Product Recalls and Audit Production

Abstract

We examine the impact of product recalls on audit production. Product recalls are events that increase engagement risk associated with greater audit effort, and by extension, greater audit quality. Consistent with this conclusion, we find firm-years with a product recall have a lower likelihood of being subsequently restated and lower levels of accrual error. While a positive association generally exists between audit effort and audit fees, prior literature shows firms experiencing economic stress negotiate lower audit fees and auditors accept lower fees to earn future returns. We expect that to increase the likelihood of retaining a client, auditors will agree to lower fees when the short-term economic stress associated with a recall injects downward pressure into fee negotiations. Supporting this conclusion, we find product recalls are associated with lower audit fees – a relation attributable to differing magnitudes of temporal fee increase. We also show the effects are transitory, the higher levels of audit quality are attributable to specialist audit offices, and the lower levels of fees are attributable to engagements where the client has greater bargaining power. While a positive association typically exists between audit effort and audit fees, our paper identifies a firm-specific event that weakens this relation.

Keywords:

• Product recall
• Restatements
• Discretionary Accruals
• Audit fees

JEL Codes:

• H22
• M42

Acknowledgement

We appreciate helpful comments and suggestions from Henrik Nilsson (editor), two anonymous reviewers, Lauren Cunningham, Emily Xu, and the workshop participants at the University of New Hampshire, Brigham Young University, the 2020 CAAA Annual Meeting, the 2019 AAA Annual Meeting and the 2022 KAA summer international conference.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplemental Data and Research Materials

Supplemental data for this article can be accessed online at https://doi.org/10.1080/09638180.2022.2130953.

Table A1: Number of product recalls and recall characteristics by calendar year.

Table A2: Correlation matrix.

Table A3: Changes analysis.

Table A4: First-stage determinants of recall issuance.

Table A5: Comparison between recall and propensity score matched non-recall observations.

Table A6: Propensity score matched analysis.

Table A7: Cross-sectional analyses.

Notes

1 The mean dollar value of the loss associated with a recall is approximately $18 million, and the associated reduction in firm value is approximately 5 percent for the average firm (Lee et al., 2015).

2 We use the term ‘economic stress’ because all the costs associated with a product recall are not immediately or fully reflected in the financial statements. According to extant literature, product recalls adversely affect brand equity and consumers’ perceptions of quality, costs that are not shown as expenses on the income statement (e.g., Chen et al., 2009; Laufer & Coombs, 2006; Van Heerde et al., 2007).

3 Information on the CPSC is publicly available at https://www.cpsc.gov/. A list of all product recalls issued since 1973, the source for data used in this study, is available at https://www.cpsc.gov/Recalls.

4 In the United States, the CSPC is one of four major regulatory agencies with the ability to initiate recall events; the United States Department of Agriculture (USDA) and Food and Drug Administration (FDA) issue food and drug related recalls, and the National Highway Transportation Safety Administration (NHTSA) issues motor vehicle related recalls. We examine CSPC recalls because their relative infrequency and previously documented financial impacts (Lee et al., 2015; Chen et al., 2009) make them the type of recall most likely to impact audit production.

5 If the manufacturing firm receives information supporting the conclusion that safety issues exist, the firm must report the issue to the CPSC within 24 h (Chen et al., 2009; Lee et al., 2015).

6 A firm also has the ability to ‘fast track’ a product recall and issue an announcement without waiting for results from the risk analysis (Chen et al., 2009). While not publicly disclosed for individual announcements, CSPC data indicates approximately 33 percent of recalls are fast-tracked (CSPC, 2017).

7 These joint recalls are considered voluntary recalls; the firm agrees with the agencies’ conclusion that a recall is warranted. In rare cases, a firm will not agree with the decision, forcing the agency to decide whether a mandatory recall must be imposed (Chen et al., 2009). However, due to the extensive and costly legal proceedings associated with the imposition of a mandatory recall, a voluntary recall is usually in the interest of both parties. Consequently, mandatory recalls are, on average, issued less than once per year (Mullan, 2004). Due to the infrequency of mandatory recalls, we do not differentiate between voluntary and mandatory recalls.

8 We believe the increase in risk from a product recall will be primarily attributable to the subjective judgments necessary to account for the event. These decisions are the focus of guidance issued by public accounting firms. However, we recognize product recalls may also affect the auditor’s risk assessment because they have an adverse impact on profitability. Product recalls impose numerous costs on the issuing firm and can have a significant effect on financial performance (Davidson & Worrell, 1992). Managers of public firms with weaker financial performance have greater incentives to overstate financial position. Because of these incentives, financially weak clients are assessed as having greater risk (Johnstone & Bedard, 2001; O’Keefe et al., 1994). We control for financial performance in our analyses.

9 While Allen et al. (2006) concludes the relation between risk and differences in the quantity of labor hours is not as strong as theory predicts, the auditor’s response to risk is not limited to a change in the quantity of labor. Johnstone and Bedard (2001, 2004) find that audit firms respond to areas of increased risk by assigning specialists or industry experts and by applying more intensive testing.

10 Chen et al. (2018) focus on the timeliness of litigation loss contingency disclosures. We complement Chen et al. (2018) by similarly examining whether expertise improves outcomes when accounting for a risky client-specific event. However, we focus on different reporting outcomes, including restatements, which represent an ‘egregious audit failure’ (DeFond & Zhang, 2014, page 277).

11 When auditors are unable to reduce the risk of material misstatement by increasing effort, they may opt to pass this residual risk on to the client in the form of a fee premium. This is referred to in the extant literature as a ‘risk premium’ (DeFond & Zhang, 2014).

12 This practice represents a threat to auditor independence because the auditor expects future returns on an initial investment. Regulators have argued this practice effectively creates a receivable from the client, incentivizing the auditor to yield to client pressure (DeFond & Zhang, 2014).

13 Alternatively, an auditor might not charge for greater effort because the ex ante risk associated with an adverse event is already impounded in the price of an engagement. Li et al. (2020) examines the association between cyber incidents and audit fees. Post-2011, when guidance suggested regulators were starting to take cybersecurity seriously, Li et al. (2020) finds ‘auditors price material cybersecurity risk prior to cyber-attacks and thus respond less severely (are less surprised) when an actual cyber incident occurs’ (Li et al., 2020, page 153). Alternatively stated, auditors don’t pass on to the client the full cost of greater audit effort the year an adverse event occurs because the potential cost is already accounted for in the engagement price.

14 While recall insurance may be purchased to cover some of the costs associated with a recall, these policies typically do not cover all expenses associated with issuance. This conclusion is supported by the disclosures of firms that purchase insurance. These disclosures often explicitly state the policies cover only direct costs and may not cover all loses.

15 While recall announcements may reference multiple parties, we ignore the upstream and downstream effects of recall issuance. Firm-years are considered recall years only if the firm issued a recall announcement.

16 To verify our results remain in the post-SOX period, we estimate Model (1) using a subsample that includes only fiscal years ending in 2004 or later.

17 Sellers et al. (2018) summarizes the approaches to operationalizing a restatement variable. The authors note that including all occurrences ‘provides the most inclusive measure of audit quality’ (page 3) and is appropriate when examining auditor or business characteristics. Francis and Michas (2013) similarly notes any restatement strongly suggests that the audit of the originally issued financial statements was of unacceptably low quality, and wider definitions of audit failure ‘provide insight on a much wider range of potentially low-quality audits than a narrower definition’ (p. 521). Consequently, for the presented analyses we adopt the same definition as Sellers et al. (2018). In the results section we discuss robustness tests using alternative definitions.

18 The modified Jones (1991) and Dechow and Dichev (2002) measures both capture accruals quality. The Jones Model estimates firm-specific parameters that can be used to predict non-discretionary accruals, and by extension, the discretionary component of total accruals. This is done by modelling total accruals as a function of assets, revenues, and fixed assets (PPE) using time-series data from a single firm. The focus of this study is an event, a product recall, that results in abnormal levels of non-discretionary accruals relative to firm-years in which no recall is issued. Since recalls are infrequent events, the firm-specific parameters generated using the Jones Model would likely under-estimate the level of non-discretionary accruals in the year a recall is issued, and by extension, over-estimate the level of discretionary accruals, biasing our results. In contrast to the Jones Model, the Dechow and Dichev (2002) measure reflects accrual estimation error and does not attempt to distinguish between discretionary and nondiscretionary accruals. Consequently, we use the Dechow and Dichev (2002) measure as our indirect proxy for audit quality. Because earnings managed in either direction meet the definition of poor audit quality, for the presented analyses we use the absolute value of the Dechow and Dichev (2002) measure. In the results section we discuss robustness tests using both income-increasing and income-decreasing accruals.

19 To verify our conclusions are not driven by omitted variables, we re-estimate Model (1) with additional controls for auditor and firm characteristics. Additional controls for auditor characteristics include variables that capture national and joint specialization, scale, and client importance (Bills, Swanquist, et al. 2016; Fung et al., 2012; Lobo & Zhao, 2013). Controls for governance characteristics include variables that reflect the characteristics of the board of directors, including size, independence, average tenure, and the presence of financial, supervisory, or accounting expertise; characteristics of the audit committee, including size, independence, and the presence of financial, supervisory, or accounting expertise; CEO characteristics, including whether the CEO is also the board chair (i.e., duality), and turnover (Baxter et al., 2013; Collier & Gregory, 1996; Hines et al., 2015; Hines & Peters, 2015). Finally, we control for other firm characteristics, including age and whether the firm is listed on the New York Stock Exchange (Baxter et al., 2013). Our main results are robust to the inclusion of these additional control variables in Model (1).

20 While our sample includes only 541 firm-years in which a product recall was issued, the effects we document likely apply to a larger population. During the sample period a total of 8357 recalls were issued, with the number of recalls in a year ranging from 217 to 654. For 6051 of the 8357 recalls issued we were unable to find the issuer in Audit Analytics. For most observations this is likely attributable to the firm being a private company.

21 To estimate the recall amount, we follow Lee et al. (2015). Recall amount is calculated for individual recalls as the number of units recalled multiplied by the mean price per unit. Since the CSPC announces the total units that may be subject to the recall, and the mean retail price, our estimation of the recall amount is likely overstated. Not all units are returned, and the refund may be less than the retail purchase price.

22 To obtain evidence the firms that issue recalls suffer economic stress, we examine the change in operating cash flows and sales in the year a recall is issued (t) and the following years (t+1 and t+2). We find issues have negative operating cash flows, and significantly less growth in year t. These effects persist in year t+1 before dissipating in year t+2. While the stress associated with a recall is likely to vary with firm, auditor, and recall characteristics, these statistics are consistent with the extant literature’s conclusion that the mean recall causes economics stress.

23 We also run a linear probability model for the analysis. Linear probability models mitigate concerns related to the incidental parameters problem that can occur with logistic regressions when including fixed effects. Untabulated results including fixed effects are similar to those reported in Table 2.

24 To examine whether the results presented in Table 2 are robust, we re-estimate our model using alternative definitions of RSTMNT, including restatements with a lag of no more than 37 months (Bryant-Kutcher et al., 2013), restatements that reduce net income, restatements that reduce stockholder’s equity, and non-reliance 8K Item 4.02 restatements. The mean value of RSTMNT is 0.12 for restatements with a lag of no more than 37 months, 0.09 for restatements that reduce net income, 0.09 for restatements that reduce stockholder’s equity, and 0.07 for non-reliance restatements. In untabulated analyses, we find the coefficient associated with RECALL is negative and significant, or close to significant, for: restatements with a lag of no more than 37 months (p = 0.01), restatements that reduce net income (p = 0.09), restatements that reduce stockholder’s equity (p = 0.04), and non-reliance restatements (p = 0.13).

25 Because the dependent variable in a logit regression is the log of the odds ratio, to interpret the economic significance of the coefficient estimates, we convert the test variables’ effect from odds to probabilities.

26 Our results hold for the subset of observations where residuals from the Dechow and Dichev (2002) measure are positive (p < 0.01) and the subset of observations where residuals are negative (p = 0.10).

27 For the subsample of firms that issued a recall during the sample period (RECALL_FIRM = 1), mean (median) audit fees are approximately $3,401,300 ($1,390,600).

28 To the best of our knowledge, no existing study examines the firm characteristics that predict recall issuance. Shipman et al. (2017) suggest: (1) it is inadvisable to include variables in a PSM model if not supported by theory, and (2) PSM models should be similar to multiple regression models. Consistent with this advice, we use the same variables from our main analysis in the first and second stage of our PSM analysis.

29 For the propensity score matched sample, mean (median) audit fees are $3,624,300 ($1,620,900).

30 In order to control for firm size, we match treatment firms to control firms in the same firm size decile. In other words, we find control firms with the closest prior-year audit fee that are in the same year, industry, and size decile.

31 In unreported tests, we replace ΔFEES with ΔLAF, calculated as the difference between LAF in year t and year t − 1. All inferences remain the same.