Auditor Conservatism and Audit Quality: Evidence from IPO Earnings Forecasts Philip J. Lee, 1 Sarah J. Taylor 2 and Stephen L. Taylor 3 1 University of Sydney, Australia 2 University of Melbourne, Australia 3 University of New South Wales and Capital Markets CRC Ltd, Australia We investigate the relation between a proxy for differential audit quality and both the (ex post) accuracy and conservatism of audited earnings forecasts provided in Australian initial public offering (IPO) prospectuses. For the period we examine, most Australian IPO prospectuses include an earnings forecast (i.e., disclosure is not ‘voluntary’), and the auditor must be satisfied prior to signing off on the prospectus. After controlling for other factors associated with forecast error, there is some evidence that forecasts audited by Big 6 auditors prove more accurate than those audited by a non-Big 6 auditor, although this result is not robust across alternative measures of forecast accuracy. In contrast, our finding of significantly less optimistic bias for forecasts associated with Big 6 auditors is robust to alternative measures of forecast bias. We interpret these results as being consistent with the argument that the economic demand for differential audit quality reflects the same factors that underlie the demand for conservative financial reporting. Key words: audit quality, initial public offering, management earnings forecasts, conservatism. SUMMARY Audit quality is not easily observed, and as a result researchers have increasingly relied on measurable attributes of audited financial statements. This is premised on the assumption that the quality of financial statement data is a joint function of management representations and the audit process. However, we observe that many prior studies rely on the assumption that higher quality financial statement data (and by inference, higher quality auditing) will be reflected in greater accuracy. One such example is the attempt to link proxies for differential audit quality (e.g., Big 6 auditors) with the absolute value of unexpected accruals. On the other hand, extant research also suggests that higher quality auditors may be associated with more conservative financial reporting. We draw attention to the potential conflict between the accuracy and conservatism of audited financial data, as conservatism implies a directional bias. Such a bias implies that attempts to link audit quality with accuracy may be confounded. Correspondence to: Stephen L. Taylor, School of Accounting, University of New South Wales, NSW 2052, Australia. Email: S.Taylor@unsw.edu.au International Journal of Auditing Int. J. Audit. 10: 183–199 (2006) ISSN 1090-6738 © 2006 The Author(s) Journal compilation © 2006 Blackwell Publishing Ltd, 9600 Garsington Rd, Oxford OX4 2DQ, UK and Main St., Malden, MA 01248, USA. We test our theory on a relatively unique setting, namely the quasi-compulsory provision of audited earnings forecasts by Australian initial public offerings (IPOs). By comparing the forecast result with the actual result subsequently reported, we are able to directly test the competing theories that high quality auditing (in this case, Big 6 auditors) are associated with either more accurate or more conservative earnings forecasts. We control for several other factors expected to be associated with forecast accuracy and/or bias. Our evidence also extends prior research which has relied on measuring attributes of earnings forecasts made by IPOs in environments where the provision of a forecast is voluntary, rather than mandatory. By examining mandatory forecasts, we effectively control for factors that may be associated with forecast accuracy and/or bias but which are also determinants of the decision to voluntarily provide such a forecast in the first place. Our results support the view that high quality auditing is associated with more conservative reporting. Although we find some evidence that forecasts audited by Big 6 auditors are more accurate than forecasts audited by non-Big 6 auditors, this result is not robust to alternative ways of measuring the forecast error. On the other hand, evidence that forecasts audited by Big 6 auditors are more conservative proves to be highly robust. We interpret our evidence as providing support for the argument that the derived demand for conservative financial reporting is also reflected in the demand for high quality auditing, and so audit quality is associated with conservative reporting. 1. INTRODUCTION It is widely accepted that auditors add value to financial statements by reducing the likelihood of deliberate misreporting (Watts & Zimmerman, 1986). However, although this suggests that higher quality auditing should be associated with more accurate (i.e., less biased) financial reporting, we also note that at least one form of bias in financial reporting, namely conservatism, has also been argued to be associated with the quality of financial reporting (Watts, 2003). Because conservative accounting can facilitate the monitoring role of financial reporting data (Ball & Shivakumar, 2005), it is not surprising that the application of conservatism within financial reporting has been argued to have evolved in conjunction with the demand for independent verification by external auditors (Watts, 2003). Indeed, recent criticisms of the accounting profession (e.g., Levitt, 1998), and especially the controversy over auditor independence, focus almost exclusively on alleged overstatements of periodic results (Ruddock et al., 2006). This further highlights the extent to which auditors are presumed to ensure a certain degree of conservatism in audited financial reports. We investigate the extent to which a widely used proxy for differential audit quality (i.e., Big 6 versus non-Big 6) is associated with more precise and/or less optimistically biased financial data. 1 Our analysis is motivated by recognition that accuracy (i.e., precision) and conservatism (i.e., less optimistic or ‘downward’ bias) are not the same, and in fact are competing, rather than complementary attributes of financial reporting data. Put simply, if higher quality auditing is associated with relatively more accurate financial reporting, we would not expect to observe an association between a proxy for audit quality and a consistent conservative bias. Following the arguments in Watts (2003), we expect that conservatism associated with the use of Big 6 auditors is particularly valuable (and hence, most likely to occur) where significant informational asymmetries are present, and where audited financial information is likely to be relied on. We therefore examine the association between a proxy for audit quality and both the accuracy and bias (conservatism) in financial reporting data in such a setting. The setting we examine is the provision of earnings forecasts in the prospectuses of Australian initial public offerings (IPOs). Earnings forecasts provided in Australian IPO prospectuses provide a unique setting in which to examine the effect of differential audit quality. Securities regulations and the threat of litigation result in almost no disclosures of forward looking financial information by United States IPOs. In contrast, relevant Australian securities regulations, operative since 1991, are widely viewed as making the provision of earnings forecasts ‘de facto’ mandatory for the majority of Australian industrial IPOs (i.e., excluding mining IPOs), despite the relatively imprecise wording of the legislation. 2 This influence is exacerbated by the relatively severe penalties imposed by the Corporations Law for the omission of ‘material information’ known to the issuer. For the sample of IPOs we initially 184 P. J. Lee et al. Int. J. Audit. 10: 183–199 (2006)© 2006 The Author(s) Journal compilation © Blackwell Publishing Ltd. 2006 identify, almost 85% provide an explicit forecast of expected earnings. This reduces the extent to which any endogenous disclosure decision potentially affects our tests relative to those that utilize less frequent, voluntary disclosures. 3 Auditors also have less flexibility in reporting on prospectus data than for periodic audits. Even when the auditor does not provide an explicit attestation to an earnings forecast, Australian professional standards prohibit (and regulatory guidelines reinforce) the auditor from signing the prospectus if there are reservations about any aspect of this document. They cannot provide a ‘qualified’ response. 4 We expect that the inability to issue a qualified audit opinion (and thereby signal uncertainty) will exacerbate the role of conservatism as an attribute of audit quality. Our expectation is that Big 6 auditors are at least as concerned with avoiding reputation costs that arise when forecasts (or, more generally, assumptions implicit in historic audited results) prove to be optimistic as they are with ensuring the precision of the information to which they attest. Likewise, we expect that the demand for audit quality incorporates at least some expectation of increased conservatism. Hence, attempts to identify a relation between use of a Big 6 auditor and forecast accuracy (i.e., the unsigned forecast error) may be inconsistent with the underlying demand for audit quality, and may be confounded by the expected greater conservatism of Big 6 auditors. Conversely, we would unambiguously expect to find that forecasts audited by Big 6 auditors prove to be more conservative, ex post, than those audited by non-Big 6 auditors. Our results are broadly consistent with these predictions. Univariate tests show significant differences when comparing measures of both forecast accuracy and bias for Big 6 and non-Big 6 auditees. After controlling for other factors expected to be associated with forecast accuracy we find some evidence that forecasts attested to by Big 6 auditors are more accurate. However, this result is not robust to alternative measures of forecast accuracy. In contrast, we find consistent evidence of greater conservatism (i.e., a lower optimistic bias) among forecasts issued by IPO firms with Big 6 auditors, irrespective of the forecast error metric used. Our paper contributes in a number of ways. First, we provide additional evidence of how auditors, and differential audit quality, may add value specifically in the IPO process. Although it has been argued that the choice of a Big 6 auditor can serve as a signalling mechanism for IPO firms (Datar et al., 1991; Titman & Trueman, 1986), the process by which auditors add value in such a setting has been subject to relatively little empirical analysis. 5 Although a few studies examine earnings forecasts voluntarily provided by Canadian IPOs (Davidson & Neu, 1993; McConomy, 1998; Clarkson, 2000), we argue that the conflicting results in these studies reflect a failure to consider the incentives which high quality auditors face in attesting to accounting projections. The results of these studies are inconsistent and sensitive to the exact model of forecast error used (Clarkson, 2000). More importantly, our sample largely avoids the potential problem of endogenous voluntary disclosure faced by these studies. Second, we provide additional evidence consistent with conservatism being one element of auditor behaviour that underlies product differentiation in auditing. Although a limited number of studies examine the relation between proxies for differential audit quality (i.e., Big 6) and the output from the accrual accounting process, these studies are limited to observing unusual (i.e., unexpected) accruals, or rely on proxies for identifying news dependent conservatism in accounting. 6 In contrast, our use of audited earnings forecasts allows us to directly measure the extent of ex post accuracy and conservatism, and consider the extent to which these are competing objectives. We also examine the association between audit quality and attributes of audited information in an environment where expected litigation costs are likely much lower than in the United States. Although expected litigation costs may be an important factor in creating a demand for differential audit quality (Basu et al., 2001), we also expect that the value of reputation effects, and the underlying economic demand for conservatism will result in evidence of an association between audit quality and conservatism in a relatively low litigation environment. The remainder of the paper proceeds as follows. Section 2 reviews prior evidence on the relation between differential audit quality and conservative financial reporting, and generates testable hypotheses. Section 3 describes our data sources, as well as providing evidence on the accuracy and bias of earnings forecasts provided in IPO prospectuses. Section 4 reports our primary results, while Section 5 summarizes additional sensitivity analysis. Section 6 concludes. Auditor Conservatism and Audit Quality 185 Int. J. Audit. 10: 183–199 (2006)© 2006 The Author(s) Journal compilation © Blackwell Publishing Ltd. 2006 2. BACKGROUND AND HYPOTHESES 2.1. Evidence of a link between audit quality and reporting conservatism As we have already noted, the underlying contracting and informational demand for financial reporting is also consistent with a demand for conservatism, at least in so far as that conservatism is a reflection of how the financial reporting process reacts to new (or revised) information. This is what Basu (1997) terms ‘news based’ conservatism, and what Ball & Shivakumar (2005) describe as ‘conditional’ conservatism. At the same time, it is logical to expect that the external verification process (i.e., external auditors) will pay heed to this demand for conservative reporting. Although the auditor does not bear legal responsibility for compiling the accounts, it is beyond dispute that they are expected to influence the outcome. Hence, it is expected that at least one dimension of what is perceived as audit quality will be the extent to which an ‘appropriate’ level of conservatism is enforced by external auditors. Evidence consistent with the conjecture that audit quality is associated with increased conservatism can be found in a number of forms. First, there is evidence that the accrual component of earnings, or at least the unexpected component thereof, is inversely related to the common auditor size-based proxy for audit quality. Several studies focus on the link between audit quality (proxied by Big 6) and the accrual component of earnings. 7 Francis et al. (1999) demonstrate that the decision to use a Big 6 auditor is positively related to firms’ endogenous propensity to generate accruals, as proxied by the length of their operating cycle (current accruals) and their capital intensity (non-current accruals). Among studies that examine the link between Big 6 auditor choice and accruals, the most consistent result is that Big 6 auditors are more conservative than their non-Big 6 counterparts. Becker et al. (1998) show that firms using Big 6 auditors typically have lower unexpected accruals than other firms. 8 DeFond & Subramanyam (1998) report that firms switching from a Big 6 to a non-Big 6 auditor appear to implement more liberal accounting, as evidenced by higher unexpected accruals. However, the method used for estimating unexpected accruals has been shown to have relatively low power in identifying the unexpected component (Dechow et al., 1995; McNicholls, 2001). As neither paper identifies any specific incentive for managers to exercise their discretion, interpretation of the results is problematic. Second, evidence of auditor conservatism is evident in auditor reporting decisions. Francis & Krishnan (1999) examine the relation between audit firms’ propensity to issue modified audit reports and the extent of accruals. They model the decision to issue a modified report, and find that the probability of a modified report increases with the (absolute) level of accruals. However, the result is strongest for firms with large positive accruals. When firms are partitioned into those with Big 6 auditors and others, the relation between accruals and the propensity to issue modified audit reports is confined to Big 6 auditors. This is consistent with Big 6 auditors being more conservative than non-Big 6 auditors. Third, there are studies that examine the extent to which earnings incorporate economic losses on a more timely basis than economic gains. Basu et al. (2001) show that the asymmetric timeliness of bad news in earnings, as reflected in unexpected stock returns, is significantly greater for Big 6 auditees than others. 9 This result largely reflects the effect of more conservative operating accruals, rather than extraordinary items or discontinued operations. Basu et al. also show that negative earnings changes are less persistent for Big 6 auditees than other firms, which is consistent with greater conservatism by Big 6 audit firms. However, tests such as these rely on the identification of news based conservatism from either contemporaneous share price movements or from the time series behaviour of earnings. We adopt a broader, but nevertheless related perspective of conservatism, and expect that forward looking financial information attested by high quality auditors will prove ex post to be relatively more conservative than otherwise. This is consistent with the view that conservative forecasts will be less likely to anticipate gains than losses, and so by deliberate understatement of the forecast and/or delayed recognition of gains (or even expected gains) until the first post-forecast result, the forecast result may provetobeex post conservative. Whether this results in reduced accuracy is an empirical question that we also address. 2.2. IPO earnings forecasts and audit quality Apart from the various methods discussed above of identifying conservatism as one dimension of 186 P. J. Lee et al. Int. J. Audit. 10: 183–199 (2006)© 2006 The Author(s) Journal compilation © Blackwell Publishing Ltd. 2006 audit quality, there are also a small number of studies that examine properties of Canadian IPO earnings forecasts and their relation to either the type of audit requirement and/or the identity of the auditor. These studies reflect the frequent voluntary provision of earnings forecasts at the time of an IPO. McConomy (1998) argues that auditing is expected to reduce the extent of any positive bias which (otherwise unaudited) information is likely to display. McConomy compares the ex post accuracy and bias of earnings forecasts prior to a 1989 requirement that the forecasts be audited, rather than reviewed, and finds a significant reduction in optimistic bias, but relatively little improvement in accuracy. Our interpretation of these results is that auditors adopt a more conservative approach as their degree of responsibility increases, although it is also possible that firms most likely to provide optimistic forecasts elected not to do so after the introduction of the audit requirements. Using a sample of Canadian IPOs from 1983–87, Davidson & Neu (1993) show that earnings forecasts reviewed by Big 6 auditors are significantly less accurate than those reviewed by non-Big 6 auditors. They explain this result as a product of less post-listing earnings management by Big 6 auditees. In effect, Davidson & Neu assume that differential audit quality has no direct effect on the quality of earnings forecasts, but at the same time does act to constrain opportunistic earnings management in the period following the IPO. Exactly why auditors would not care about earnings forecasts, yet actively intervene to constrain accounting policies is not clear, except that the review (as distinct from audit) requirement that applied to Canadian IPO earnings forecasts may effectively reduce the significance of auditor reputation. Another explanation for the result reported by Davidson & Neu (1993) is that the relation between differential audit quality and forecast properties may be sensitive to the choice of variables used to control for firm-specific and period-specific uncertainty. Clarkson (2000) examines forecast accuracy and bias in both the review (1984–87) and audit (1992–95) regimes, and shows that Davidson & Neu’s primary result is sensitive to the choice of variables used to control for business risk, which in turn is expected to affect forecast accuracy. When a similar test is performed on the audit regime sample, earnings forecasts audited by Big 6 auditors are significantly more accurate than others. For a measure of forecast bias, Clarkson finds no significant difference between those audited by Big 6 and non-Big 6 auditors, in either the audit or review regimes. 10 However, while the results reported by Clarkson (2000) suggest that forecasts audited by Big 6 auditors are significantly more accurate but not significantly less biased, we note that the decision by Canadian IPO firms to provide an earnings forecast is voluntary. Canadian IPO firms that hire non-Big 6 auditors are typically riskier (Clarkson & Simunic, 1994), and auditors may be relatively less willing to face possible adverse effects of attesting to forecasts by risky firms. Hence, the endogenous forecast decision potentially biases Clarkson’s tests towards finding that forecasts audited by Big 6 auditors are significantly more accurate. A better specified test is possible in an environment where the earnings forecasts are a ‘routine’ part of the prospectus, as they are for the Australian IPOs we examine. 11 2.3. Hypotheses Based on the evidence outlined above, we adopt the view that differential audit quality acts to constrain, or at least delay, relatively aggressive reporting practices. Hence, the primary evidence of audit quality effects is most likely to occur as greater conservatism, rather than improved accuracy. If high quality auditing results in the application of relatively conservative constraints, then users have less concern at possible ‘overstatements’ than otherwise. In the context of IPOs providing earnings forecasts at the time of going public, we expect that conservatism will be realized via forecasts which prove, ex post,tobe more conservative. Even if ex post evidence of conservative forecasts reflects some degree of upwards earnings management in the first post-listing result, this still reflects the deferral of ‘good news’ and/or more aggressive reporting to a later point than explicitly incorporating it in the forecast result. Moreover, it is hard to imagine that an auditor who attests to conservative forecasts would simply allow aggressive accounting in the subsequent result. High quality auditors also will likely face higher costs where they are found to have endorsed over-optimistic forecasts, as compared to ‘excessive’ conservatism. On the other hand, systematically conservative estimates of future results would be expected to reduce the overall level of accuracy that would Auditor Conservatism and Audit Quality 187 Int. J. Audit. 10: 183–199 (2006)© 2006 The Author(s) Journal compilation © Blackwell Publishing Ltd. 2006 arise from otherwise unbiased estimates, with the result that higher quality auditors may not be associated with forecasts which prove, ex post,tobe more accurate. Ultimately, the extent to which audit quality is associated with either conservatism or accuracy is an empirical question. Hence, we test the following two hypotheses, both of which are stated in the null form: H1: There is no association between audit quality and forecast accuracy. H2: There is no association between audit quality and forecast bias. 3. DATA 3.1. Sample Our sample begins with all Australian industrial IPOs between 1991 and 1998. Our data begins at 1991 to reflect the effect of changes to the Corporations Law. Mining IPOs are excluded, as they typically do not forecast earnings, primarily because this practice is actively discouraged for exploration firms. Trusts and pooled development funds are also excluded, due to the differences in operating structure and taxation treatment of dividends and capital distributions. The resulting sample of IPOs comprises 220 firms, of which 184 provide earnings forecasts with a horizon of at least 60 days. However, five of these firms were eliminated because the company was delisted prior to the end of the forecast period and/or the forecast could not be matched with actual results. Table 1 provides a summary of the temporal distribution and industry membership of the sample firms. Panel A shows that there is some temporal clustering, consistent with the existence of ‘hot’ and ‘cold’ issue markets. Panel B indicates that the sample is relatively evenly distributed across the five broad industry groupings, with the exception of Financial services, which has a lower representation. In order to test our hypotheses, a proxy for differential audit quality is required. Consistent with theory (DeAngelo, 1981) and an over- whelming amount of empirical evidence (Hay et al., 2006), we use the conventional Big 6/non-Big 6 distinction. Our focus on the most common method of identifying high quality auditors is also motivated by our desire to highlight the underlying tension between the effects of audit quality (as conventionally measured) on forecast accuracy as distinct from the effect of forecast bias (i.e., conservatism effects), rather than complicating the analysis with more equivocal proxies for differential audit quality. Other possible indicators of high quality auditors such as client industry specialization have mixed support. For example, although Craswell et al. (1995) report evidence that industry specialist auditors earn significant audit fee premiums, more recent evidence suggests that this premium was eroded as the audit market was consolidated from a Big 8 through to a Big 6 and, finally, a Big 4 (Ferguson & Stokes, 2002). This is consistent with a reduced number of large international audit firms making it more difficult for any one of those firms to be seen as an industry specialist, simply because the ‘random’ market share in each client industry increases as the number of competing audit firms declines. Table 1: Summary of temporal distribution and industry group membership details of 179 initial public offerings for the period January 1991–June 1998 Panel A: Temporal distribution of industrial ipos disclosing an earnings forecast Year Number of firms Per cent 1991 4 2.2 1992 19 10.6 1993 54 30.2 1994 36 20.1 1995 12 6.7 1996 20 11.2 1997 29 16.2 1998 5 2.8 Total 179 100 Panel B: Industry group distribution Group Number of firms Per cent Services 50 27.9 Construction & Development 39 21.8 Retail & Consumer/ Household Goods 29 16.2 Financial 11 6.1 Industrials 50 27.9 Total 179 100 Firms are classified as belonging to one of five industry groups which are formed based on Australian Stock Exchange Industry Classifications. These groups are: Services, Construction Development, Retail Consumer/Household Goods, Financial, and Industrials. Mining firms are excluded from the sample. 188 P. J. Lee et al. Int. J. Audit. 10: 183–199 (2006)© 2006 The Author(s) Journal compilation © Blackwell Publishing Ltd. 2006 3.2. Forecast errors We hand collect data, and so carefully match the forecast income number with the actual result. In many cases, forecasts are provided for several income definitions. When this occurs we match forecast operating profit before tax (OPBT) with actual. We prefer this measure relative to after-tax earnings because discussion in several prospectuses suggests that firms often forecast tax using the nominal corporate tax rate rather than the expected rate applicable to the calculation of income tax expense. 12 Where firms do not forecast OPBT, an alternative definition is used, either operating profit after tax but before abnormal and extraordinary items, operating profit after tax, or earnings before interest and tax. 13 In all cases, forecast error is the difference between forecast and actual, so that a positive forecast error indicates optimism. Descriptive data for forecast accuracy (i.e., absolute error) and forecast bias (i.e., signed error) are presented in Table 2. Earnings forecast errors are scaled two ways. First, we utilize issue size as a deflator. This gives a feel for the possible ‘economic significance’ of these forecast errors, relative to the funds raised through the IPO. However, earnings are for the firm as a whole, but the use of issue size as a deflator reflects only the interest of the ‘new’ shareholders. Hence, we also measure earnings forecast errors on a per share basis, where the deflator is the issue price per share. Both of the forecast error measures we report provide a more intuitively ‘economic’ measure than simple error percentages with respect to actual or forecast earnings, and most closely corresponds with the measures of forecast error (or ‘earnings surprise’) used in other studies. 14 From Panel A of Table 2, the absolute forecast error expressed relative to share price at the time of the offering has a mean (median) error of 4.93% (1.48%). Turning to the extent of possible bias, the data are consistent with forecast errors being, on average, optimistic. However, the median forecast error is negative which Table 2: Descriptive statistics of forecast accuracy (Panel A) and bias (Panel B) for 179 IPO firms Forecast accuracy is measured as the absolute value of (Forecast earnings less Actual earnings), while bias is measured as (Forecast earnings less Actual earnings). Reported forecast measures (accuracy and bias) are scaled by two alternate deflators (i) issue size and (ii) on a per share basis with the deflation by the issue price per share. This results in two measures of forecast accuracy: error as a percentage of issue size and per share error deflated by issue price. All figures are expressed as percentages. Panel A: Forecast error for full sample (n = 179) Mean Std. Dev Min. Median Max. Abs (F-A)/Issue size 13.65 27.56 0 3.18 246.15 Abs (F-A) per share/Issue price per share 4.93 8.37 0 1.48 61.48 (F-A)/Issue size 2.46 30.68 -246.15 -0.33 107.71 (F-A) per share/Issue price per share 2.00 9.51 -34.22 -0.14 61.48 Panel B: Forecast error for non-Big 6 auditees (n = 54) Mean Std. Dev Min. Median Max. Abs (F-A)/Issue size 19.08 29.47 0 4.78 107.71 Abs (F-A) per share/Issue price per share 7.76 12.20 0 2.29 61.48 (F-A)/Issue size 13.70 32.38 -41.08 0.21 107.71 (F-A) per share/Issue price per share 5.34 13.46 -15.41 0.12 61.48 Panel C: Forecast error for Big 6 auditees (n = 125) Mean Std. Dev Min. Median Max. Abs (F-A)/Issue size 11.31* 26.48 0.03 2.29* 246.15 Abs (F-A) per share/Issue price per share 3.71*** 5.65 0.02 1.08* 34.22 (F-A)/Issue size -2.40*** 28.72 -246.15 -0.38 65.40 (F-A) per share/Issue price per share 0.55*** 6.74 -34.22 -0.22 23.09 */**/*** = statistically significantly different from non-Big 6 auditees at 10%, 5%, 1% levels. Auditor Conservatism and Audit Quality 189 Int. J. Audit. 10: 183–199 (2006)© 2006 The Author(s) Journal compilation © Blackwell Publishing Ltd. 2006 indicatesthat there are actually more forecasts which are ex post pessimistic than optimistic (100/179). Expressed on a per share basis, the mean (median) signed forecast error measured as a percentage of the issue price is 2.00% (-0.14%). 15 Panel B of Table 2 contains descriptive statistics for accuracy and bias of non-Big 6 auditees while data for Big 6 auditees is contained in Panel C. Both measures of absolute forecast error (i.e., forecast accuracy) have means and medians that are significantly lower for Big 6 auditees. However, for signed forecast errors (i.e., forecast bias), only the mean is significantly lower for Big 6 auditees. This is consistent with the distribution of forecast errors for non-Big 6 auditees being more skewed, with a relatively larger proportion of forecasts that prove, ex post,to have large optimistic errors. However, given the numerous systematic differences between Big 6 and non-Big 6 auditees documented in Table 3, which may also be associated with the sign and size of forecast errors, we caution against relying on these univariate comparisons. 3.3. Control variables In order to identify the effect of differential audit quality on earnings forecast accuracy and bias, we regress measures of forecast error on our proxy for audit quality (i.e., a Big 6 dummy variable), a measure of underwriter quality and three composite control variables. Because we are interested in the incremental effect of differential audit quality rather than the determinants of forecast error per se, we use principal component analysis (Harman, 1976) to construct three composite control variables (‘factors’), which are intended to capture firm specific risk (FIRMRISK), forecast characteristics (FORECAST) and managerial incentives (INCENTIVES), respectively. 16 The objective is simply to establish a limited number of composite control variables where each composite variable comprises a number of measures that intuitively capture similar attributes. We do not use orthogonalization procedures to specifically minimize factor correlation. 17 Rather, we pre-select the components of each factor, and use principal component analysis to create the three factors purely to simplify the presentation of our analysis and to reduce the focus on individual determinants of forecast error and bias. Of course, where the components of a factor are highly correlated, there are some efficiency gains from simply maximizing the extent of the explained dispersion across this set of variables before attempting to explain the variation in forecast error or bias. 18 Principal component analysis allows us to isolate linear combinations of the potential control variables that are likely to capture similar aspects of the firm, its forecasting environment or the incentives to make more accurate and/or less biased forecasts. Hence, we construct an artificial variable (i.e, factor) that is an optimally weighted linear combination of the original variables. Each factor is the normalized linear combination of the assigned set of control variables with maximum variance. Importantly, all of our results with respect to the relation between audit quality and forecast accuracy or bias are robust to simply estimating a model with all of the individual control variables rather than the three factors as independent variables. Our selection of possible control variables (and the three resulting factors) is guided by prior evidence on the determinants of IPO earnings forecast accuracy and/or bias (e.g., Clarkson, 2000), as well as prior studies showing a relation between proxies for firm-specific risk and complexity that are ‘priced’ by auditors (Craswell et al., 1995). Due to uncertainty about the future and the inherent complexity of the firm’s operations, managers typically forecast earnings with some error. Possible proxies for firm specific risk and complexity include the age of the firm, firm size, leverage, the number of subsidiaries of the IPO firm, whether or not the firm has foreign operations, proportion of issue price not backed by net tangible assets (i.e., a measure of ‘growth options’ for the firm), 19 whether or not the firm had a loss in the previous three years and the number of risk factors highlighted in the IPO prospectus. We include each of these measures in our first estimated composite proxy, which we label FIRMRISK. 20 Many of these variables are also used to control for aspects of audit risk (i.e., number of subsidiaries) in audit pricing models, which also demonstrate evidence of differential audit quality (Craswell et al., 1995). In addition to the firm specific characteristics included in our composite FIRMRISK measure described above, forecast specific characteristics may also be associated with the size and/or sign of forecast errors. The length of time between the date of making the forecast and the end of the period to which it relates will affect the degree of confidence 190 P. J. Lee et al. Int. J. Audit. 10: 183–199 (2006)© 2006 The Author(s) Journal compilation © Blackwell Publishing Ltd. 2006 with which predictions can be made about the future. Accordingly, the forecast horizon, measured as the number of months between the prospectus date and the end of the forecast period, is included in our second composite proxy. We also control for the level of detail with which the forecast is made. A more detailed forecast likely reflects greater confidence in the accuracy of the forecast. Forecast detail is captured using a self constructed index, which awards a score from 1 to 9 based on how many of the following are disclosed in the forecast: earnings, revenues, expenses, capital expenditures, financing details, cash flows, dividends, assumptions used in forecasting and sensitivity analysis. 21 Finally, a dummy variable is used to indicate whether the firm provided forecasts for more than one financial year. Firms providing forecasts for multiple years are expected to be more confident about future outcomes, and as such are likely to have lower forecast errors. These three measures are combined in our second composite proxy, which we label FORECAST. It is also possible that IPO earnings forecasts reflect some degree of moral hazard on the part of those providing them. Lack of publicly available information may provide managers with opportunities to exploit investors, since the costs of relying on an inaccurate earnings forecast are generally borne by new investors. In contrast, if managers intend to return to the capital market they will have incentives to provide more accurate forecasts to maintain investor confidence. Accordingly, four variables are used to proxy for competing managerial incentives: the level of retained ownership, the proportion of newly- raised funds paid to vendor shareholders, and dummy variables for whether the firm conducted a seasoned equity offering (SEO) in the two years following the IPO and whether or not the offer is a packaged or unit offer. The distinction between IPOs where the offering is a package of current (i.e, shares) and deferred (e.g., options) equity purchase reflects evidence that among Australian IPOs, unit IPOs represent a signalling strategy intended to address concerns about the quality of the firm’s business model that would otherwise result in increased underpricing (Lee et al., 2003a). We combine these four measures into our third composite proxy, which we label INCENTIVES. Apart from our focus on the effect of differential audit quality, another external party expected to serve a monitoring function in relation to the prospectus is the underwriter. Although not directly responsible for the forecast provided in the prospectus, underwriters typically have access to superior information about the strategy and future prospects of the firm which is relevant to valuation of the IPO. Similar to Big 6 auditors, high quality underwriters likely have their reputation at stake in the event that a forecast is found to be extremely inaccurate/biased. This leads to an expectation that high quality underwriters will encourage IPO firms to provide more accurate earnings forecasts. A dummy variable is used to indicate if the IPO firm used a high quality underwriter or not. 22 Finally, we include industry dummies in our regressions to reflect possible industry-specific variation in forecast attributes. This is especially relevant to IPOs, where it is also possible that IPOs cluster by type according to market conditions. We use the broad industry groupings summarized in Table 1. Our model used to identify the effect of audit quality on forecast accuracy and bias is therefore as follows: Forecast e rror a b IA_Big6 b FIRMRISK b FORECAST b INCENTIV 12 34 =+ + + + EES bUWQ bINDUSTRY e 56 + ++ (1) where: IA_BIG 6 equals 1 if the auditor is a BIG 6 auditor, otherwise 0; FIRMRISK is a composite factor capturing firm-specific attributes that are likely to be associated with the riskiness of the forecasting task; FORECAST is a composite factor capturing attributes of the forecast which are likely to be associated with increased variation and/or bias; INCENTIVES is a composite factor capturing variables which are likely to be associated with managers’ incentives to make accurate and/or biased forecasts; INDUSTRY is a numeric variable distinguishing sample firms on industry groupings as outlined in Table 1; and e is an error term. Descriptive data on the variables used to construct our composite proxies, the underwriter quality proxy and the auditor quality variable is reported in Table 3. Panel A reports mean, standard deviation, minimum, median and maximum values for the continuous variables for the full sample and the Big 6/non-Big 6 sub-samples. Data relating to Auditor Conservatism and Audit Quality 191 Int. J. Audit. 10: 183–199 (2006)© 2006 The Author(s) Journal compilation © Blackwell Publishing Ltd. 2006 Table 3: Descriptive statistics of firm characteristics for a sample of 179 Australian industrial IPOs with a matched earnings forecast between January 1991 and June 1998 Panel A contains mean, standard deviation, minimum, median and maximum for continuous variables for the full sample and the Big 6 and non-Big 6 audit sub-samples, as well as t-statistics comparing mean values for Big 6 and non-Big 6 auditees. Panel B contains frequency data for binary variables, partitioned by audit quality and a chi-squared statistic comparing frequencies between Big 6 and non-Big 6 auditees. Panel A: Descriptive statistics – continuous variables Variable Mean Std. Dev Min. Median Max. Mean diff Student T Prob (t) Firm age All 7.90 3.27 0 10 10 -1.05 -1.99 0.05** Big 6 8.22 3.02 0 10 10 non-Big 6 7.17 3.73 0 10 10 Total assets ($’000) All 549,290 3,108,500 2,058 33,682 30,218,000 -725,000 -1.44 0.15 Big 6 767,867 3,702,444 3,358 46,243 30,218,000 non-Big 6 43,324 82,082 2,058 18,453 543,400 Issue size ($’000) All 147,440 1,066,600 61 14,000 14,153,260 -187,000 -1.08 0.28 Big 6 203,908 1,273,569 61 20,000 14,153,260 non-Big 6 16,730 33,801 1,300 8,000 235,000 Leverage (%) All 39.59 23.90 0 41.73 94.95 -6.82 -1.76 0.08* Big 6 41.64 23.38 0 43.80 94.95 non-Big 6 34.83 24.63 0 36.86 91.94 No. of subsidiaries All 6.70 12.29 0 4 94 -3.39 -1.7 0.09* Big 6 7.72 13.60 0 5 94 non-Big 6 4.33 8.11 0 3 58 Growth options All 0.61 0.34 0 0.67 1 -0.02 -0.27 0.78 Big 6 0.62 0.33 0 0.66 1 non-Big 6 0.60 0.34 0 0.69 1 Risk factors All 8.20 4.65 0 8 32 0.03 0.04 0.96 Big 6 8.19 4.95 0 8 32 non-Big 6 8.22 3.93 0 9 18 Retained ownership (%) All 45.47 26.90 0 50 99.20 4.23 0.97 0.34 Big 6 44.19 28.02 0 49.50 99.20 non-Big 6 48.42 24.09 0 50.50 95.40 Funds to vendor (%) All 26.54 39.54 0 0 100 -17.20 -2.72 0.01*** Big 6 31.72 41.87 0 0 100 non-Big 6 14.53 30.64 0 0 100 Forecast detail index All 4.93 1.45 1 5 9 -0.80 -3.47 0.00*** Big 6 5.17 1.44 1 5 9 non-Big 6 4.37 1.34 2 4 7 Horizon (months) All 8.66 3.36 2.1 8.4 18.2 0.16 0.29 0.77 Big 6 8.61 3.49 2.1 8.3 18.2 non-Big 6 8.77 3.06 2.5 8.4 16.6 192 P. J. Lee et al. Int. J. Audit. 10: 183–199 (2006)© 2006 The Author(s) Journal compilation © Blackwell Publishing Ltd. 2006 [...]... (1994), ‘The association between audit quality, retained ownership, and firm-specific risk in U.S vs Canadian IPO markets’, Journal of Accounting and Economics, Vol 17, pp 207–28 Craswell, A T., Francis, J R & Taylor, S L (1995), Auditor brand name reputations and industry Int J Audit 10: 183–199 (2006) Auditor Conservatism and Audit Quality specializations’, Journal of Accounting and Economics, Vol 20,... G & Hughes, J (1991), ‘The role of audits and audit quality in valuing new issues’, Journal of Accounting and Economics, Vol 14, pp 3–49 Davidson, R & Neu, D (1993), ‘A note on the association between audit firm size and audit quality , Contemporary Accounting Research, Vol 9, pp 479–88 DeAngelo, L E (1981), Auditor size and auditor quality , Journal of Accounting and Economics, Vol 3, pp 183–99 Dechow,... choice of auditor, and the optimal degree of conservatism (or accuracy) are endogenously determined Of course, by focusing on IPO forecasts, we may not be able to generalize to other settings such as forecasts made by listed firms, where the incentives for making accurate and/ or biased forecasts may be somewhat different from the IPO setting 4 Australian Auditing Standard (AUP) 3.1 ‘Special Purpose Auditor s... avoiding excessively optimistic statements relative to conservatism (Basu et al. , 2001) Also, our results may reflect a better specified test of the relation between forecast bias and differential audit quality than Clarkson (2000), who Int J Audit 10: 183–199 (2006) Auditor Conservatism and Audit Quality is faced with the potentially confounding effect of an (endogenous) voluntary disclosure decision... forecast errors between Big 6 and non-Big 6 auditees is essential Panel B of Table 3 contains descriptive data relating to the binary variables, partitioned into Big 6 and non-Big 6 auditees For the sample companies, approximately 70% have a Big 6 auditor, with 55% having a high quality underwriter There is a statistically significant association between auditor and underwriter quality (chi-square significant... 10: 183–199 (2006) Auditor Conservatism and Audit Quality 195 Table 4: Multivariate analysis of determinants of forecast error for a sample of 179 Australian industrial IPOs between January 1991 and June 1998 Panel A reports regression models of determinants of accuracy, while Panel B reports regression models of determinants of bias The first model in each of Panels A and B (i.e., Models 1 and 3) report... indirect evidence of differential audit quality, there is relatively little direct evidence based on ‘output’ from the audit process Given that there is evidence consistent with IPO firms choosing audit quality based on signalling considerations, the properties of the most observable financial output in that setting (i.e., earnings forecasts) is likely of considerable interest Our paper is also motivated... Economics, Vol 31, pp 105–231 Lee, M., Lee, P J & Taylor, S L (2003a), ‘Unit initial public offerings: Staged equity or signaling mechanism?’ Accounting and Finance, Vol 43, pp 63–86 Lee, P J., Stokes, D J Walter, T S & Taylor, S L (2003b), ‘The association between audit quality, accounting disclosures and firm-specific risk: Evidence from initial public offerings’, Journal of Accounting and Public Policy, Vol... analysis shows all of our conclusions are robust to this choice 21 This measure has previously been used in an examination of the relation between audit quality, IPO pricing and voluntary disclosure (Lee et al. , 2003b) 22 We are unaware of any ‘established’ rankings for underwriters of Australian IPOs, and so we designate underwriters as high quality if they © 2006 The Author(s) Journal compilation... for sale of shares FC detail index A score from 1–9 used to capture the detail of forecast information provided Horizon The number of months between the prospectus date and the end of the financial period to which the forecast relates */**/*** Statistically significant at 10%, 5%, 1% Variable defintions Age Assets Issue size Leverage Subsidiaries Auditor Conservatism and Audit Quality 193 Int J Audit . the common auditor size-based proxy for audit quality. Several studies focus on the link between audit quality (proxied by Big 6) and the accrual component of earnings. 7 Francis et al. (1999) demonstrate. Auditor Conservatism and Audit Quality: Evidence from IPO Earnings Forecasts Philip J. Lee, 1 Sarah J. Taylor 2 and Stephen L. Taylor 3 1 University of Sydney, Australia 2 University. statistically significantly different from non-Big 6 auditees at 10%, 5%, 1% levels. Auditor Conservatism and Audit Quality 189 Int. J. Audit. 10: 183–199 (2006) © 2006 The Author(s) Journal compilation