Fair Value and Audit Fees

Transcription

1 Fair Value and Audit Fees Igor Goncharov WHU Otto Beisheim School of Management Edward J. Riedl * Harvard Business School Thorsten Sellhorn WHU Otto Beisheim School of Management This version: May 2011 Abstract: We investigate the effect of fair value reporting and its attributes on audit fees. We use as our primary setting the European real estate industry around the time of IFRS adoption, which provides unique heterogeneity on a number of dimensions to assess fair value. We first contrast audit fees across firms applying fair value versus amortized cost as the reporting model for their property assets. We find that firms reporting under fair value exhibit lower audit fees, and that this is (in part) driven by impairment tests that occur only under amortized cost. We then find that audit fees are increasing in the complexity of the fair value estimation, and are higher for fair values that are recognized (versus only disclosed). Overall, the results complement prior findings that fair values can reduce information asymmetries by suggesting they can also lead to lower contracting costs. However, the results further suggest that any reductions in audit fees vary with salient characteristics of the fair value reporting, including the difficulty to measure and treatment within the financial statements. Key Terms: fair value, audit fees, audit pricing, real estate industry, IFRS * Corresponding author: Harvard Business School Morgan Hall 365 Boston, MA Phone: Fax: eriedl@hbs.edu Acknowledgements: We appreciate helpful comments from Jere Francis, Martin Glaum, Paul Michas, Frank Moers, George Serafeim, Suraj Srinivasan, Ann Vanstraelen, and workshop participants at HEC Paris, University of Essen, University of Giessen, University of Innsbruck, University of Maastricht, University of Missouri, WHU Otto Beisheim School of Management, and the 2011 EAA Annual Congress. We also appreciate the insights from audit partners participating in interviews. Electronic copy available at:

2 Fair Value and Audit Fees Abstract: We investigate the effect of fair value reporting and its attributes on audit fees. We use as our primary setting the European real estate industry around the time of IFRS adoption, which provides unique heterogeneity on a number of dimensions to assess fair value. We first contrast audit fees across firms applying fair value versus amortized cost as the reporting model for their property assets. We find that firms reporting under fair value exhibit lower audit fees, and that this is (in part) driven by impairment tests that occur only under amortized cost. We then find that audit fees are increasing in the complexity of the fair value estimation, and are higher for fair values that are recognized (versus only disclosed). Overall, the results complement prior findings that fair values can reduce information asymmetries by suggesting they can also lead to lower contracting costs. However, the results further suggest that any reductions in audit fees vary with salient characteristics of the fair value reporting, including the difficulty to measure and treatment within the financial statements. Key Terms: fair value, audit fees, audit pricing, real estate industry, IFRS 1 Electronic copy available at:

3 Fair Value and Audit Fees 1 Introduction This paper examines the effect of fair value reporting and its attributes upon audit fees. Prior research documents the magnitude and determinants of audit fees (for a review, see Hay et al., 2006). Prior research also documents that fair value reporting is both priced by the equity market (e.g., Easton et al., 1993), and has the capacity to reduce information asymmetries across investors (e.g., Muller et al., 2011). We combine both literatures to investigate how fair value reporting, and various salient attributes of how it is implemented, affects the pricing of audit services a major contracting cost (Jensen and Meckling, 1976). We use as our primary setting real estate firms domiciled in Europe during the period , a setting which provides several benefits. First, the within-industry design holds constant other factors that could drive audit fee differences across industries, allowing us to structure analyses that better isolate the effect of fair value reporting. Second, European real estate firms exhibit substantial variation in the reporting of fair values for their property assets, which we exploit in our analyses. Specifically, prior to adoption of International Financial Reporting Standards (IFRS) by the European Union effective 2005, domestic standards varied in their requirements for the reporting of real estate assets on the balance sheet: either at fair value, or at amortized cost subject to impairment. Subsequent to IFRS adoption, the applicable standard allowed firms to report property assets either under the fair value or cost models; however, IFRS also introduced a mandatory footnote disclosure requirement of property fair values for those firms electing the cost model. Third, this industry setting allows detailed examination of four specific attributes of fair value reporting: the proportion of a firm s assets reported at fair value; the difficulty of estimating the fair values reliably; whether the fair values are recognized on the face of the financial statements or disclosed in the footnotes; and the effect of using an alternative external monitor an 2

4 external appraiser to derive the fair value estimates. No other institutional setting allows these features to be concurrently examined. Empirical results reveal that, controlling for other determinants, audit fees are significantly lower for firms reporting property assets at fair value relative to those reporting property assets at amortized cost. To further assess this arguably unexpected result, we conduct interviews with real estate audit partners: these interviews suggest that (potential and actual) impairments are likely a significant source of higher audit effort for amortized cost firms, as are other reporting requirements (such as component depreciation) which arise only within amortized cost contexts. 1 Testing these expectations, we indeed find that impairments reported by amortized cost firms are a significant driver of observed higher average fees. Our analyses of particular attributes of fair value implementation reveal that audit fees are lower for firms with above-average exposure to assets recognized or disclosed at fair value, consistent with auditors being able (on average) to reduce effort and/or risk when a greater proportion of the client firm s assets are reported at fair value. In addition, audit fees are higher for more complex property portfolios, consistent with multiple-sector portfolios (e.g., retail, office, industrial, residential) being more difficult to value and audit. Finally, audit fees are higher for fair values that are recognized on the face of the financial statements versus only disclosed in the footnotes, consistent with incremental audit effort for reporting elements recognized within the financial statements (e.g., Libby et al., 2006). We fail to find evidence that use of an external appraiser attenuates audit fees. As above, the viability of these results is again confirmed via interviews with real estate audit partners. We then re-assess these findings by exploiting two alternative settings. First, to mitigate concerns that results are confined to the European setting, we reassess the relation of 1 As discussed later, component depreciation is required under the amortized cost model, and requires that firms allocate portions of buildings to particular functions and depreciation schedules (e.g., piping, structure, office improvements). Fair value reporting does not require such allocations, which are effectively captured in the estimate of the property s fair value. 3

5 fair value reporting to audit fees by comparing audit pricing for UK versus US real estate firms; these firms are selected using propensity score matching. Since the UK and US are among the world s largest and most developed real estate markets, this analysis benefits from a larger sample size and isolates the main difference between both countries for this industry: the financial reporting model. Specifically, UK firms report property assets at fair value, while US firms report them at amortized cost. Further, while the US generally is considered to have higher audit litigation risk than other countries, prior research fails to designate the real estate industry as a high risk setting, suggesting that litigation differences are unlikely to be the primary source of audit pricing differences across the US and UK for this sector. Consistent with our previous results, we document that audit fees are significantly lower for firms reporting property assets at fair value (i.e., the UK firms) relative to amortized cost (i.e., the US firms). Further, impairments again appear to be a primary driver of higher audit fees for amortized cost firms. Second, we re-assess the effects of exposure to fair value and complexity in fair value measurement using a sample of UK investment trusts. Restricting the analysis within the UK eliminates cross-country differences in institutional features as a potential source of variation in audit fees. Further, this industry provides a potentially stronger assessment of the difficulty to measure fair value by exploiting fair values calculated based on market inputs (level 1 fair values) versus those based on less reliable valuation inputs (level 2 and 3 fair values). Consistent with our previous results, we find that audit fees are decreasing in the firm s exposure to assets reported at fair value, and increasing in the firm s exposure to more difficult-to-measure (i.e., levels 2 and 3) fair values. The results are also robust to sensitivity analyses including (1) removal of firm-year observations potentially affected by the global financial crisis, (2) control for country effects, and (3) alternative measurement of the dependent variable. 4

6 Overall, the results reveal that reporting assets at fair value (on average) reduces audit fees, where a primary driver of higher audit fees observed under amortized cost appears to be impairments. Further, we find any reduction in audit fees relative to amortized cost depends on several salient characteristics of the fair value reporting: the overall exposure to fair value measurement; the complexity of the fair value measurement; and whether the fair values are recognized in the primary financial statements versus only disclosed in the footnotes. Combined with previous findings that fair value reporting can reduce information asymmetries, these results suggest that fair value reporting can have benefits both for decision usefulness as well as contracting, fostering both objectives of financial reporting. These findings are potentially informative to standard-setters in their ongoing deliberations about the role of fair value measurement in general-purpose financial statements. Section 2 presents the prior literature, primary setting, and hypothesis development. Section 3 presents the base audit fee regression. Section 4 presents the primary analyses. Section 5 presents alternative settings to assess the hypotheses. Section 6 presents sensitivity analyses, and Section 7 concludes. 2 Prior literature, primary setting, and hypothesis development 2.1 Prior literature This study builds upon two broad literatures: that examining the determinants of audit fees, and that examining the effects of fair value reporting. There is a substantial literature on audit pricing, with Simunic (1980) among the earliest to provide theoretical and empirical evidence on the determinants of this contracting cost. 2 Hay et al. (2006), in a survey of the literature on the determinants of audit fees, suggests that under a competitive audit market 2 We follow Jensen and Meckling (1976) in viewing audit fees as one of the agency costs arising from a contractual arrangement between the owners (principal) and the management (agent) of a firm; that is, audit fees represent monitoring (bonding) costs. See also Watts and Zimmerman (1986). 5

7 these determinants may be classified into several major categories: client attributes, auditor attributes, and characteristics specific to the audit engagement. Among these, client attributes have received the most attention, and commonly reflect the firm characteristics of size, risk, and complexity. In particular, consistent with theory on audit effort and litigation, audit fees generally are found to increase in the client s size (e.g., Simunic, 1980), risk (e.g., Stice, 1991), and complexity (e.g., Hackenbrack and Knechel, 1997). The literature examining the effects of fair value reporting is also extensive, with many papers analyzing the relation of fair values and equity prices. Earlier empirical papers, such as Barth (1994) and Eccher et al. (1996) exploit disclosed fair values relating to financial instruments and provide evidence that fair values are value relevant: that is, reflected in stock prices. Similar results are found for alternative non-financial asset classes (e.g., Easton et al., 1993). Other studies examine the effect of fair values on the information environment, documenting that asymmetry is reduced when fair values are disclosed (e.g., Muller et al., 2011), but that information risk is higher when fair values are based on unobservable inputs (Riedl and Serafeim, 2011). Among the few papers speaking to contracting implications of fair value accounting, Barth et al. (1995) suggests that fair value accounting, if applied to assess banks regulatory capital, can address struggling financial institutions problems earlier than does amortized cost. Benston (2006) argues that fair value accounting is ill-suited as a basis for accounting-based management compensation. We combine and contribute to these literatures by investigating the effect of fair value reporting upon audit fees in two ways. First, we contrast observed audit fees across firms using fair value relative to those using amortized cost as the principal reporting model for their primary operating assets. Second, we exploit known variation in the salient characteristics of reported fair values to identify predictable differences in audit fees. 6

8 2.2 Primary setting: the European real estate industry Although fair value accounting has been adopted by standard setters for numerous reporting elements spanning a number of industries, it is difficult to identify a single setting in which multiple attributes of fair value reporting can be observed simultaneously. For this reason, we use the European real estate industry as the primary setting, which has several advantageous features. First, firms in this industry share a common primary operating asset: real estate (i.e., investment property). That is, these firms acquire (through purchase, lease, or development), manage, and sell real estate to generate profits through rentals and/or capital appreciation. Typically, firms either acquire legal ownership through a purchase or hold the property under a finance lease. Accordingly, the primary asset for our sample firms is held constant: long-lived tangible real estate assets. 3 Second, the industry is well-developed. Within Europe, there are over 180 publicly traded real estate firms, with an aggregate equity market value exceeding 150 billion on December 31, The firms are domiciled across most European countries, with the larger economies (e.g., France, Germany, and UK) having a higher representation. Third, this industry exhibits substantial variation in how firms report property assets. Prior to the mandatory transition by many European countries from domestic standards to IFRS in 2005, domestic standards varied considerably in the required accounting treatment for real estate assets: some required reporting at amortized cost (e.g., Germany) and some required reporting at fair value (e.g., the UK). Even upon IFRS adoption, variation in reporting continued. Specifically, the relevant standard addressing the accounting for real estate assets, International Accounting Standard (IAS) 40: Investment Property (IASB, 2000), allows firms to choose between reporting these assets at fair value or at amortized cost on the 3 Note that real estate assets for our sample firms reflect real estate assets used for investment, as opposed to real estate assets used in production or operations. 7

9 balance sheet. Thus, the reporting of real estate assets exhibits significant variation, both prior to and subsequent to IFRS adoption, which we exploit in our analyses. Finally, firms in this industry exhibit substantial variation in salient fair value attributes beyond the principal reporting model, allowing for further analysis. Specifically, real estate firms vary in their exposure to assets reported at fair value, owing to other operating segments or balance sheet items. The firms also vary in the complexity of the fair value measurement, which is driven in large part by variation in property portfolios, particularly across sectors such as retail, office, industrial, and residential. The firms also vary in the recognition versus disclosure of investment property fair values. Under IAS 40, firms can elect to either recognize real estate fair values on the balance sheet under the fair value model, or provide mandatory footnote disclosure of these fair values if electing the cost model. Finally, these firms vary in the use of a key alternative external monitor to derive fair value estimates: the external property appraiser. 2.3 Hypothesis development We first examine the effect of reporting model on observed audit fees, contrasting fair value versus amortized cost. Our focus on the real estate industry allows us to partition firms into those reporting their primary operating asset under either of these two models. The effect of reporting assets at fair value relative to amortized cost upon audit effort, as well as assessed reputation and litigation risk, is unclear a priori. Critics maintain that fair value reporting introduces substantial discretion into management estimates (e.g., Watts, 2006; Ramanna and Watts, 2010). This increased discretion can compound agency costs, leading auditors to increase their assessment of reputation and/or litigation risk and, consequently, their efforts to verify fair value estimates. This risk may be greater in contexts, such as real estate, where market prices for identical assets are generally unavailable. Interviews with 8

10 real estate audit partners confirm that fair values for real estate assets are viewed as either level 2-type values (with market values available for similar, but not identical, properties) or as level 3-type values (with simplistic discounted cash flow analysis used in the absence of property-specific market parameters). Alternatively, fair values may reduce auditor reliance on management estimates and litigation risk (e.g., to the extent these values are derived from observable inputs). For example, firms with real estate assets typically have rent rolls for clients having multi-year leases within their properties, allowing auditors to more clearly identify and support management forecasts of future cash flows, and thus to more easily ascertain property value. In addition, amortized cost based reporting has several aspects that introduce complexity and uncertainty into the auditing process. Two examples most relevant to the real estate setting include component depreciation and impairment testing. Under component depreciation, a property is depreciated based on the lives of individual elements within it. For example, within a given property, certain electrical/plumbing components can be assigned a 20-year life, the roof a 15-year life, and the foundation/structure a 50-year life. The assignment process across multiple components for multiple properties can increase audit costs; such costs do not directly arise for firms reporting under fair value, where such factors are incorporated into the fair value of the property. Second, impairment testing requirements under amortized cost can lead to higher audit costs. Under fair value reporting, firms must establish a process of valuation; this process becomes the focal point of the annual audit, regardless of upward or downward changes in property values. That is, the valuation process provides a basis for repeated annual discussions with the auditor; to the extent this process does not change, the audit process is focused primarily on updating assumptions and inputs into the fair value process for the current year. In contrast, if a property value decline leads to impairment 9

11 considerations for a firm applying amortized cost, the auditor must now assess both the firm s valuation and process in a non-routine (and likely contentious) setting. Further, this process can differ from an assessment of fair value as impairments anchor on the notion of recoverable amount. Recoverable amount reflects the higher of fair value or firm-specific value in use, where the latter may substantially deviate from the former and involve higher discretion by management. Thus, it is probable that impairments lead to substantial frictions, and thus higher audit fees, for firms reporting under amortized cost. Taken together, this reasoning leads to our first two hypotheses (in alternative form): H 1A Audit fees differ for firms applying amortized cost versus fair value reporting to their primary operating assets. H 1B Impairments lead to higher audit fees for firms reporting their primary operating assets under amortized cost. We next examine attributes of fair value reporting likely to lead to heterogeneity in auditor efforts. To assess these effects, we focus on four attributes of fair value reporting likely to affect audit fees: the firm s exposure to fair value reporting; the complexity of the fair value measurements; whether fair value is recognized on the face of the financial statements versus disclosed in the footnotes; and the use of alternative non-auditor external monitors in the fair value measurement process. All four attributes have observable variation within the real estate industry. First, we examine the firm s exposure to fair value reporting. Firms with greater proportions of their primary operating assets reported at fair value may require additional audit efforts, owing to incremental procedures necessary to confirm additional fair values. Alternatively, to the extent the audit focuses on the process, firms with greater proportions of their operating assets reporting under fair value may require fewer efforts (e.g., due to a lack of component depreciation or impairment testing, or due to economies of scale arising 10

12 through audit of the fair value process versus audit of particular valuation estimates). This leads to the following hypothesis: H 2A Audit fees differ for firms reporting a higher proportion versus a lower proportion of their primary operating assets at fair value. Second, we assess the role of complexity in fair value measurement, which reflects challenges inherent in the estimation process. These include estimation using a long time series of cash flows, or the availability of benchmarks to approximate fair value. Both notions are considered in applicable US accounting principles and IFRS regarding fair value measurement. Consistent with this framework, we expect that audit effort will be higher for fair values requiring more complex estimation procedures (e.g., Hackenbrack and Knechel, 1997). That is, property portfolios that are more complex due to spanning multiple sectors, or due to unavailability of market-derived valuation inputs, likely require additional audit procedures. This leads to our second hypothesis: H 2B Audit fees are higher for firms with more difficult-to-measure fair values. Third, we assess the role of recognition versus disclosure. Numerous standards have increased fair value disclosure requirements; others have required that fair value levels and changes be incorporated into the primary financial statements. 4 Consistent with prior research (Libby et al., 2006), we predict that audit fees are higher for fair values that are recognized versus disclosed, consistent with auditors expending more effort to validate information recognized in the primary financial statements. This test allows us to exploit the IAS 40 option for real estate firms to elect recognition under the fair value model, or disclosure of fair values under the cost model. This leads to our third hypothesis: 4 Examples of standards requiring disclosure of fair values include in the US SFAS 107 Disclosures about Fair Value of Financial Instruments (FASB, 1991), and in the EU IFRS 7 Financial Instruments: Disclosures (IASB, 2005). Examples of standards requiring recognition of fair values include in the US SFAS 115 Accounting for Certain Investments in Debt and Equity Securities (FASB, 1993) and SFAS 123 Accounting for Stock-Based Compensation (FASB, 1995), and in the EU IFRS 9 Financial Instruments (IASB, 2010b), IAS 39 Financial Instruments: Recognition and Measurement (IASB, 1999), and IAS 41 Agriculture (IASB, 2001). 11

13 H 2C Audit fees are higher for firms reporting assets at fair value that are recognized on the balance sheet relative to firms only disclosing them in the footnotes. To the extent audit effort does not differ across recognized versus disclosed reporting items, this will bias against H 2C. Finally, other external monitors may provide inputs into the audit process, with real estate firms frequently employing external appraisers in the fair value measurement process. Auditing standards recognize the role of experts, such as International Standard on Auditing 500: Audit Evidence (International Federation of Accountants, 2010), which states that auditors may accept the findings of a specialist hired by management as appropriate audit evidence. This suggests a substitution role: that is, specialists may provide expertise and insights, which can potentially reduce necessary efforts by the auditor to achieve a particular level of audit risk. Prior research provides evidence consistent with this notion: Muller and Riedl (2002) documents that information asymmetry is lower for property firms employing external (versus internal) appraisers in the real estate industry, and Cotter and Richardson (2002) similarly document that revaluations of property, plant and equipment by independent appraisers are more reliable than those conducted by directors. Accordingly, we predict that audit fees are lower for firms employing external monitors as part of the fair value reporting process, reflecting potential substitution of efforts. This leads to our final hypothesis: H 2D Audit fees are lower for firms reporting assets at fair value derived using (nonauditor) external monitors relative to firms that do not use such monitors. To the extent there is variation in the quality of external appraisals, this can bias against H 2D. 3 Research design: base model of audit fee determinants Prior research has accumulated evidence of a number of audit fee determinants. For our main research setting, the European real estate industry, we maintain the relevant determinants, using the following basic audit fee model: 12

14 LogFees it = α 0 + α 1 LogTA it + α 2 IFRS_Adopt it + α 3 Foreign it + α 4 NSegm it + α 5 ADR it where: + α 6 ROA it + α 7 Loss it + α 8 Receiv it + α 9 Lev it + α 10 Distress it + α 11 Qualified it + α 12 Volatility it + α 13 BigN it + α 14 Yearend it + ψ it (1) LogFees it is the log of total auditor fees paid by firm i for year t; LogTA it is the log of firm i s total assets at the end of year t; IFRS_Adopt it is an indicator variable equal to 1 for the first fiscal year t, and immediately preceding fiscal year t-1, of first-time IFRS adoption for firm i, and 0 otherwise; Foreign it is international assets divided by total assets for firm i for year t; NSegm it is the number of firm i s operating segments for year t; ADR it ROA it Loss it is an indicator variable equal to 1 if firm i is cross-listed in the United States for year t, and 0 otherwise; is firm i s net income, net of impairment losses, divided by total assets, both measured for year t; is an indicator variable equal to 1 if firm i reports negative net income for year t, and 0 otherwise; Receiv it is firm i s receivables divided by total assets, both measured for year t; Lev it is firm i s total debt divided by market value of equity for year t; Distress it Qualified it is an indicator variable equal to 1 if firm i reports negative book value of equity for year t, and 0 otherwise; and is an indicator variable equal to 1 if firm i receives a qualified audit opinion for year t or t-1, and 0 otherwise; Volatility it is the standard deviation of monthly stock returns for firm i over year t; BigN it is an indicator variable equal to 1 if firm i uses a large auditor (i.e., Big 4 or Big 6) during year t, and 0 otherwise; and Yearend it is an indicator variable equal to 1 if firm i has a fiscal year end between December and March (corresponding with the audit busy season) for year t, and 0 otherwise. 13

15 Our dependent variable is LogFees, the log of total auditor fees. 5 Consistent with prior studies examining the determinants of audit fees, we express the dependent variable in log form to mitigate the effects of non-linear relations (see Hay et al., 2006). We then include variables assessed in prior research to drive audit fees. We use two primary groupings: characteristics of the audit client, and those of the audit firm. Regarding audit client characteristics, we first include LogTA; as audit effort is expected to increase in the scale of the client, the predicted sign is positive (Simunic, 1980). Given the assetintensive nature of the real estate industry, this variable appears particularly relevant. To capture audit complexity, we include IFRS_Adopt, Foreign, and NSegm. As audit effort is expected to be higher for real estate firms transitioning to IFRS, having more international operations, or having more complex operations, the expected sign on the coefficients for these three variables is positive. We also include ADR to capture additional audit effort or litigation risk arising from exposure to the US capital market through a crosslisting; the predicted sign is positive. Next, we include variables to capture firm risk, which has been documented to have a positive association with audit fees (e.g., Stice, 1991). We include accounting measures of firm performance, both assessed as a continuous variable (ROA) and as a distress indicator variable (Loss); the predicted signs are negative and positive, respectively. We then include several balance sheet constructs to capture audit risk. Consistent with prior research (e.g., Hay et al., 2006), we include Receiv; as receivables may be subject to higher risk of error, the predicted sign is positive. 6 Real estate firms maintain moderate amounts of receivables, reflecting amounts due from tenants. We also include Lev; as more leveraged firms face 5 6 Similar to prior research, this variable is based on the Thomson Reuters Worldscope data item 01801, Auditor Fees, which comprises fees paid to the firm s auditor for the statutory audit of the financial statements (statutory audit fees) as well as fees paid to the firm s auditor for other services (non-audit fees). We examine alternative definitions in section 6.3. While prior research proposes inventory as another risk construct leading to higher audit fees, we exclude this variable as real estate firms do not typically hold material amounts of inventory. Nonetheless, results are unchanged if we include inventory (scaled by total assets) in the regression. 14

16 greater financing constraints, the predicted sign is positive. Real estate firms typically employ substantial leverage, suggesting it is a very relevant construct for this industry. Next, we include Distress and Qualified to capture extreme negative performance; as firms with negative equity or qualified audit opinions are more likely in distress, the predicted signs for both variables are positive. Finally, we include Volatility to reflect overall market risk; as more volatile stock returns reflect riskier firms, the predicted sign is again positive. The second group of variables includes audit characteristics, which would affect audit pricing for real estate firms. We include BigN to capture perceived higher quality and/or reputational effects of larger audit firms (i.e., the large audit firm premium ; see Francis, 1984); the predicted sign is positive. Finally, we include Yearend to capture the higher fees charged when audits occur during periods of constrained auditor resources (i.e., during the audit busy season ; see Ireland and Lennox, 2002); the predicted sign is positive. In the sections that follow, we refer back to this base model, augmenting it by adding experimental variables to test our hypotheses. 4 Primary tests 4.1 The effect of reporting model on audit fees: evidence using European real estate firms upon mandatory IFRS adoption Our first test examines whether audit fees decrease when European real estate firms switch from amortized cost to fair value. Specifically, we use mandatory adoption of IAS 40 in Europe as a natural experiment to conduct a difference-in-differences analysis. We compare how audit fees change upon mandatory IFRS adoption for two sets of European real estate firms: those domiciled in countries requiring that property assets be reported at amortized cost under pre-ifrs domestic standards (treatment group), and those requiring that property assets be reported at fair value under pre-ifrs domestic standards or early IFRS 15

17 adoption (control group). 7 If audit fees vary with the firms primary reporting model (H 1A ), we should find a significantly larger change in audit fees for the treatment group (which transitions from a cost regime to a fair value regime) compared to the control group (which remains on a fair value regime throughout the analysis period). We then explore a possible reason for differential audit fees across reporting regimes. If impairment testing (which only firms reporting under amortized cost are subject to) drives up audit fees (H 1B ), we should observe higher audit fees for firm-year observations with recognized impairment losses compared to those without such charges. To implement our difference-in-differences analysis within the linear regression framework of equation (1), we estimate the following augmented model: LogFees it = ξ 0 + ξ 1 LogTA it + ξ 2 Foreign it + ξ 3 NSegm it + ξ 4 ROA it + ξ 5 Loss it + ξ 6 Receiv it where: + ξ 7 Lev it + ξ 8 Distress it + ξ 9 Qualified it + ξ 10 Volatility it + ξ 11 BigN it + ξ 12 Yearend it + ξ 13 HC it + ξ 14 IFRS it + ξ 15 HC it IFRS it + ξ 16 Impair_D it + ω it (2) HC it IFRS it Impair_D it is an indicator variable equal to 1 if firm i is domiciled in a country that required property assets to be carried at amortized cost under pre-ifrs domestic standards, and 0 otherwise (i.e., is domiciled in a country that required property assets to be carried at fair value under pre-ifrs domestic standards or under early IFRS adoption). Countries requiring amortized cost include Austria, Belgium, Finland, France, Germany, Italy, Norway, Poland, Spain, and Switzerland; those requiring fair value include Denmark, the Netherlands, Sweden, and the United Kingdom; is an indicator variable equal to 1 if firm i is reporting under IFRS (reporting property assets at fair value) in year t, and 0 otherwise (i.e., is reporting under pre-ifrs domestic standards in year t); and is an indicator variable equal to 1 if firm i reports impairment charges during year t, and 0 otherwise. All other variables are as defined previously (see Appendix A). Equation (2) excludes IFRS_Adopt as the years captured by that variable are excluded from the analysis. The 7 We exclude the year of first-time IFRS adoption and the immediately preceding year to avoid the increased audit effort due to the implementation of a new accounting framework. Our inferences are unchanged when these transition years are included in the analysis. 16

18 experimental variables are HC, IFRS, their interaction of HC IFRS, and Impair_D. HC captures the difference in audit fees between treatment firms (applying amortized cost) and control firms (applying fair value) before IFRS adoption. IFRS captures the effect of switching to IFRS for the control firms (i.e., for firms applying fair value accounting before IFRS adoption). HC IFRS captures the incremental effect on audit fees of moving from an amortized cost to a fair value regime; its coefficient ξ 15 tests H 1A. Impair_D captures the effect on audit fees of reporting impairment charges; its coefficient ξ 16 tests H 1B. Table 1 presents the sample selection, with the final sample including real estate firms domiciled in the European Economic Area with the necessary data for the period After deleting outliers and observations from the IFRS adoption year and the preceding year, the sample includes 480 firm-year observations representing 172 unique firms. Columns (1) and (2) of Table 2 present means and medians for this sample. On average, sample firms have 1.7% foreign revenues, 1.9 segments, 24% of observations reporting losses, a leverage ratio of 1.6, and 56.9% of observations being audited by Big N auditors. Table 3 presents the empirical results, which are based on robust standard errors. 8 In Column (1), we report a base model, which is supplemented with the experimental variables in Columns (2) and (3). Throughout all models, the coefficients on most of the control variables are statistically significant with the expected signs. For the base model in Column (1), we find that audit fees are increasing in total assets (0.551, t-stat = 23.80), number of segments (0.123, t-stat = 3.84), receivables (1.763, t-stat = 4.45), volatility (1.742, t-stat = 2.57), use of Big N auditor (0.253, t-stat = 2.57), and audit occurring within the busy season (0.263, t-stat = 3.08), and decreasing in ROA ( 0.472, t-stat = 2.06). 8 An alternative approach would be to cluster standard errors by firm and/or year. Our tests, however, do not have a sufficiently large number of clusters across the year and/or firm dimensions, which is a prerequisite for using clustered standard errors (Petersen, 2009). We perform a sensitivity test and control for clustering by using standard errors clustered by year and firm, which leads to qualitatively similar results. 17

19 Introducing our first set of experimental variables in Column (2), the coefficient for HC is insignificant, consistent with no difference in audit fees between treatment firms and control firms before IFRS adoption. Likewise, the coefficient for IFRS, which captures the effect on audit fees of switching to IFRS for the control firms that used fair value accounting before IFRS, is also insignificant. However, the coefficient on the interaction HC IFRS is significantly negative ( 0.759, t-stat = 2.84), indicating that treatment firms reduction in audit fees upon switching to IFRS is significantly stronger than that experienced by the control firms during the same period. This finding suggests that, ceteris paribus, firms reporting under a cost-based accounting regime experience a significant decrease in audit fees when switching to a fair value-based accounting regime, providing support for H 1A. Turning to the effect of impairment testing on audit fees, Column (3) reports a significantly positive coefficient for Impair_D (0.259, t-stat = 2.31), indicating that firm-year observations in which impairment losses occur are associated with higher audit fees. This result provides support for H 1B, and suggests that the risk and effort associated with impairment testing procedures are major drivers of audit fees, and likely contributes to the results above for H 1A. In interpreting the result on H 1B, we note two items. First, impairment testing occurs only for firms reporting under a pre-ifrs amortized cost-based accounting regime or adopting the cost model under IAS 40; fair value model firms do not conduct impairment testing. Second, Impair_D captures only recognized impairment charges. This suggests that any incremental audit efforts associated with impairment tests that do not lead to recognized impairment charges are not captured by this variable, likely leading Impair_D to understate the full effect of impairment testing on audit fees. 18

20 4.2 The effect of fair value characteristics on audit fees: evidence using European real estate firms after mandatory IFRS adoption To test our second set of hypotheses, we adjust our setting by retaining our focus on the European real estate industry, but turning to the post-ifrs adoption period ( ) during which all firms report under IAS 40. This setting is beneficial as all sample firms report under a uniform fair value regime while nonetheless exhibiting variation across each of the four fair value attributes we wish to examine: exposure, complexity, recognition versus disclosure, and use of an alternative external monitor. Accordingly, we augment the basic audit fee model of equation (1) as follows: LogFees it = β 0 + β 1 LogTA it + β 2 Foreign it + β 3 NSegm it + β 4 ROA it + β 5 Loss it + β 6 Receiv it where: + β 7 Lev it + β 8 Qualified it + β 9 Volatility it + β 10 BigN it + β 11 Yearend it + β 12 FV_TA_RE it + β 13 FV_Complex it + β 14 FV_Recog it + β 15 FV_Ext it + ø it (3) FV_TA_RE it FV_Complex it FV_Recog it is the firm s exposure to assets measured at fair value, calculated in two steps. First, we calculate the proportion of firm i s total assets measured at fair value. For firms reporting property assets on the balance sheet at fair value, it is the ratio of property fair values to total assets; for firms reporting property on the balance sheet at amortized cost, it is the ratio of disclosed property fair values to the sum of total assets less recognized property at amortized cost plus disclosed fair value of property. Second, FV_TA_RE equals 1 if this proportion is higher than the sample mean (indicating higher exposure to assets reported at fair value), and 0 otherwise (indicating lower exposure to assets reported at fair value); is the complexity of firm i s property portfolio in year t, calculated in two steps. First, we sum the square roots of the percentages of property for firm i within each of eleven sectors: land, residential, office, retail, parking, industrial, gastronomy, health care, education, leisure, and other. Second, FV_Complex equals 1 if this measure is above the sample mean for firm i in year t (indicating higher portfolio complexity), and 0 otherwise (indicating lower portfolio complexity); 9 is an indicator variable equal to 1 if firm i recognizes property fair values on the balance sheet in year t, and 0 otherwise (that is, only discloses property fair values in the footnotes); and 9 Alternative measures of FV_Complex (e.g., reducing from 11 to 4 main sectors, or based on reference to median sample complexity values) yield similar results. 19

21 FV_Ext it is an indicator variable equal to 1 if firm i uses an external appraiser to provide investment property fair values in year t, and 0 otherwise. All other variables are as defined previously (see Appendix A). Note that we exclude the control variables IFRS_Adopt, ADR, and Distress due to a lack of variation in this sample. The four experimental variables correspond to H 2A through H 2D. First, we include FV_TA_RE, which captures real estate firms exposure to assets reported (i.e., either recognized or disclosed) at fair value. If higher exposure to fair value reporting for assets requires additional effort by the auditor (e.g., to validate the fair values), the predicted sign is positive: that is, audit fees will be higher for firms having greater exposure to assets reported at fair value. Alternatively, if higher exposure reduces audit effort (e.g., by simplifying procedures necessary to validate the fair values, or due to reductions attributable to more costly audit procedures related to assets carried at cost, such as auditing component depreciation and impairment testing), the predicted sign is negative. Accordingly, the sign is not predicted; and β 12 is our primary test of H 2A. Next, we include FV_Complex, which captures the complexity of the fair value measurement through the heterogeneity of the firm s property portfolio. If greater complexity leads to additional audit effort, the predicted sign is positive; and β 13 is our test of H 2B. We then include FV_Recog, which captures the recognition of fair values (fair value changes) on the balance sheet (income statement). If recognition leads to incremental audit effort, the predicted sign is positive; and β 14 is our test of H 2C. Finally, we include FV_Ext, which captures differences in audit fees due to firms employing external appraisers to provide fair value estimates of their property assets. If the use of this alternative monitor reduces the effort necessary by the auditor, the predicted sign is negative; and β 15 is our test of H 2D. Table 1 reveals that the sample of firms having the necessary hand-collected data includes 159 firm-year observations representing 96 unique firms. Columns (3) and (4) of Table 2 present means and medians for this sample. On average, property assets represent 20

22 72.3% of total assets, 81.1% of observations reflect recognition (versus disclosure only) of property fair values, and 88.7% employ external appraisers. Among the control variables, 9.7% of assets are international, firms average 2.7 segments, on average ROA performance is 6.4%, and 79.2% employ large auditors. Table 4 presents the empirical results. Similar to our previous analysis, Column (1) presents results including only the control variables, while Column (2) additionally incorporates the experimental variables. Focusing on Column (2), among the control variables, we find as predicted that audit fees are increasing in total assets (0.635, t-stat = 13.38), receivables (2.853, t-stat = 2.29), leverage (0.158, t-stat = 3.07), volatility (2.248, t- stat = 2.12), use of Big N auditor (0.322, t-stat = 1.90), and whether the audit occurs during the busy season (0.522, t-stat = 2.81). The remaining control variables are insignificant. 10 Among the four experimental variables, we first find that FV_TA_RE is significantly negative ( 0.413, t-stat = 2.28). This variable captures the shift in average audit fees for firms having above average exposure to operating assets requiring fair value measurement. Accordingly, the significantly negative coefficient supports H 2A, and is consistent with audit firms charging lower audit fees for firms reporting higher proportions of property assets at fair value, relative to those reporting lower proportions. Second, FV_Complex is significantly positive (0.247, t-stat = 2.24). This variable captures the mean shift in audit fees for firms having above average complexity in their property portfolios. The significantly positive coefficient supports H 2B, and is consistent with audit firms charging higher fees when auditing more difficult-to-value, i.e., more complex, property portfolios. Third, FV_Recog is significantly positive (0.383, t-stat = 2.18). This result supports H 2C, and is consistent with audit firms charging higher audit fees for fair values that are recognized on the primary financial statements versus disclosed only in the footnotes. Finally, FV_Ext is 10 The variance inflation factors (VIF) across all of our specifications do not exceed 4, suggesting multicollinearity is not an issue (Neter et al., 1985). 21

23 insignificant (0.025, t-stat = 0.14). Thus, we fail to find evidence supporting H 2D that use of an alternative external monitor reduces audit fees. Overall, the results are consistent with audit fees decreasing in the firm s proportion of total assets reported at fair value, increasing in the difficulty to measure the fair values, and increasing if the fair value is recognized (versus only disclosed). In the next section, we exploit other settings to provide additional evidence on our two sets of hypotheses. 5 Alternative settings In this section, we reassess the results of our primary analyses by examining two alternative settings to test our predictions. We first compare UK versus US real estate firms to reassess H 1A and H 1B. We then use UK investment trusts to reassess H 2A and H 2B. Finally, we use UK manufacturing, real estate, and investment trust firms to reassess H 1A and H 2B. 5.1 The effect of reporting model on audit fees: evidence using UK and US real estate firms We compare audit fees for UK versus US real estate firms over the period , which provides an advantageous alternative setting. First, it maintains the within-industry setting, and thus the nature of the assets being examined. Second, the UK and US both have highly developed real estate industries, reflected in a large number of publicly traded real estate firms within each country. Third, this setting exploits the primary difference between the UK and US real estate industries: the financial reporting model. Specifically, real estate firms in the UK report property assets on the balance sheet at fair value: either as required under UK domestic GAAP prior to mandatory IFRS adoption, or as implemented under IAS 40 by this industry in the UK. In contrast, US real estate firms report property assets on the balance sheet at amortized cost, as US GAAP prohibits firms from reporting tangible assets, including real estate, at fair value. Further, industry practice in the US is such that 22