Osma, Thomas Jeanjean, Anne Jeny, Sanjay Kallapur, Itay Kama, Reuven Lehavy, Andreea Moraru-Arfire, Naomi Soderstrom, Samuel Tan (discussant), and workshop participants at ESSEC Business School and the EAA Annual Congress 2019 in Paphos for their helpful comments and suggestions. Junqi Liu also gratefully acknowledges the financial support from the National Natural Science Foundation of China (NSFC), grant number 72202190, from the Social Science Foundation of Fujian Province, grant number FJ2022C034, from the Fundamental Research Funds for the Central Universities, grant number 20720221042, and from the Association Francophone de Comptabilité (AFC). All remaining errors are our own.Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1 Cost stickiness arises primarily from the asymmetry in managers’ real resource commitment. When activity levels increase, managers add resources to meet growing demand; when activity levels fall, they retain some of their underutilized resources, because they perceive the adjustment costs of reducing the resources as higher than the costs of holding them (Anderson et al. Citation2003). In this sense, cost stickiness is distinct from managers’ financial reporting choice.2 In this study, we focus on accrual-based income smoothing, which reflects managerial effort to reduce earnings volatility through accounting methods (in contrast to real activities); see Section 2.2 for more detail. We do not discuss income smoothing through real activities such as discretionary adjustments of R&D or marketing expenditures because such activities are likely detrimental to firm performance in the long run.3 For example, if each employee uses a truck or a software license (i.e. complements to labour), then greater labour cost stickiness will lead to greater stickiness in the associated truck or software costs. Alternatively, if the firm uses temporarily idle employees to perform tasks that are normally outsourced to external contractors (substitutes to in-house labour), then greater labour cost stickiness could lead to lower stickiness for these external costs. To this end, operating costs properly capture these cross-resource effects.4 Recent literature also identifies conditions under which costs exhibit anti-stickiness (e.g. Kama and Weiss Citation2013, Banker et al. Citation2014, Chen et al. Citation2019). We focus our discussion on cost stickiness because this is a significantly more common phenomenon, and our setting of WDL adoptions primarily increases the adjustment costs of labour and therefore lead to stronger operating cost stickiness.5 Untabulated analyses show that cost stickiness is negatively associated with earnings persistence and positively associated with pre-managed earnings volatility and operating cash flow volatility, providing further evidence that cost stickiness likely reduces earnings informativeness and increases ex-ante earnings volatility. Our definition of pre-managed earnings is outlined in Appendix B.6 Theoretically, low cost elasticity (i.e. the symmetric response of costs to activity increases and decreases) may as well lead to high earnings volatility. However, the symmetric pattern of cost changes in response to changes in activity can be anticipated by investors (Banker and Chen Citation2006). Therefore, unlike high cost stickiness, low cost elasticity exhibit no impact on the predictability of earnings (Hutton et al. Citation2012), as it does not hinder investors from correctly evaluating negative shocks.7 When retaining unused resources during a sales decrease, managers sacrifice current earnings but save on expected future adjustment costs, which leads to a future economic benefit. However, unsmoothed reported earnings for the current period do not incorporate such a future benefit. Managers who are privately informed of this benefit could smooth earnings to communicate this information and reflect the “true economic performance” of the firm.8 The legal profession distinguishes three types of WDLs: the good-faith exception, the public-policy exception, and the implied-contract exception. For a detailed discussion of the WDL adoption, see Walsh and Schwarz (Citation1996), Miles (Citation2000), Kugler and Saint-Paul (Citation2004), and Autor et al. (Citation2006).9 Serfling (Citation2016) shows through a survival analysis that the adoption of WDLs is effectively unpredictable.10 WDLs may have real effects on firms, such as increased innovation (Acharya et al. Citation2013) and share repurchases (Dang et al. Citation2021), and decreased profitability (Bird and Knopf Citation2009), financial leverage (Serfling, Citation2016), firm growth (Bai et al. Citation2020), and tax aggressiveness (Fairhurst et al. Citation2020). The latter set of channels are associated with lower, rather than higher, income smoothing (see Carlson and Bathala Citation1997, Mayberry et al. Citation2013). Hence, if the adoption of WDL exceptions affected income smoothing through a real activities channel, our instrumented diff-in-diff results would be more prone to showing a negative, rather than positive, impact of cost stickiness on income smoothing. Despite potential underestimation, we still find a positive impact, which validates the adoption of WDL exceptions as an instrument for cost stickiness.11 We thank an anonymous reviewer for suggesting this method to address the mechanical link between sticky costs and accruals.12 We also used linear and quadratic spline corrections, and the results remain qualitatively similar.13 Weiss (Citation2010) constructs an alternative measure of cost stickiness, which is premised on quarterly data but conceptualized to generate firm-year observations. This measure constructs a standalone variable of cost stickiness at the firm-year level, simplifying the regression analyses. However, this procedure is, by construction, more likely to retain observations with more volatile earnings, leading to a potential selection bias (Banker and Byzalov Citation2014), and the validity of this procedure is also econometrically unwarranted. Hence, we follow the majority of cost-stickiness studies and capture this construct using a cross-sectional regression. Our untabulated results remain qualitatively similar when we apply the Weiss (Citation2010) measure of cost stickiness.14 Industry (or firm) fixed effects are not included in the first-stage regression, because doing so would lead to prohibitively too many coefficients for interaction terms to be estimated and included in the construction of the variable STICKYt. Our results remain qualitatively similar if we do not control for industry (or firm) fixed effects in the second-stage estimation.15 There is a slight discrepancy in the WDL adoption dates in the literature. Utah is not considered as having adopted the good-faith exception in Autor et al. (Citation2006), but other works (e.g. Walsh and Schwarz Citation1996, Serfling Citation2016) code Utah as having adopted the good-faith exception in 1989. Our results are robust to the different versions of coding.16 A firm may operate in multiple states, and its employees are protected by WDLs only if they work in the state in which WDLs are in force. Given that the information about employee location is not available on Compustat, we follow the extant research to capture each firm’s primary location of operations through its headquarters state. Another issue is that the headquarters state recorded on Compustat is the firm’s most recent headquarters location. Nonetheless, if a firm relocated to another state during the sample period, the measurement error will bias against finding our hypothesized results.17 New Hampshire (adopted the good-faith exception in 1974) and Oklahoma (adopted the good-faith exception in 1985) reversed the passage of the good-faith law in 1980 and 1989, respectively. The 11-year window excludes post-reversal observations from the sample.18 We retain all observations within the full 35-year sample period for the control firms. Since control firms do not have an adoption date, we cannot drop observations following the same criteria applied to treated firms. Furthermore, since all control firm observations lie in the 10-year window for at least one treated firm, we cannot drop any control observations due to the lack of a treated counterpart. Even though the number of observations differs significantly between treated and control firms, it should not bias our inference as each treated observation has its (possibly multiple) control counterparts, and each control observation has its treated counterpart.19 In untabulated tests, we remove the 11-year window restriction and keep all observations within the full 35-year sample period for the treated sample, and our inferences remain qualitatively similar.20 The descriptive statistics for the income smoothing measure (SMOOTHt+1) is similar to those reported in other studies such as Hamm et al. (Citation2018) and Tucker and Zarowin (Citation2006).21 We caution that our regression models differ from Heath et al.’s (2022) standard specifications, and the difference might lead to different critical values for a statistic inference purpose.22 We do not control for state fixed effects in columns (1) and (2), as state fixed effects become effectively redundant after controlling for firm fixed effects. Untabulated results show that our results in columns (1) and (2) do not change after adding state fixed effects.23 Relative to the standard error in an estimation without the bootstrapping adjustment (0.035), the standard error for STICKYt more than doubles with the bootstrapping adjustment (0.074).24 Firms could self-select to relocate into/out of a state if they are for/against a forthcoming WDL adoption. To alleviate this concern about firms’ self-selection being associated with the observable variables, we replicate our tests of both stages in an entropy balanced sample in an untabulated test. The qualitatively similar results alleviate the concern that firms’ self-selection of location drives our results.25 We caveat that discretionary revenues consist only part of firms’ discretionary accruals. An income-smoothing measure based on discretionary accruals might omit other accrual components that managers use to smooth earnings, leading to potentially biased findings. As such, we do not use this measure in our main analyses.26 We obtain the same inference by replacing ROAt-1 with CFOt-1, and by excluding ROAt-1, following equation (3) and footnote 17 in Baik et al. (Citation2020).27 The data and methodology for calculating different versions of EPL scores are available on www.oecd.org/employment/protection. We use the first version of EPL scores since it covers the whole sample period. Version 2 has only scores for regular contracts, whereas version 3 starts coverage from 2008, which would significantly reduce our sample size. In untabulated tests, we also use the scores in versions 2 and 3 as alternatives to capture the strictness of EPL, and the results are qualitatively similar.28 Untabulated results show that EPL scores are not associated with other components of operating costs than labour costs (i.e. operating costs minus labour costs), indicating that the results for operating costs primarily reflect the effect for labour costs.Additional informationFundingJunqi Liu gratefully acknowledges the financial support from the National Natural Science Foundation of China (NSFC), grant number 72202190, from the Social Science Foundation of Fujian Province, grant number FJ2022C034, from the Fundamental Research Funds for the Central Universities, grant number 20720221042, and from the Association Francophone de Comptabilité (AFC). All remaining errors are our own.