The Influence of Auditor State-Level Legal Liability on Conservative Financial Reporting in the Property-Casualty Insurance Industry
DeFond and Francis (2005) observe that the legal system affects auditor behavior and, by implication, audit quality, through the standard of care auditors must meet to legally satisfy audit responsibilities and avoid litigation arising from deficient audits. The auditor's liability is affected by statutory law and common law. Statutory responsibilities were initially defined by the Securities Act of 1933 and the Securities Exchange Act of 1934, and were largely unchanged until the Private Securities Litigation Reform Act of 1995, which reduced the auditor's liability exposure under federal securities law. Unlike statutory law, which is applied at the federal level, the auditor's common law liability to third parties is based on court cases and legal precedents that are decided at the state level. A fundamental distinction is whether the state follows the privity doctrine, which restricts third-party claimants to those with contractual privity with the auditor. The audit profession has argued that any departure from privity raises litigation risk to unacceptable levels (Cook et al. 1992).
Although prior research suggests that audit quality is higher when auditors face greater liability exposure under statutory law (Lee and Mande 2003; Venkataraman et al. 2008), no research has addressed the association between state liability standards and auditor behavior. The difficulty of addressing this issue is noted by Talley (2006), who notes that settlement information in private suits in state courts is frequently sealed, which restricts the historical information available for analysis. However, the notion that auditor litigation risk is higher in non-privity states is supported by Linville (2001), who finds that auditors operating in these states pay higher malpractice premiums.
This study provides the first evidence that the legal standard in a state-privity or non-privity-is associated with the quality of audits conducted within its borders. We equate audit quality with earnings management, and limit the analysis to property-casualty insurers. The insurance setting is advantageous for two reasons. First, an objective measure of earnings management is available for these firms. Petroni (1992) reports that financially struggling insurers tend to inflate their apparent financial position by understating their loss reserves. Mandated disclosures to insurance regulators can be used to measure this bias. Second, by focusing on the insurance industry, we can identify a sample of private firms that only operate in a single state. Auditor liability is not influenced by federal statutory law when the client is private, which allows us to focus on variation in state-level liability standards. The potential for forum shopping is minimized when the insurer is only licensed to write policies in the domicile state, which increases the precision with which we can identify the relevant auditor liability standard.1
The question that we address is whether less under-reserving by weak insurers is observed when the insurer is domiciled in a non- privity state, where the auditor is presumably subjected to greater third-party litigation risk. The results, based on a sample of 3,107 observations from 1993 through 2004, suggest that auditors are less likely to allow reserve understatements by struggling insurance clients domiciled in non-privity states, and more likely to allow reserve understatements in privity (and near-privity) states.
We recognize that reserve bias among struggling insurers is potentially influenced not only by litigation risk, but also by the regulatory stringency in the state of domicile (Gaver and Paterson 2000). Therefore, we control for the regulatory climate in the insurer's state of domicile in our empirical tests. Our measures of litigation risk and regulatory stringency, although both state- specific variables, are not significantly correlated. Thus, auditor legal liability to third parties under common law, as defined by the state of domicile, does appear to matter. We also find that under- reserving by weak insurers in non-privity states is lower when the auditor is a Big 6 firm and when the auditor is an insurance industry expert. Prior researchers report similar findings regarding auditor size using research designs that involve statutory changes (Lee and Mande 2003; Geiger et al. 2006) and cross-country designs (Francis and Wang 2008).
The paper proceeds as follows. The second section describes the institutional setting of the study. It explains industry practices unique to property-casualty insurance firms that are important to the study. It also provides background on alternative judicial standards for auditor liability to third parties and explains how these standards differ across states. The third section presents the hypotheses. The fourth section describes the sample selection process, defines the variables, and provides descriptive statistics. The fifth section reports the results, and the sixth section concludes.
Insurer Loss Reserves
Insurance firms follow what are known as statutory accounting principles (SAP) for reporting to state insurance commissioners. Under SAP, the loss reserve is the insurance firm's estimated liability for unpaid claims on all losses that occurred by the balance sheet date. Petroni (1992) contends that the loss reserve is the most likely account through which insurance managers adjust the reported financial position of their firm. It is the largest liability on insurer balance sheets, and requires substantial managerial judgment. Of particular concern is the possibility that a weak insurer has understated the reserve in an effort to inflate its apparent financial health. Accordingly, while the auditor's overall charge is to evaluate the fairness of the insurer's financial statements in their entirety, the loss reserve is singled out as an account requiring special scrutiny.2
Prior researchers gauge reserve management by comparing the insurer's initial estimate of losses arising during a period to the total claims resulting from those losses that are eventually filed against the insurer (Petroni 1992; Petroni and Beasley 1996; Beaver et al. 2003; Gaver and Paterson 2004, 2007). Such a comparison is possible because of the extensive disclosure requirements of SAP. Table 1, excerpted from the 1998 Statutory Annual Statement of General Electric Mortgage Insurance Company (GEMIC), provides an example of these disclosures.3 The loss reserve nets aggregate estimated losses against cumulative cash payments for current and previous loss years. Thus, the loss reserve reported in the 1993 GEMIC balance sheet reflects the sum of all losses estimated for 1993 and prior years (Column 5 of Panel A, a cumulative amount of $856.216 million), less the sum of all cash payments for losses incurred in 1993 and earlier (Column 5 of Panel B, a cumulative amount of $499.293 million). This amount is $356.923 ($856.216 , $499.293) million.
Although cash payments are a matter of record, losses are subject to judgment. At the end of 1993, estimated losses for years up to and including 1993 totaled $856.216 million. By the end of 1998, the estimate for the same loss period had been increased to $972.798 million (the sum of the first five rows of Column 10, Panel A). The difference between the revised estimate of cumulative losses ($972.798 million) and the cumulative cash payments ($499.293 million) is known as the ''developed reserve.'' Thus, the 1998 developed reserve for 1993 (and earlier) losses is $473.435 ($972.728 minus $499.293) million.
In our study, we use a five-year development period to determine reserve bias, which is consistent with Petroni (1992), Petroni and Beasley (1996), and Gaver and Paterson (2000, 2004, 2007).4 For GEMIC, the five-year developed reserve for 1993 is $473.435 million, and the original 1993 reserve is $356.923 million. The comparison shows that GEMIC is under-reserved by $116.512 million in 1993. This makes sense because it is one of the insurers in our study that is categorized as financially weak, and Petroni (1992) finds that weak insurers tend to under-reserve.
State insurance commissions have used IRIS (Insurance Regulatory Information System) ratios since the early 1970s as an initial screen to identify firms for further regulatory scrutiny. A ''usual range'' is developed for each ratio, which encompasses results expected from the majority of companies during a normal year. Because economic conditions are not static, the components of each ratio are reviewed annually and revised when deemed necessary (National Association of Insurance Commissioners [NAIC] 1994).
Appendix A provides a definition of each ratio, explains how it is affected by the loss reserve estimate, and states the usual range for the ratio during the sample period. Under-reserving boosts reported policyholder surplus, a statutory account analogous to the combined retained earnings and paid-in capital accounts of a company following generally accepted accounting principles. Eight of the 12 IRIS ratios are improved by understating the loss reserve, and only one ratio (Ratio 11) is worsened through understatement. In the three other cases, reserves either do not affect the ratio (Ratios 2 and 5), or the effect is indeterminate (Ratio 6). The IRIS system has been a boon to researchers seeking an objective measure of insurer financial health because virtually all propertycasualty insurers must participate in the program, and ratio definitions and insurer results are publicly available from the NAIC. Auditor Liability to Third Parties for Negligence
Maroney et al. (2000) describe three basic judicial viewpoints that underlie state liability standards for auditors: privity, Restatement, and reasonable foreseeability. Privity imposes the lowest litigation risk on auditors. Strict privity requires a direct contractual relationship between an accountant and a third party for the latter to be able to sue the practitioner for negligence; nearprivity requires the suing party to be an intended third-party beneficiary of the contract between the accountant and the client. Compared to privity, the Restatement standard increases auditor liability by allowing third parties to sue if they suffer a pecuniary loss in a transaction that the audited information was intended to influence. The reasonable foreseeability rule imposes the greatest liability on auditors of all three standards. Under this rule, auditors can be sued by any third party that they should reasonably foresee as receiving and relying on their work.
The accounting profession has argued that any departure from privity increases auditor litigation risk to unacceptable levels (Cook et al. 1992). In our analysis, we code states that follow the strict or near-privity rule as low litigation risk environments and those that follow the Restatement or the foreseeability standard as high litigation risk environments. Appendix B specifies the liability standard that we assign to each state during each year of the sample period, and our rationale for doing so.
We argue that in jurisdictions where the likelihood of third- party litigation is high, auditors will demand more conservative reporting from client firms. Weak insurers have incentives to inflate their apparent financial condition by understating their loss reserves (Petroni 1992). In the context of our study, therefore, auditors take a conservative position when they oppose reserve understatement by financially struggling insurance clients. This leads to our first hypothesis:
H1: The magnitude of the reserve understatement by financially weak property-casualty insurers is lower when the insurer is domiciled in a state that has a higher likelihood of third-party auditor litigation.
Prior research indicates that audit firm size influences the relation between the legal environment and the degree of conservatism in financial reports.5 Lee and Mande (2003) and Geiger et al. (2006) observe a decrease in conservatism following the passage of the Private Securities Litigation Reform Act of 1995 for auditees of Big 6 firms, but not for auditees of non-Big 6 firms. Similarly, Francis and Wang (2008) report a greater degree of conservatism among firms with Big 4 auditors when a country's investor protection regime gives stronger protection to investors. In contrast, among firms with non-Big 4 auditors, conservatism is largely unaffected by different investor protection regimes. These findings lead to H2:
H2: The reserve understatement by financially weak insurance clients in high-liability states is less pronounced when the auditor is a Big 6 firm.6
Bruynseels et al. (2011) argue that industry-specialist auditors develop a high level of knowledge of a client's industry and also are more vulnerable to litigation and reputation loss in the event of audit failure. Thus, compared to non-specialist auditors, industry specialists have a greater ability and a greater incentive to produce high-quality audits. Supporting this, prior researchers find that auditor industry specialization is associated with fewer violations of GAAS reporting standards (O'Keefe et al. 1994), greater ability to generate alternative hypotheses when trying to identify accounting errors (Wright and Wright 1997), more insight into non-error explanations for unexpected ratio fluctuations in analytical procedures (Solomon et al. 1999), more effectiveness at detecting errors in staffworking papers during the audit review process (Owhoso et al. 2002), lower levels of earnings management (Krishnan 2003; Balsam et al. 2003) and lower incidence of fraud (Carcello and Nagy 2004). Further, audit pricing studies document price premiums for industry-specialist auditors (Ferguson and Stokes 2002; Ferguson et al. 2003; Francis et al. 2005). Auditor industry expertise is likely to have a particularly strong impact on audit/ reporting quality in the insurance industry because it is regulated (Petroni and Beasley 1996). This motivates H3:
H3: The reserve understatement by financially weak insurance clients in high-liability states is less pronounced when the auditor is an insurance industry expert.
The initial sample consists of all 32,411 insurer-year observations in the property-casualty database of the NAIC during the years 1993 through 2004.7 We begin the analysis in 1993 because it is the first year that the name of the auditor is reported in the statutory annual statement. We end the analysis in 2004 in order to allow a sufficient loss reserve development period, and also because auditor data must be hand-coded, making data collection costly.8 An advantage of our sample period is that it includes years where the property-casualty industry as a whole tended to over-reserve (1993- 1997), as well as years when the industry under-reserved (1998- 2002) and returned to over-reserving (2003-2004).9 The insurance industry reacted to the liability crisis in the 1980s by over- reserving during the first part of our sample period. This trend reversed in 1998 after multiple years of excess reserves, and returns to over-reserving in 2003. Data are drawn from A.M. Best Aggregates and Averages (Part 2 Summary). This provides a natural control for exogenous industry trends and allows us to focus on the influence of auditor legal liability on insurer reserve errors.
For an observation to be retained for analysis, we require that the insurer is domiciled within the United States and organized as either a stock company or a mutual company. The insurer must also have loss reserves subject to managerial discretion. For this reason, we drop observations if the insurer engages in a pooling arrangement or cedes all premiums to other insurers. We also delete observations if the insurer writes more than 25 percent of premiums for surety and credit or if the insurer writes more than 25 percent of premiums for reinsurance, accident and health, or workers' compensation.10 From the remaining set of 15,144 observations, we delete 2,873 cases where the insurer lacks sufficient data to estimate the models described in the ''Results'' section, 1,148 cases lacking auditor data, and 104 cases where the insurer is exempt from audit.
To isolate the influence of state-level liability standards, we limit the analysis to private insurance companies that are not subject to Securities and Exchange Commission (SEC) purview. Thus, we delete 3,515 observations where the insurer is either publicly traded or owned by a publicly traded entity.11 We assume that the insurer will be sued in its state of domicile, but recognize the potential for ''forum shopping'' by third parties seeking alternative venues where there is a greater chance that the court will find that the auditor owes them a duty of care. To minimize this possibility, we delete insurers that are licensed to write policies outside of their state of domicile. This eliminates 4,138 observations from the sample.12 Although borrowing by property- casualty insurers is rare, we also delete 259 cases where the insurer has outstanding debt to eliminate creditors as a potential third-party claimant.
The final sample consists of 3,107 observations from private property-casualty insurance companies that operate in a single state and have loss reserves subject to managerial discretion. Sample observations comprise 9.68 percent of total firm-year observations for the U.S. propertycasualty industry during 1993-2004. On average, sample insurers reported $84.68 million in admitted assets and $27.28 million in net premiums written during the sample period (untabulated). This compares to U.S. property-casualty industry averages for admitted assets and net premiums of $386.93 million and $116.10 million, respectively. The relatively small size of the insurers in our sample reflects the fact that these are private firms that are not licensed to write policies outside their state of domicile. As noted by Hope and Langli (2010), although the majority of companies worldwide are not publicly listed, extant research on auditor independence focuses almost exclusively on public companies. The insurance data allow a rare glimpse into auditor independence issues for private clients. The sample selection process is detailed in Table 2.
Variable Definitions and Descriptive Statistics
Our hypotheses specify four key variables. For each insurer-year we must estimate the loss reserve estimation bias and the insurer's financial health, the auditor, and the state of domicile. In this section, we describe the measurement of these variables. Descriptive statistics are reported in Table 3.
We use a five-year development period to determine loss reserve bias, which is consistent with Petroni (1992), Petroni and Beasley (1996), Beaver et al. (2003), and Gaver and Paterson (2004, 2007). To compute the loss reserve bias for an observation in 1993, for example, we subtract the original loss reserve reported in the 1993 annual statement from the five-year developed reserve reported in 1998. We then divide the result by the admitted assets reported in the 1992 annual statement to control for variation in insurer size.13 The bias is positive if the manager initially underreserved, and vice versa. As reported in Table 3, Panel A, the median reserve bias (BIAS) is ,2 percent of lagged admitted assets, indicating initial overstatement of the reserve. Gaver and Paterson (2004, 2007) also note a general tendency for property-casualty insurers to overstate reserves. Beaver et al. (2003) explain that the claim loss provision is tax deductible. Thus, financially strong insurers have incentives to over-reserve to minimize the present value of tax payments (Gaver and Paterson 1999). Over-reserving also provides a credible signal of insurer financial strength (Petroni 1992), and smooths income in unusually profitable years (Petroni et al. 2000). Appendix A explains that most IRIS ratios are improved by understating the loss reserve. Gaver and Paterson (2007) find that managers deliberately understate reserves to avoid IRIS ratio violation. Almost two-thirds of the firms in their sample that violated four or more IRIS ratios on a pre-managed basis held the number of violations to less than four on a reported basis. The important point for our study is that using reported results to compute IRIS ratios is likely to misclassify some financially struggling insurers as non-WEAK. Thus, following both Beaver et al. (2003) and Gaver and Paterson (2007), we ''back out'' the effect of the loss reserve bias on the financial results before computing the ratios. In other words, we restate the financial statement items as if no bias is present, and then compute the IRIS ratios using these ''unbiased'' financial statement numbers. For example, the 1993 reserve bias for GE Mortgage Insurance Company (based on the 1998 developed reserve) is $116.512 million. Understated reserves result in understated losses and liabilities. Thus, we compute the unbiased losses and liabilities of GEMIC by adding $116.512 to the reported amounts. Understated reserves affect policyholder surplus on an after-tax basis. We assume a marginal federal tax rate of 35 percent and compute the unbiased surplus by subtracting [$116.512 3 (1 - 0.35)] from the reported surplus.14
We classify an insurer as financially weak (WEAK) if it has more than three (unbiased) ratios outside of the bounds considered normal by the NAIC. We select this cutoffbecause it usually triggers regulatory attention (Belth 1987; Petroni 1992; NAIC 1994; Troxel and Bouchie 1995; Gaver and Paterson 2004). As shown in Panel A of Table 3, around 14 percent of the sample meets this criterion. Consistent with prior research, financial weakness is associated with reserve understatement. Panel B reports that the mean reserve bias for the 442 observations from weak insurers is about 17 percent of lagged admitted assets (indicating under-reserving), and the corresponding mean for the 2,665 observations from healthy firms is ,5 percent (indicating over-reserving). This difference is significant at the 0.0001 level. Panel C partitions the weak subsample into high (LITRISK = 1; n = 245) and low (LITRISK= 0; n= 197) litigation groups. Consistent with H1, under-reserving among financially struggling insurers is significantly lower when litigation risk is high (p < 0.0001).
The name of the auditor in each of the years 1993 through 2004 is hand-collected from either the insurer's statutory annual statement or Best Insurance Reports. We find that Big 6 auditors are used in 1,302 of our 3,107 insurer-years, with non-Big 6 auditors used in the remaining 1,805 cases (untabulated). The sample includes 513 unique firms, of which 87 are coded as weak in one or more sample years. Sample insurers are domiciled in 49 states, with the largest concentrations in New York (52 insurers), Pennsylvania (48 insurers), Texas (37 insurers), Florida (36 insurers), California (24 insurers), and Illinois (23 insurers).15
Reichelt and Wang (2010) report that audit quality is higher when the auditor is both a national and city-specific industry specialist, arguing that auditors' national network synergies and individual auditors' deep industry knowledge at the office level are jointly important factors in delivering higher audit quality. Accordingly, EXPERT is a dummy variable that takes on the value of 1 if the insurer's auditor is an insurance industry expert at both the national and city levels, and is 0 otherwise. In our sample, Ernst & Young and Coopers & Lybrand (PricewaterhouseCoopers from 1998-2004) have the two largest market shares of the national insurance market and are designated national experts. To be a city expert, an audit firm must have the largest market share of the insurance industry in its city by at least 10 percent.16 We find that industry-expert auditors (at both the national and the city levels) are used in 921 of our 3,107 insurer-years, with non-expert auditors used in the remaining 2,186 cases (untabulated).
Pacini et al. (2000) construct a liability index to quantify the stringency of the auditor liability standard in each state. This index, which is described in detail in Appendix B, ranges from 1 (for strict privity, the lowest liability case) to 9 (for reasonable foreseeability, the highest liability case). In our empirical tests, we convert the liability index to a dummy variable (LITRISK) to achieve a more compact representation of liability. LITRISK takes on the value of 1 if the liability index in the insurer's state of domicile is equal to or exceeds 4.0 (the high liability case), and is 0 otherwise. We choose 4.0 as our cutoffbecause it provides a clear demarcation between privity (low liability) and Restatement/ reasonable foreseeability (high liability). LITRISK has a mean of 0.58.17
Our model for testing hypotheses about the association between auditor legal liability and under-reserving by weak insurers, described in the following section, includes eight control variables intended to capture influences on reserve bias that are unrelated to insurer financial condition or auditor litigation risk. The variables are: OVER 3 LENGTH, UNDER 3 LENGTH, MAL, MUTUAL, TAX, STATETAX, NI_SMOOTH, RATE_REG, REG_BUDGET, and REINSURE. Although we defer discussion of the control variables until the ''Results'' section, their descriptive statistics are included in Table 3.
The Association between Auditor Liability and Reserve Bias
Under H1, the magnitude of the reserve understatement by financially weak property-casualty insurers is lower when auditors face higher state-level litigation risk. To test this hypothesis, we estimate the following model:
A variant of this model originally appeared in Petroni and Beasley (1996), and was later modified by Gaver and Paterson (2007).
The variables of interest in Equation (1), BIAS, WEAK, and LITRISK, are defined in the ''Data'' section. The first three control variables (OVER 3 LENGTH, UNDER 3 LENGTH, and MAL) are used by both Petroni and Beasley (1996) and Gaver and Paterson (2007) to control for product mix. LENGTH is claim loss reserves expressed as a percentage of total liabilities. It is included because the longer the claim cycle, the more difficult it is to forecast total claims. Petroni and Beasley (1996) find that insurers with long-tailed product lines tend to have more pronounced reserve errors of both signs. Following them, we parse LENGTH into two variables: (UNDER 3 LENGTH) and (OVER3LENGTH). UNDER is an indicator variable that takes on the value of 1 if the reserve error is positive (indicating initial understatement), and 0 otherwise. The companion variable, OVER, takes on the value of 1 if the reserve error is negative, and 0 otherwise. MAL is the percent of net premiums written for malpractice insurance. Both Petroni (1992) and Petroni and Beasley (1996) find that reserve bias is associated with malpractice writings, although the sign of the relationship varies across years.18 We add a fourth product-line variable, MUTUAL (coded as 1 if the insurer is organized as a mutual, and 0 if the insurer is a stock company), because Mayers and Smith (1988) and Lamm-Tennant and Starks (1993) find that differences in insurer ownership structures lead to differences in the lines each type of insurer tends to underwrite. Compared to mutual insurers, stock insurers tend to write more insurance in lines with higher risks and higher managerial discretion.19
The remaining control variables (TAX, STATETAX, NI_SMOOTH, RATE_REG, REG_BUDGET, and REINSURE) capture additional influences on reserve bias. TAX is coded 1 if the insurer either paid taxes or received a refund in the observation year, and is 0 otherwise. We expect TAX to be negatively associated with reserve bias, reflecting the incentive for insurers in higher tax brackets to over-reserve (Petroni 1992; Gaver and Paterson 1999; Beaver et al. 2003). STATETAX is a qualitative variable coded as 1 if the insurer is domiciled in Connecticut, Florida, Kansas, Indiana, Illinois, New Hampshire, New York, Mississippi, or Oregon. Insurers in these states pay both federal and state taxes on income and, therefore, have enhanced incentives (relative to insurers that do not face state income taxes) to overstate reserves. Thus, we also expect a negative coefficient on STATETAX. NI_SMOOTH is the current year's unbiased net income minus the previous year's reported net income, divided by the absolute value of the previous year's reported net income. Managers wishing to smooth results will under-reserve in lean years and over-reserve in fat years. We, therefore, predict a negative relation between NI_SMOOTH and reserve bias (Grace 1990; Beaver et al. 2003).
Two variables are used to control for variation in regulatory stringency and the ability of states to scrutinize insurers. RATE_REG is coded as 1 if the insurer is domiciled in a state with stringent rate regulations defined as state-made rates, and prior approval rate regulation rules (Harrington 2002; Grace and Leverty 2010). It is coded as 0 otherwise. Insurers facing rate regulation have less opportunity to under-reserve, which suggests a negative relation between RATE_REG and reserve bias. BUDGET_REG is the state of domicile's insurance department budget divided by the number of insurers domiciled in the state. It is used as a proxy for state regulatory stringency (Petroni and Shackelford 1995). Even though a state may have the legal authority and inclination to closely monitor the reserving practices of insurers domiciled in its borders, its ability to do so may be inadequate. State insurance departments with fewer resources are expected to be less rigorous, which suggests a negative relation between REG_BUDGET and reserve bias. Our last variable controls for the insurer's use of reinsurance. Prior to our sample period, insurers used reinsurance transactions to manage their reported results (Adiel 1996). The effectiveness of reinsurance to achieve reporting goals was diminished by SFAS No. 113 (effective for fiscal years beginning after December 15, 1992), which eliminated the practice of retrospective reinsurance. Using data from 1990-1997, Grace and Leverty (2010) find that insurers hide underreserving with reinsurance, and it is possible that there is a lingering reinsurance effect in our sample period, as well. We, therefore, include REINSURE in the model, computed as the amount of reinsurance ceded divided by the direct premiums written. We predict a negative sign on REINSURE because reinsurance lessens insurers' incentives to under-reserve.
The test of H1 rests on b3, the coefficient on the interaction between financial condition and litigation risk. A significantly negative value means that the tendency of weak insurers to underreserve is attenuated when the audit is conducted in a state that imposes high litigation risk on the auditor. This would indicate that auditors are more likely to find and less likely to allow reserve understatements by weak insurers in high-liability jurisdictions. We make no prediction concerning b2, the coefficient on litigation risk. H1 states that reserve understatement is restricted to firms with the incentive to manipulate; in other words, the weak firms. It makes no prediction about the direct relation between litigation risk and reserve error. In other words, privity laws only play a role in a high-risk audit, and the financially troubled insurers are high-risk auditees.
We estimate Equation (1) adjusting for clustered standard errors across both firms and time.20 Results for the sample of 3,107 insurer-years described in the ''Data'' section are reported in Table 4. The model is highly significant, with an average F- statistic of 123.24 and an average adjusted R2 of 0.34. Consistent with Petroni (1992), the coefficient on WEAK is significantly positive (p = 0.002), indicating that financially struggling insurers understate their loss reserves. The coefficient on litigation risk (LITRISK), on the other hand, is insignificant (p = 0.412), indicating that there is no direct relation between insurer reserve bias and auditor liability to third parties. However, the coefficient on the interaction between financial condition and LITRISK is significantly negative (p =0.022). This suggests that weak insurers are less likely to under-reserve if their auditor is subject to higher expected litigation costs, a finding that supports H1. The sum of the coefficients on WEAK and WEAK3LITRISK is significantly positive (p , 0.0001, untabulated), indicating that although under-reserving by weak insurers is reduced in high- liability states, it is not eliminated.
The coefficients on the control variables in Table 4 are not uniformly significant. Of the product mix variables, OVER3LENGTH and UNDER3LENGTH are significant in the predicted directions (p L 0.03 for both coefficients), but MAL and MUTUAL are insignificant (p=0.224 and p=0.484, respectively). This suggests that the relationship between an insurer's product mix and its reserve error is primarily determined by length of the loss development period. The coefficient on NI_SMOOTH is significantly negative (p , 0.0001), as predicted, which suggests that insurers under-reserve in lean years and over-reserve in fat years. The coefficient on RATE_REG is also significantly negative (p = 0.008), implying that insurers facing rate regulation have less opportunity to under-reserve. On the other hand, the coefficient on REG_BUDGET is insignificant (p = 0.702), as are the coefficients on the tax variables, TAX and STATETAX (p = 0.975 and p = 0.669, respectively). Finally, the coefficient on REINSURE is significantly negative (p = 0.049). The ability to transfer risk via reinsurance transactions seems to reduce insurers' incentives to under-reserve, even after SFAS No. 113 reduced their ability to use this strategy retroactively.
H2 and H3 concern the association between auditor type and under- reserving by weak insurance clients in high-liability states. H2 posits that the reserve understatement by financially weak insurers in high-liability states is less pronounced when insurers have Big 6 auditors. H3 predicts that the reserve understatement by financially weak insurers in high-liability states is less pronounced when insurers have auditors who are insurance industry experts. To test these hypotheses, we estimate Equations (2) and (3) for the subsample 1,810 sample observations from high-liability (LITRISK=1) states. Among this subsample, there are 956 observations with a Big N auditor and 617 observations where the auditor is an insurance- industry expert at both the national and the city levels. Our estimation technique adjusts for clustered standard errors across both firms and time (Petersen 2009).
The results for Equation (2) are presented in Column 3 of Table 5. As in Table 4, the coefficient on WEAK is significantly positive, indicating that financially struggling insurers understate their loss reserves. The coefficient on BIG6 is insignificant, indicating that there is no direct relation between insurer reserve bias and auditor size for insurers domiciled in high litigation risk states. This result is consistent with Kim et al. (2003), who find that Big 6 auditors are only more effective than non-Big 6 auditors when managers have incentives to prefer income-increasing accrual choices. In our setting, it is the financially weak insurers that prefer under-reserving, an income-increasing accrual choice. Accordingly, the coefficient on the interaction between financial condition and BIG6 is significantly negative. Consistent with H2, weak insurers domiciled in high-liability states are less likely to under-reserve if their auditor is a Big 6 firm. The coefficients and significance of the control variables echo the results in Table 4.
Results for the test of H3 are reported in Column 4 of Table 5. The results are qualitatively the same as in Column 3. Just as the coefficient on WEAK3BIG6 (Column 3) is significantly negative, so too is the coefficient on WEAK 3 EXPERT (Column 4). Thus, H3 is also supported. Weak insurers domiciled in high-liability states are less likely to under-reserve if they have an industryexpert auditor. Auditors are more likely to identify manipulations and demand adjustments in the loss reserves of weak insurance clients when the audit is conducted in a state that imposes a high degree of legal liability on the auditor. The effect is more pronounced for Big 6 auditors and auditors that are insurance industry experts.21
The principal finding of the paper is that the tendency of weak insurers to under-reserve is attenuated when the insurer is domiciled in a state that imposes greater legal liability on the auditor. These results are robust to several perturbations in research design. For example, Petroni (1992) argues that three reserve ratios are designed to identify insurers that are under- reserved, and including them in the identification of financially troubled insurers could induce a bias in favor of finding that weak firms under-reserve. She identifies financially troubled insurers as those that have more than one unusual non-reserve ratio without adjusting the ratios for the ex post estimation errors. Redefining weak firms using this approach does not alter our inferences. Neither does changing the threshold for WEAK (using all IRIS ratios on a pre-managed basis) to three or more unusual IRIS ratios. The same is true for alternative liability index cutoffs (J 3.5; J 4.5) to code LITRISK. We also reestimate the models by year, because Petroni (1992) finds that malpractice insurance reserves are systematically understated in some years and overstated in others. The coefficient on MAL is significantly negative (p , 0.05) in three years and insignificant in the other nine years.22 Our primary findings are unchanged.
In Table 4, we define auditor legal liability using the legal standard in the insurance client's state of domicile and limit the potential for forum shopping by restricting the sample to insurers that only write policies in the domicile state. However, the results are qualitatively unchanged when we estimate Equation (1) for the subset of 1,530 observations where the client headquarters and the audit practice office are also located in the state of domicile.23 For this subsample, forum shopping is virtually impossible. We also explore the possibility that the negative sign on b3 reflects the fact that weak insurers domiciled in high liability states are ''less weak'' than the other weak firms and, therefore, less likely to under-reserve. In the full sample of 3,107 observations, 442 are categorized as weak. We partition the weak sample according to LITRISK, and find no difference between two subgroups in terms of the amount of under-reserving needed to avoid reporting four or more unusual IRIS ratios (p = 0.2835).24 This suggests that the degree of financial distress among the weak firms is not related to their domicile state. We, therefore, reject the notion that weak insurers in states with high auditor liability risk are actually ''less weak'' than the other weak firms, and for that reason are less likely to under-reserve. Instead, we believe that auditors in high- liability states are less likely to allow reserve understatements, because the higher liability standard increases the cost of an audit failure.
The notion that the litigation risk affects auditor behavior and, by implication, audit quality, is well established in the auditing literature. However, extant research focuses on the impact of federal-level statutory law on audit quality, and largely ignores the auditor's liability to third parties under state-level common law. This paper provides the first evidence that state-level liability standards affect auditor behavior. We are able to isolate the impact of state liability standards on audit quality because we limit our sample to private firms that are not subject to federal statutory law. We can also pinpoint the relevant state liability standard because our sample insurance firms only operate in a single state.
The principal focus of property-casualty audits is the adequacy of the loss reserve, with particular attention to financially struggling clients that have incentives to under-reserve. Under- reserving by weak insurers is well documented and occurs in our sample. Our contribution is to explore whether this behavior is reduced when the insurance client is domiciled and exclusively licensed in a state that allows a broader class of third parties to sue the auditor for negligence. We find reduced under-reserving among weak insurers in states that use either the Restatement of Torts or the reasonable foreseeability standard to determine the auditor's liability to third parties. Compared to the case where the auditor's liability is defined by the legal concept of privity, these standards impose greater legal costs on auditors for ordinary negligence. We also find that under-reserving by weak insurers in non-privity states is lower when the auditor is a Big 6 firm and when the auditor is an insurance industry expert.
States differ not only by the liability standard established for auditor negligence, but also by regulatory stringency. However, our data show that empirical measures of these constructs are not significantly correlated. This suggests that our results are not an artifact of the impact of regulation, which has also been shown to influence the reserve choices of weak insurers. Auditor legal liability to third parties under common law, as defined by the state of domicile, independently affects this behavior. Our results reinforce the notion that auditor independence is enforced by the threat of litigation against the auditor, and that variation in perceived litigation risk across states is important. The fact that differential client conservatism is observed for audits conducted in different states by the same audit firm suggests that cross-state differences in auditor liability regimes have an incremental influence on client conservatism beyond auditor incentives for reputation protection.
Copyright American Accounting Association Aug 2012
(c) 2012 Auditing: A Journal of Practice & Theory. Provided by ProQuest LLC. All rights Reserved.