PRACTICE DIVERSIFICATION BY LARGE AUDIT FIRMS #

PRACTICE DIVERSIFICATION BY LARGE AUDIT FIRMS# GEORGE DELTAS RAJIB DOOGAR University of Illinois, U. C. University of Illinois, U. C. 1206 South ...
0 downloads 0 Views 315KB Size
PRACTICE DIVERSIFICATION BY LARGE AUDIT FIRMS#

GEORGE DELTAS

RAJIB DOOGAR

University of Illinois, U. C.

University of Illinois, U. C.

1206 South Sixth Street Champaign, IL 61820 [email protected] http://www.staff.uiuc.edu/~deltas

1206 South Sixth Street Champaign, IL 61820 [email protected] http://www.cba.uiuc.edu/doogar

JUNE 2003

#

We are indebted to Ira Solomon for numerous comments and suggestions. We also thank Rashad Abdel-Khalik, Kevan Jensen, Ken Koga, Kathryn Kadous, Clive Lennox, Linda Myers, Mark Peecher, Dave Ricchiute, Theodore Sougiannis, Martin Wu and seminar participants at UIUC, the 2003 mid-year auditing meeting, and ISAR 2003 for helpful discussions and Jonathan Hamilton and Public Accounting Report for sharing data on audit firm leverage and revenues. We alone are responsible for any errors.

PRACTICE DIVERSIFICATION BY LARGE AUDIT FIRMS Abstract This study contributes to our understanding of the structure of audit markets. In particular, we develop a framework based on a portfolio view of the firm’s audit practice and use it to investigate four important questions on the determinants of audit firm clienteles and audit firm mergers. We find that firms’ practice diversification strategies reflect the hypothesized trade-offs between riskdiversification and incentives to exploit scale economies. We also find systematic differences among large audit firms in the degree of practice concentration driven, in part, by differences in their internal organization. Further, all firms that were party to mergers (Big Eight to Big Six, Big Six to Big Five) were characterized by higher than average degree of practice concentration, while the resulting entities are characterized by lower than average degree of practice concentration. This suggests that a possible incentive for large audit firm mergers may have been the desire to diversify the practice. Finally, we examine and find evidence that the Ernst and Young merger was not followed by a change in the merged firm’s diversification strategy, while the Deloitte and Touche merger was followed by an increase in practice concentration.

Keywords: Practice Concentration, Auditor Industry Specialization, Audit Clienteles, Audit Firm Mergers. J.E.L.: L19, L84, M41. Data Availability: All data are obtained from publicly available sources identified in the text.

I. INTRODUCTION We examine practice diversification by large (Big Five and predecessor) audit firms during the period 1980-1998. Intuitively, a large firm’s clientele can be thought of as a collection of risky assets because the auditor’s payoff, net of expected litigation losses, from any single client is effectively a random variable (Simunic and Stein 1990). In this view, an audit firm’s clientele reflects a trade-off between the benefits of risk diversification and the loss of potential economies of scale in auditing similar clients. A portfolio-based approach thus appears to be a natural unifying framework for studying the economic determinants of audit firm clienteles. In this study, we develop a portfolio-based statistical framework that measures (clientele) practice diversification at the firm level and use it to analyze some aspects of the structure of audit markets. In particular, we study four important questions on the characteristics of equilibrium audit firm clienteles and on audit firm mergers. First, we examine the degree to which measures of client risk and economies of scale explain the composition of large audit firm practices. We investigate this issue by documenting the extent to which practice diversification strategies of large firms reflect the hypothesized trade-off between risk diversification and economies of scale. Second, since practice diversification is likely to vary with a firm’s product mix and production technology, we investigate whether firms differ systematically in the extent (degree) of practice diversification.1 Third, our statistical approach allows us to investigate whether diversification itself might be a motive for large

1

For example, Cushing and Loebbecke (1986) find significant differences in large firms’ audit methodologies. More specifically, they find differences in the degree of structure exhibited by large firms’ audit methodologies and they argue that the degree of structure has implications for the distribution of audit staff at various levels within the firm, i.e., the firm’s span of control. As we argue below, the firm’s span of control or leverage is also expected to have implications for the firm’s optimal degree of practice diversification. 1

audit firm mergers, unlike prior studies which have focused on economies of scale or market power. In particular, we examine whether mergers occur between firms with similar or dissimilar extent of diversification, or between firms of higher than average or lower than average degree of diversification. Finally, we examine whether the Big Eight to Big Six and Big Six to Big Five mergers were associated with changes in the degree of practice diversification by the merging firms. Our methods and findings should be useful to both researchers and regulators interested in better understanding strategic conduct in audit markets.2 We define the weight of an industry in a firm’s practice as the ratio of the firm’s share of the industry to the firm’s overall share of the audit market. We say a firm’s practice is perfectly diversified if every industry is assigned a weight of 1, i.e., if the firm’s market share in each industry is equal to its overall market share. In our statistical framework, the deviation of the observed weight of an industry in a firm’s practice from unity is the contribution of that industry to the firm’s overall departure from perfect diversification. For expositional convenience we refer to departure from perfect diversification as being equivalent to practice concentration.3

2

Section 701 of the Sarbanes-Oxley Act of 2002, for instance, requires the U.S. General Accounting Office (GAO) to study and review “... The factors that have lead to the consolidation of public accounting firms since 1989 and the consequent reduction in the number of firms capable of providing audit services to large national and multi-national business organizations subject to the securities laws, including the present and future impact of this consolidation on capital formation and securities markets, and the solutions to any problems discovered including ways to increase competition .” (GAO 2002). 3

Our analysis is the first empirical assessment of the extent and determinants of practice diversification by large U.S. audit firms. Kwon (1996) and Krishnan (2001) use Yardley et al.’s (1992) measure of above-average portfolio shares, i.e., industries with a larger than average share in the firm’s client portfolio, to define a specialist auditor. However neither study considers the measurement of overall (firm-wide) diversification. In contrast, our research focuses on the extent to which a firm’s presence in each industry is larger or smaller than might be expected given the firm’s overall size. This focus allows us to examine possible differences in audit firm 2

Our findings can be summarized as follows: First, we find evidence that the extent of departure from perfect diversification in a firm’s practice in any given industry is systematically related to measures of industry riskiness and returns to scale in auditing, in a way that is consistent with our theoretical predictions.4 Departures from perfect diversification are decreasing in the industryspecific audit risk associated with auditing clients in an industry. Conversely, departures from perfect diversification are more pronounced for industries in which returns to scale in auditing are greater. Second, we find that the extent of diversification varies significantly among audit firms, suggesting some degree of strategic differentiation in their approaches to constructing a large audit practice. We show that while firm organization structure (leverage or the staff-to-partner ratio) partially explains diversification, product mix (the proportion of non-audit-service revenue) does not. Third we find that the Big Eight to Big Six and Big Six to Big Five mergers occurred among firms with similar approaches to practice diversification: Firms with relatively more concentrated practices are the ones that chose to engage in merger activity. Finally, we ask whether the Big Eight to Big Six mergers were followed by changes in diversification strategies.5 We do so by combining the pre-merger practices of the constituents of the Deloitte and Touche and Ernst and Young mergers from the beginning of our sample period and comparing the degree of practice concentration of the post-

strategic behavior. 4

As we explain below, industries in which auditor business risk is high are likely to be characterized by distributions of auditor market shares similar to the distribution of market shares in the economy as a whole. By contrast, industries in which economies of scale in auditing are large are likely to be characterized by more extreme distributions of auditor market shares. 5

Since our sample period ends in 1998, we have only one year of data for the postPricewaterhouseCoopers merger period, which is too limited to permit any meaningful conclusions to be drawn about the impact of that merger. 3

merger firms with their pre-merger (combined) counterparts. We find that the Deloitte and Touche merger was followed by a reduction in diversification, while the Ernst and Young merger had no discernable diversification effect. The rest of the paper is organized as follows: Section II outlines the essence of our conceptual and econometric framework. Section III discusses in some detail economic forces likely to influence the determination of audit firm clienteles and section IV formally presents the statistical framework. Section V details the hypothesized relationships between departures of firm industry shares from those of a fully diversified practice and measures of economies of scale, client industry risk and client industry competitiveness. Data sources are identified in section VI, section VII reports results, and section VIII summarizes and concludes.

II. THE MODELING AND STATISTICAL FRAMEWORK IN A NUTSHELL The conceptual and econometric framework is presented in more detail in sections III and IV. In this section we illustrate the central premises of our analysis using a short example. We also explain below how our “complete” model differs from this stylized example. The central premise of our framework can be summarized as follows. We assume that the audit market can be divided into a set of sub-markets. On the one hand, both client audit risk and client business risk are correlated across clients in each sub-market. This correlation in risk provides an incentive for audit firms to diversify their practices across many sub-markets. On the other hand, developed audit expertise is partially “portable” across clients. This leads to returns to scale in auditing activity in any given sub-market, which in turn provides an incentive for audit firms to concentrate (or lump) their practices in a few sub-markets (rather than spread them equally across 4

all sub-markets). In this paper the audit sub-markets are defined on the basis of client industries, and in particular on the basis of two-digit SIC codes.6 A simple example can illustrate how differences in audit risk and audit economies-of-scale across industries affect the distribution of audit firm shares across industries. Suppose that there are only 2 auditors of equal economy-wide size, i.e., each with aggregate market share of ½. Suppose also that there are only three industries, each of equal size in terms of audit revenue/effort. Note that the practices of the two firms are fully diversified if each firm’s share in every sub-market is 50%, i.e., their “expected” or fully diversified market share in each sub-market is 0.5. Further suppose that the magnitude of audit economies to scale relative to audit risk is summarized by a variable X, with high values of X being associated with low risk/high economies-of-scale industries and low values of X being associated with high risk/low economies-of-scale industries. First, consider Industry A which is characterized by high audit risk/client business risk and no auditing returns to scale. For industry A, the variable X takes the value XA. We posit that in such an industry, auditor shares will be close to aggregate market shares. In particular, suppose that the industry share of firm 1 equals 0.55 and that of firm 2 equals 0.45 (see Figure 1). Each firm’s “excess” market share, defined as the absolute value of the deviation of its actual market share from its “expected” market share (i.e., from the firm’s fully-diversified market share), equals 0.05. Denote the ratio of the absolute value of a firms’s “excess” market share to the firm’s expected market share

6

In doing so, we follow a long tradition in auditing literature starting with Eichenseher and Danos (1981), Danos and Eichenseher (1982, 1986) and continuing to the present (see Hogan and Jeter 1999, and many others). More broadly, defining audit sub-markets on the basis of client industries is not limited to the U.S. audit market or to the use of SIC codes (see, for example, Dopuch and Simunic 1980, Craswell et al. 1995, and Ferguson and Stokes 2002). 5

by σ (since there are only two firms of equal size, σ will be the same for both firms). Thus, the value of σ for industry A is equal to 0.10 for both firms. Next, consider industry B1 which is characterized by low audit risk and significant audit returns to scale (for this industry the variable X takes the value XB). We posit that in such an industry, one firm (in this example firm 1) will emerge as the winner, pushing the other firm to the margin. In particular, suppose that the market share of firm 1 equals 0.75 while that of firm 2 equals 0.25 (see Figure 2a). This does not, however, imply that firm 1 goes long in all industries with high values of X, while firm 2 goes short in all such industries. In an otherwise identical industry B2, firm 2 may emerge as the winner, with a market share of 0.80 (see Figure 2b). Consequently, while the average market share of each firm in industries with a value of X equal to XB may be close to ½, in each individual industry firm shares will deviate substantially from ½. Nevertheless, notice that the value of σ for both industries is high, regardless of the identity of the dominant firm: in industry B1 σ equals 0.50 while in industry B2 σ equals 0.60.



A regression framework, such as that used in our study, can be used to investigate the extent to which shares deviate from “expected” shares (normalized by the expected share of the firm) as a function of X which captures industry risk and economies-of-scale attributes (in other words, how X explains the variation in σ across industries). On the basis of the numerical values given above, the positive estimated slope on variable X obtained from this regression would be consistent with

6

the central premise/hypothesis given at the start of this section that firms trade off the benefits of risk diversification against the benefits of economies of scale. Though the intuition embodied in the above example carries through to the “complete” model, the complete model presented in section III differs from the simple example presented above in a number of ways. First, following prior literature, we identify a number of characteristics, each of which is hypothesized to affect the risk and economy-of-scale attributes of the industry. Therefore, we estimate the importance of each of these risk and returns-to-scale industry attributes on the value of σ. Second, there are more than two firms in the economy, each with different economy-wide market shares. Therefore (i) the fully diversified sub-market shares differ across firms and (ii) in each sub-market there will be a different value of σ for each firm. In other words, unlike the example above, in the general model σ varies both across client industries and across audit firms. Third, firms may systematically differ in the way they trade off between risk and economies of scale. Therefore, the values of σ can be systematically related to the identity of the firm or to firm characteristics that affect the risk/economies-of-scale trade-off. In section IV, we develop an econometric framework that allows the simultaneous incorporation of all the above features.

III. EXPLAINING AUDIT PRACTICE DIVERSIFICATION In this section we outline in more detail our framework for understanding the determination of equilibrium audit firm clienteles. We argue that the equilibrium profile of audit firm client portfolios in the economy is determined both by client and auditor preferences. Clients may have varying preferences for sharing or avoiding common auditors or willingness to pay a premium for specialist auditors. Auditors face a trade-off between the benefits from diversification across industries and 7

the benefits of reaping economies of scale. Further, an auditor’s ability to exploit returns to scale and tolerance for risk may depend on firm characteristics, such as firm organization structure and product mix. Jointly, both client and auditor preferences influence the degree of a firm’s practice diversification or concentration. We discuss these factors in more detail below. Both theoretical reasoning and anecdotal evidence suggest multiple reasons that large audit firms would diversify their practices. First and foremost, auditor business risk and litigation are often tied to client business risk, so firms have incentives to limit their exposure to systematic client-industry risk. For instance, increased diversification (that is, decreased concentration) insures the firm from segment specific shocks in litigious industries (chemicals and pharmaceuticals, e-commerce and high technology, S&Ls, or REITs). Second, to maintain market share in a constantly evolving economy, a large audit firm is likely to seek foothold shares in every sector of the economy. Third, audit firms need to credibly communicate to the market that their knowledge bases remain current and comprehensive. Consequently, they may find it necessary to maintain a presence in a large number of industry segments to credibly communicate their status as a leading provider of audits.7 Fourth, the Big Five audit firms collectively audit the vast majority of clients going public in the U.S. capital markets. Since successful clients rarely change auditors, large firms have incentives to “acquire” IPO clients regardless of the client’s line of business in the expectation that some fraction of them will be successful, providing growth in future business. This process is likely, over time, to lead to increasingly diversified practices. Finally, clients in the same industry may have competitive reasons to avoid sharing auditors (Kwon 1996). These considerations constitute the benefits of diversification 7

Informally, one may think of this as “maintaining bragging rights”: in bidding for new clients, a firm that does not have at least a few “leading” clients in related sectors of the economy may be perceived as noncompetitive or not sufficiently credible to lenders and investors. 8

or, equivalently, the costs of practice concentration for large audit firms. In this paper, we focus on risk diversification and client industry competitiveness as the driving forces for practice diversification.8 There are, however, countervailing incentives pushing firms away from diversification and toward concentration. First, firms can reap economies of scale by developing a body of industryspecialist auditors. Generally accepted accounting and auditing standards vary from industry to industry and a great deal of the auditor’s work is directed towards ensuring that client financial statements comply with the accounting standards (e.g., GAAP) and the audit work complies with auditing standards (e.g., GAAS). Prior research has identified a number of regulated industries such as utilities, banks, mining and extractive operations which are subject to specialized accounting and auditing standards (Eichenseher and Danos 1981). Second, focusing on an industry may enable the auditor to develop a deeper understanding of the client’s business and therefore better manage client business risk. Recent models of audit risk suggest that successful auditing of complex businesses cannot be carried out absent a holistic understanding of the client’s business in the context of the industries in which the client operates (Bell et al. 1997). In this view, auditing multiple clients in an industry confers a positive externality by enabling the auditor to develop a deeper understanding of each client’s business in the context of the client’s industry as a whole. This deeper understanding both increases the quality of the audit and reduces the cost of providing it. Third, clients may prefer sharing auditors to signal their higher quality (Bernheim and Whinston 1985, Dye 1995, and Datar et al. 1991). If this is the case, establishing a large practice in an industry, i.e., market leadership by 8

Note, however, that the framework developed in this study can be extended to also investigate the importance of the other considerations for diversification using appropriate proxies for them. 9

an auditor, may result in an audit fee premium (Craswell et al. 1995). Collectively, these considerations constitute the benefits of practice concentration, or equivalently the costs of complete diversification. We argue that firm market shares in industries for which the incentives for concentration are relatively weak and the costs of concentration are high (chemicals, e-commerce and high-technology) are expected to be approximately equal to the aggregate, economy-wide, firm shares. Our intuition is as follows. Consider such an industry in which each large firm’s market shares are equal to its economy-wide shares. No firm would like to expand its share, as doing so would not result in lower audit costs than that of its rivals, while adding more clients in that industry increases the firm’s risk exposure. Therefore, if the client industry is large enough, the client allocation in a risky industry with negligible economies of scale should end up closer to the efficient risk-bearing allocation, i.e., industry shares proportional to firm size, and prices would adjust to reflect the risk auditors bear under this market allocation.9 Note that even though one might argue that auditors with larger than optimal exposure to concentrated industry risk can charge a higher fee premium to reflect the additional risk, such an incremental premium (over the premium charged by an optimally diversified auditor) is not sustainable in the long run.10 Consider next industries for which the incentives for

9

We cast the discussion here in terms of large industries because audits are indivisible, and therefore, industry shares in industries with a small number of clients will depart from aggregate market shares due to integer constraints. We take such effects into account in our analysis (see discussion on measures of client industry competitiveness in section V). 10

An implicit assumption here is that re-insurance markets are not totally efficient, i.e., we assume that there are some frictions, either through transaction costs or agency costs, in the re-insurance market. If that were not the case, then the allocation of risk to auditors would be irrelevant and therefore risk considerations would not lead to any incentives for practice diversification. 10

concentration are strong and the costs of concentration low (utilities, banks, mining and extractive industries). For such industries, firm market shares are expected to differ substantially from the aggregate, economy-wide, firm shares. The competition for clients in these industries has features of a “winner (or small number of winners) take all” contest. Consequently, the shares of the “winners” are substantially larger than their economy-wide counterparts, while the converse will be true for the shares of the “losers.” In other words, in industries characterized by larger economies of scale in auditing, firm market shares are likely to differ from their economy-wide market shares as some firms are able to dominate these markets by exploiting returns to scale, pushing other firms to lower than expected shares. Of course, the winners in some industries may be the losers in other industries that are also of a “winner-take-all” nature (cf. the example discussed in Section II). Our econometric methods are, in large part, chosen with these observations in mind. Though the trade-off between risk diversification and returns to scale is expected to affect the equilibrium profile of every firm’s clientele, firms are likely to differ in the degree to which they are averse to risk or able to exploit economies of scale. As a result, firms are likely to differ systematically in the degree of practice concentration, even after accounting for client industry characteristics. These differences are likely to be driven not only by differences in the strategic philosophy of the firm (which is a purely idiosyncratic feature of firm culture) but also by differences in the internal organization structure of the firm. For example, a firm that adopts a highly structured audit approach may be better able to exploit economies of scale by standardizing the audit of similar clients. However, highly structured firms may have a greater aversion to risky client industries.11 11

As Cushing and Loebbecke (1986, 43) point out, a highly structured approach is also more likely to be inflexible and “it cannot be applied easily to atypical audit environments” and may consequently “lead to . . . the formulation of an inappropriate audit opinion.” 11

There may be discrepancies between a firm’s desired and attained degree of practice concentration, possibly due to changes in competitive conditions (Mintzberg and Waters 1985, Mintzberg 1987). Because changing practice concentration is costly and takes time (there is substantial inertia in audit firm clienteles), firms may be motivated to achieve the desired degree of practice concentration by combining practices, i.e., through merger. A merger can affect the degree of practice concentration simply due to practice agglomeration (if firms differ substantially in the sectors in which they focus, the merged practice is likely to be markedly less concentrated). Further, a merger can affect the degree of practice concentration through a calculated or intended change in the strategic focus of the merged entity, as the merger might affect both the internal structure of the firm and its strategic philosophy. Our econometric analysis is chosen to also reflect the above considerations.

IV. MEASURING PRACTICE DIVERSIFICATION The first element of our framework is the construction, from first principles, of a measure of practice concentration, which we define as the degree of departure from perfect diversification. This measure defines the weight assigned to an industry in a firm’s portfolio to be the ratio of the firm’s market share in that industry to the firm’s economy-wide market share. The statistical framework we develop below postulates that the observed variability in the weights assigned to industries in a firm’s portfolio is driven by the trade-offs between the costs and benefits of diversification. In particular, let the size of the j th auditor’s total practice in year t, µ jt , be given by µ jt' G skt k0Kjt

12

(1)

where skt is the size of the kth client in year t and Kjt is the set of clients of auditor j in year t. The size of a client can be (i) one, in which case µ jt is the simple client count, (ii) the total assets or sales of a client, or, (iii) any index of client activity which would be a proxy for the amount of audit effort (where traditionally audit effort is modeled as a concave function of client size or activity). The size of industry i, Sit , is defined by Sit ' G skt k0Kit

(2)

where Kit is the set of clients in segment i and year t. The size of the (audited) economy in year t, St, is St ' j Sit

(3)

i

If auditor j’s practice is distributed across all I industries in proportion to their size, then the size of its practice in industry i equals Sit

µˆ ijt '

St

µ jt

(4)

This practice is fully diversified (i.e., it exhibits no practice concentration). As the firm’s practice becomes less diversified (more concentrated) the actual size of practice will be larger than µˆ ijt in some industries while in others it will be smaller than µˆ ijt . Then, the actual size of auditor j’s clientele in industry i and year t, µ ijt , can be written as µ ijt ' µˆ ijt ε˜ ijt

(5)

2

where ε˜ ijt is a random variable with variance σijt and expected value of 1. In other words, for nonperfectly diversified portfolios, µˆ ijt can be thought of as the expected (anticipated) size of the firm’s 2

practice in industry i. An unbiased estimate of σijt is given by

13

2 σˆ ijt

'

µ ijt µˆ ijt

2

& 1

(6)

Note that µ ijt /µˆ ijt is the relative weight of industry i in firm j’s portfolio in year t. In our 2

specification, the variance of the relative weights, σijt , is a function of auditor identity, industry segment, and time. Before we proceed to the discussion of our hypotheses and tests, it is instructive to contrast our research with the rather large literature on auditor industry specialization. The specialization literature generally uses measures of market leadership, e.g., firm market share in a given industry, as proxies for specialization (e.g., Craswell, Francis, and Taylor 1995, Ferguson and Stokes 2002).12 In contrast, we examine the weight assigned to an industry by a firm within the context of that firm’s overall practice. The two views are complementary in that share-based leadership measures represent a client’s-eye view of the leading auditor in an industry, while our portfolio-weights approach reflects an audit-firm’s-eye view of various client industries, enabling the researcher to study the entire pattern of an audit firm’s market presence.13 Moreover, by controlling for differences in the absolute sizes of a firm’s clientele this approach permits comparisons of practice diversification strategies across firms of varying size.

12

In this literature, an auditor is considered a specialist if its market share in a particular segment exceeds a cutoff point. Traditionally, this cutoff has been set at 10%, while recently (after the 1989 mergers) it has been 15%. There is some movement to more “continuous” measures, in which auditors are differentiated with respect to the degree of specialization (Dunn et al. 2000). Palmrose (1986), by contrast, uses a combination of rank and market share based criteria to identify industry specialists (or market leaders). 13

Craswell et al. (1995) and Willenborg (2002) discuss strengths and weaknesses of share-based leadership measures. 14

An example may help highlight the differences between the two approaches. Consider an auditor who by traditional measures is classified as a “non-specialist” in a particular industry but for whom that industry represents a larger share of its entire practice than that of a (bigger) firm classified as a specialist in that industry.14 The smaller “non-specialist” auditor’s relative involvement in the industry is greater than that of the “specialist” auditor. Our measure highlights this relative practice concentration. Traditional market-share-based measures of specialization on the other hand, are insensitive to this aspect of an audit firm’s strategic choice to concentrate or diversify its practice in particular sectors of the economy.15

V. METHODS: VARIABLES AND HYPOTHESES Preliminaries and Overview. To examine the extent to which deviations in auditor market shares can be explained by auditor and market segment characteristics, and time, we estimate regressions of the form

14

This is not a fine point. For almost the entire period of our study, the largest Big firm is almost twice as large as the smallest of these firms. Notwithstanding these size differences, the Big firms audit almost every client industry segment. Consequently, the larger Big firms are a priori more likely to show up as specialists using absolute market share based measures. For example, the results in Table 5 of Hogan and Jeter (1999) show that Arthur Andersen, which has the largest absolute market share of any firm during the period covered by that Table, has the largest number of specializations. 15

In effect, our measure suggests an alternative concept of specialization, one that is based on the extent to which firm shares in an industry deviate from those of a fully diversified practice. Summing the absolute value of the deviation of firms industry-shares from their perfectly diversified shares across all firms in an industry yields a measure of industry dominance that takes into account the entire (i.e., economy-wide) practice profile of every firm as opposed to traditional measures (such as HHI, CR4, and CR8) which depend only on firm shares in that industry. Indeed, one could in principle carry out our statistical analysis using the practice profile of a subset of the Big Five firms (and their predecessors), while the traditional analysis requires the market shares of all auditors in a number of industries. 15

log σˆ ijt ' α % β Xijt % u˜ijt

(7)

where Xijt is a set of regressors which represents market segment and auditor characteristics (including, in some specifications, audit firm leverage, proportion of consulting revenues, and indicator variables for auditor identity and time).16 We compute σijt using each of three measures of market share (number of clients, log client assets and square-root of client sales) and estimate (7) using multiple specifications. These regressions examine the extent to which industry and firm characteristics explain departures from perfect diversification. They also allow us to examine changes in diversification due to merger-induced practice agglomeration effects as well as changes in post-merger diversification strategy. Since practice diversification may vary across auditors for firm-specific reasons as well as client industry characteristics and over time, all three sets of factors motivate our choice of regressors Xijt .17 We follow prior literature in our use of particular market and auditor characteristics as proxies for underlying variables of interest (such as risk and audit-economies-to scale). Our choice of proxies is extensive, but, of course, it could be complemented by the incorporation of additional variables in our regression framework. In the remainder of this section we discuss, in turn, our choice of auditor characteristics, client industry-specific characteristics and period indicator variables used as regressors.

In our sample, the value of σˆ ijt is never zero, so taking the log does not result in any missing observations. 16

17

We confine our tests to the Big Six and their predecessor firms because these few firms are in a class of their own. 16

Regressors based on Auditor Characteristics Each audit firm can choose to organize production in a unique way, be it in the degree of centralization or structure imposed on the audit (Cushing and Loebbecke 1986), the use of leverage (Doogar and Easley 1998), or the use of different audit approaches (Bell et al. 1997, Bell et al. 2002). Moreover, large audit firms differ in the extent to which they provide non-audit services. Differences in centralization, audit approach and product mix can affect a firm’s trade-off between exploiting economies of scale and its willingness to bear risk in audit markets. Therefore, these differences in internal firm structure might result in observable differences in firm strategies with respect to the degree of practice concentration. We use firm-specific indicator variables and controls for the nature of the firm’s practice (Leverage, i.e. staff-to-partner ratios, and %MCS, i.e., proportion of non-audit fees) to test whether firm diversification strategies differ across firms and, if they do, how strategies change over time and in response to mergers.18 Since merger is often either a precursor to, or a consequence of, strategic change (Porter 1980, Prahlad and Hamel 2000), we investigate whether mergers are followed by changes in a firm’s practice diversification strategy. To examine whether a merger changes the merging firms’ diversification strategy itself, we re-compute the portfolio weights and deviations for a hypothetical 2

“merged” firm. That is, in constructing σijt from equations (1) through (6), we assume that the parties to the merger have been merged throughout our sample. We then re-estimate the set of regressions in (7) including as regressors merger indicator variables and their interaction with time.

18

Though in the long-run clientele composition and the staff-to-partner ratio is interrelated, in the short-run the resources (especially partners) under a firm’s disposal are likely to drive the composition of its clientele (at the margin) rather than vice-versa (see Doogar and Easley, 1998). Therefore, endogeneity of leverage (in econometric terms) is not an issue. 17

Creating such hypothetical firms allows us to compute the effect of merger on the strategy of the combined firm controlling for the agglomeration effect that is purely an artefact of the merger process: merging two firms while holding their clienteles fixed is likely to increase the diversification of the merged firm relative to that of the constituent firms. Due to the highly concentrated supply side of the U.S. audit industry (Dopuch and Simunic 1980, Simunic 1980, Danos and Eichenseher 1986, Doogar and Easley 1998), mergers among large audit firms are a subject of considerable and ongoing attention (Minyard and Tabor 1991, Copley 1992, Wooten et al. 1994, Ivancevich and Zardkoohi 2000, Sullivan 2002, GAO 2002). Prior research on audit firm mergers has investigated whether these mergers were conducted to extend the reach of the firm or to increase market power in select markets. By contrast, our analysis investigates practice diversification as a motive for merger. In particular, we examine whether the Big Eight to Big Six and Big Six to Big Five mergers occurred between firms with similar or dissimilar degrees of practice concentration, or between firms of higher than average or lower than average degrees of practice diversification.

Client Industry Specific Regressors Prior research on the determinants of audit firm market shares, audit market concentration, and auditor specialization suggests numerous client characteristics that may be used as measures of potential economies of scale in audit production, client industry competitiveness and client industry risk. Moreover, since audits of most U.S. public entities are usually indivisible, the lumpiness of the

18

client industry can also affect the extent of practice diversification (as we measure it).19 In what follows we first discuss factors likely to affect the magnitude of economies of scale followed by factors which might reflect industry competitiveness, and finally, factors that might reflect client industry risk. Given that the unit of observation is at the firm-industry-year level, all of our measures are either industry characteristics or industry averages of the respective variable (measured at the individual client level). Determinants of Returns to Scale in Auditing First and most obviously, incentives to specialize may vary across market segments because of client technological characteristics. A principal concern of the auditor is to ensure that the client’s financial statements provide reasonable values for the assets, liabilities and earnings of the client. We expect auditors to enjoy larger economies of scale resulting from specialized knowledge of industry conditions in industries in which the principal assets of the client are not tradeable or are thinly traded, i.e., in fixed-asset intensive industries and in industries with significant intangible assets (knowledge, or equivalently, human-capital intensive industries).20 Conversely, economies of scale due to specialized knowledge of the sort discussed above are likely to be smaller in industries

19

If an industry is characterized by a few large clients, some auditors may be shut out simply because there are only a limited number of audits to go around. 20

Thin markets create difficulties for the auditor in estimating initial, carrying and terminal values of assets. For instance, Lev (2001, 3) notes: “ ... I trace the measurement and reporting problems of intangibles to the unique attributes of these assets: high risk, lack of full control over benefits and absence of markets.” Similar arguments relating to fixed-asset valuation in general and asset-impairment accounting in particular have been advanced in the auditing and financial statement analysis literature (Ricchiute 1998, 569-72; Revsine et al. 2001, 475-77). 19

in which the principal assets are readily marketable or are commodities.21 We use asset fixity (the ratio of non-current to total assets, AFix) as a measure of fixed-asset intensity and the natural log of Market to Book Value of Equity (log(MTB)) as a measure of intangible intensity (Lev 2001, 8; Hall 2001).22 [In computing industry average MTB scores, we excluded all observations for which the ratio was negative since a negative BTM ratio makes no economic sense.] Similarly, regulated industries (REG) call for a detailed knowledge of applicable regulations, leading to pronounced economies of scope relative to unregulated industries. Prior studies have found higher auditor concentration in regulated industries, as it only pays a few firms to invest the substantial resources necessary to master the industry-specific regulatory regime (Danos and Eichenseher 1986). We follow the classification used in Hogan and Jeter (1999) to identify regulated industries. We expect asset fixity, log of employee compensation and regulation to be positively related to the degree of departure from perfect diversification.

21

Examples of such industries might be retail trade, traditional manufacturing, transportation, and agriculture. However, each of these industries may also exemplify other influences such as regulation, litigation, or financial riskiness to different degrees. 22

U.S. generally accepted accounting principles (GAAP) do a poor job of measuring (valuing) intangibles since most R&D activity is expensed in the year in which the outlay occurs. Moreover neither the book value of R&D or patenting costs are good estimates of the value of self-generated intangible assets. Fama and French (1992), Lakonishok et al. (1994), Chan et al. (1995, 2001) have used the Market-to-Book ratio as a proxy for the firm’s growth expectations which is equivalent to imputing a premium attributable to unrecorded or intangible assets. Since the generation of intangible assets is usually a function of the quality of human capital employed, we also used the natural log of average employee compensation as an alternative measure of the importance of human capital in the client’s input mix. The explanatory power of this variable, like that of log(MTB), was very weak. Therefore, we do not further discuss any of these results. 20

Determinants of Client Industry Competitiveness The degree of departure from perfect diversification is also likely to be influenced by demand side factors, such as the degree of competitiveness in the client industry. Oligopolistic rivalry in client markets is more likely to lead to higher diversification as the main competitors are unlikely to have the same auditor. For example, Coke and Pepsi, or General Motors and Ford do not share auditors (see Kwon 1996). Since the auditor of one of the big clients is effectively shut out from the other big clients, the ability of that auditor to gain a larger than expected market share is diminished. Traditionally, industry concentration (a frequently used proxy for industry competitiveness) is measured by the Herfindahl-Hirschman Index (HHI), the value of which is a combination of the number of clients and the inequality in their relative sizes. We distinguish between the two components of the HHI by including separate measures of client inequality and industry size because each component affects diversification through a different mechanism. Other things being equal, larger industries should be more diversified because the scale of operations at which returns to scale fade out is likely to be smaller relative to the size of the industry, permitting several large firms to compete effectively in a larger industry. Moreover, holding inequality in client size constant, auditor selection is less likely to be affected by clients’ unwillingness to share auditors.23 Inequality in client size, holding industry size constant, increases the likelihood that some of the clients will be direct

23

Since clients are indivisible, an industry cannot have more auditors than it has clients. Thus, industries with only a few clients are likely to have a lumpier distribution of auditor market shares. As a result, from a purely statistical point of view, small industries are likely to result in a more marked departure from perfect diversification. This makes it important to directly control for industry size independent from the distribution of client sizes. Furthermore, we perform our analysis both using the entire sample of industries and a sub-sample of industries with a large number of clients (n>30 and n>60). The results are qualitatively similar to those using the entire sample on all variables except the number of clients. 21

competitors and therefore unwilling to share auditors.24 In this paper, we use the natural log of the number of clients in an industry (log(nSIC2)) as our measure of industry size, and the adjusted Gini coefficient as our measure of inequality in client sizes (adjGINI).25 Determinants of Client Industry Riskiness We now turn to a discussion of the risk dimensions likely to affect diversification incentives. First, industries with high risk of auditor litigation are unlikely to attract a disproportionate weight in any firm’s portfolio. We measure client industry litigiousness (LITIG) by a set of industry dummies identified by recent studies (Palmrose 1986, Hogan and Jeter 1999). Consistent with prior research and in accordance with the discussion in section III, we expect litigation to decrease the likelihood that a firm will “go long” in these industries. Consequently, we expect auditors to have a presence in these industries that is proportionate to their overall economy-wide size. We expect higher client financial and business risk measures to have the same effect as litigation risk, i.e., to be associated with industry market shares that are close to the firm’s overall economywide share. We use Altman’s (1968) Bankruptcy (Z) Score as a measure of overall client financial

24

For example, if there 10 clients in a industry of all of equal size, each of these clients is not likely to care about the identity of the auditor of the other nine clients. If, however, 3 of the 10 clients are dominant firms, then these 3 clients would likely not want to share auditors. This effect is likely to be stronger for industries with a large number of clients. Suppose that an industry only has two clients. Then, the relative shares of these two clients are not likely to be important in determining their aversion to sharing auditors. In our empirical analyses we restrict our attention to industries with more than 30 clients, so that the effect of GINI on practice concentration is expected to be negative. 25

The Gini is the most widely used measure of inequality in economics, political science and sociology. It is equal to the expected value of the difference between the sizes of a randomly chosen pair of firms in an industry scaled by the average firm size in that industry. It is however, biased when the number of firms in the industry is small. The adjusted Gini is used to account for this small sample bias. See Deltas (2003) for a discussion of the Gini coefficient, its smallsample properties and the adjustment that corrects for the small-sample bias.. 22

condition. We also use Acquisition Expenses (ACQ) as a risk factor since acquisition activity is associated with increased financial statement risk (Palmrose 1991). To the extent that particular sectors of the economy are more prone to this sort of business reorganization they would pose higher auditor business risk and therefore lead auditors to diversify away from those sectors. Prior research also identifies several client financial characteristics as client business risk measures (Lev and Thiagarajan 1993, Brealey et al. 1999, Choi et al. 2002). Smaller client size, lower liquidity, solvency, profitability, and low asset turnover are commonly identified sources of risk and therefore likely to lead to higher diversification (lower portfolio weights). We choose Relative Client Sales (Individual Client Sales/Average Economy-Wide Client Sales, RelSls) as a client size measure, Current Ratio (log(CR)) and Cash Flow/Total Liabilities (CFTL) as liquidity/solvency measures, Return on Total Assets (ROA) as a profitability measure, and Asset Turnover defined as Sales/Total Assets (ATurn) as an activity measure. Since a number of the variables are financial ratios, their distributions are highly non-normal with very long tails. In the regression analyses reported below, we use logarithmic scaling of a number of these variables to improve model fit. Since Altman’s findings suggest that clients with Z scores above 3 are considered “safe” and thus possibly qualitatively different than clients with lower Z scores (Altman 1968; Penman 2001, 730), we use the fraction of clients with Z scores below 3 (Bankruptcy) as our measure of industry risk of Bankruptcy. In other words, the larger the fraction of clients with Z scores below 3, the more risky the industry. Since the Z-score is primarily used to assess the riskiness of manufacturing or industrial borrowers, we set the industry Bankruptcy score to zero for all industries with 2 digit SIC codes greater than 59. Finally, since both CFTL and ROA

23

have extreme outliers, we also estimate spline models which allow for the marginal effect of these variables to differ over the range of the variables.

Time Indicator Variables as Regressors There may be trends in the degree of practice concentration even after controlling for the auditor and industry characteristics. Further, the degree of practice concentration may depend on the number of firms. We therefore break the period of our study into five periods, 1980-1984 (Period1), 19851988 (Period2), 1989-1993 (Period3), 1994-1997 (Period4) and 1998 (Period5) to examine if there is a systematic change in firm behavior over time. We choose these periods to coincide with the dates at which the number of active auditors in the economy change and to also yield roughly balanced windows around the 1989 Big Eight to Big Six mergers. Table 1 lists all the regressors used in the analysis and the expected direction of their effect on deviations from perfect diversification.



VI. DATA Our sample consists of all clients of the Big Five (and predecessor) firms for which annual financial data for the period 1980-1998 are available in the Compustat database. We also collected data on audit firm partner and staff and product mix from Public Accounting Report for the period 19901998. Table 2 reports descriptive statistics for the variables of interest. Table 3 sets out the market shares of the firms over the entire sample period. Each of the firms audits between seven and twentytwo percent of all U.S. entities issuing publicly traded securities and each has clients in almost every 24

sector of the U.S. economy. During this period three mergers occurred: in 1989, Deloitte, Haskins and Sells (DHS) and Touche Ross (TR) merged to form Deloitte and Touche (DT) and Arthur Young (AY) and Ernst and Whinney (EW) merged to form Ernst and Young (EY) reducing the number of large firms from eight to six. In 1997, Price Waterhouse (PW) and Coopers and Lybrand (CL) merged to form PricewaterhouseCoopers (PwC), further reducing the set to five.



VII. MODELS AND RESULTS Analyzing the Degree of Practice Concentration To investigate the questions outlined above, we estimate the following base model (Model (1a)) log(σijt) '

β1 % β2 AFixit% β3 REGi % β4 log(MTBit) % β5 AdjGiniit % β6 log(nSIC2it) % β7 LITIGi % β8 log(Acq it) % β9 Bankruptcy % β10 RelSls it % β11 ROAit % β12 ATurn it % β13 log(CRit) % β14 CFTL it % β15 PERIOD2 t% β16 PERIOD3t % β17 PERIOD4 t % β18 PERIOD5t % β19 AAjt % β20 AY jt % β21 CL jt % β22 KPjt % β23 DT jt % β24 DHSjt % β25 EY jt % β26 EWjt % β27 PWjt % β28 PwCjt % εijt

where i indexes client industries, j indexes audit firms, t indexes time and εijt is a disturbance term of mean zero. The variables are as defined in Table 1. Note that TR is the omitted firm dummy. We construct log(σijt) using log-asset-weighted market shares and estimate the above model via OLS after dropping industries with 30 or fewer clients to mitigate effects arising from client

25

indivisibility.26 The means of σijt and log(σijt) are 0.32 and -3.20. Therefore, the average percentage deviation of a firm’s share in an industry from that implied by a fully diversified practice is 32%. This shows that audit firms’ practices are far from being fully diversified. The interpretation of some of the results below is facilitated by knowledge of the standard deviations of σijt and log(σijt) . These are 0.29 and 2.33, respectively. The results of the regression are reported in Table 4.



Table 4 shows that audit risk and industry characteristics affect diversification strategies as hypothesized. Measures associated with the presence of returns to scale in auditing a sector are either positive and statistically significant ( AFixit and REGi ) or not statistically significant ( log(BTMit) ). To quantify the “economic significance” of the results, we also calculate the standardized coefficients of the regression (not reported) [see Chatterjee and Price 1978, Maddala 1977]. The standardized coefficients are the change in the expected value of the dependent variable (measured in standard deviations of the dependent variable), when the value of a regressor increases by one standard deviation of that regressor. The 0.496 coefficient on AFixit (average client asset fixity in the sector) corresponds to a standardized coefficient of 0.067, which implies that an increase in the value of AFixit by one standard deviation (of AFixit ) leads to an increase of the expected value of log(σijt) by 0.067 standard deviations (of log(σijt) ). The 0.182 coefficient of the binary variable REGi implies that expected value of σijt in a regulated industry is 18.2% higher than the expected value 26

We also perform our analysis both using the entire sample of industries and a subsample of industries with more than 60 clients. The results using all three samples are quantitatively similar to each other except for the coefficient of log(nSIC2it) . 26

of σijt in an otherwise identical unregulated industry. Both of these results show that the returns to scale in auditing have a quantitatively important effect on the standard deviation of departures of firm market shares from the market shares implied by perfect practice diversification. Asymmetries in the client size in an industry (measured by AdjGINIit ) are strongly associated with smaller values of log(σijt) : An increase in AdjGINIit by one standard deviation leads to a reduction in log(σijt) by 0.082 standard deviations. This result is consistent with the hypothesis that dominant clients in an industry would prefer not to share auditors, making it more difficult for a single audit firm to dominate that industry. The number of clients in an industry is strongly (and negatively) associated with log(σit) , consistent with the idea that in large industries most audit firms will have sufficiently large clienteles to exhaust economies of scale. Recall that the coefficient of log(nSIC2it) is expected to have a negative sign also due to the indivisibility of audit engagements. Measures associated with the auditor’s exposure to client and client industry risk are either negative and statistically significant ( LITIGi and log(Acqit) ) or not statistically significant ( Bankruptcyit , RelSlsit , ROAit , ATurnit , log(CRit) , and CFTLit ).27 The standardized coefficient of log(Acqit) is -0.032, implying that a one standard deviation increase in log(Acqit) decreases the expected value of log(σijt) by 0.032 standard deviations of log(σijt) . The parameter estimate of

27

The variables ROAit and log(CFTLit) have long tails and we investigate the possibility that they do have an effect on log(σit) at the tails of their distribution by fitting a spline regression. This regression allows for coefficients that vary over the range of the variables, while preserving the continuity of the regression plane with respect to these variables (see Johnston 1984). For the most part, almost all the slopes associated with these two regressors remain statistically not significant. 27

LITIGi shows that the expected value of σijt in a litigious industry is 21.2% lower than the expected value of σijt in an otherwise identical but non-litigious industry. 28 The period indicator variables are not significant, suggested that the degree of practice concentration is not changing over time or as the number of large audit firms declines. In contrast, there appears to be significant heterogeneity in the degree of practice concentration among the Big Five audit firms and their predecessors. The auditor indicator variables are jointly statistically significant at any conventional level of significance. Note that in our specification we consider merged firms as different entities than either of their predecessor firms, allowing for change in the extent of diversification arising from the agglomeration of practices or from post-merger strategic repositioning. The three least concentrated practices are those of PricewaterhouseCoopers, Ernst and Young, and KPMG. The three most concentrated practices are those of Arthur Young, Touche Ross (the omitted indicator variable), and Ernst and Whinney. Notice that two of the least concentrated practices belong to entities created by merger, while none of the three most concentrated practices belong to merged firms. Further, all three mergers create firms that are more diversified than either of their constituents. The fact that merged firms are more diversified than other firms and more diversified than their predecessor firms is not surprising. If there is no post merger strategic repositioning of the practice of the merged firm, practice agglomeration (holding everything else constant) is likely to increase diversification. Moreover, if mergers are taking place between

28

The fact that most of the financial risk variables are not significant (once industry litigiousness and acquisitions are taken into account) may be due to the fact that auditors can adjust audit procedures to minimize the litigation risk arising from publicly observed measures of client financial distress. 28

randomly selected firms, the set of merged firms would have practices that are less concentrated than the average firm. Interestingly, it appears that mergers are not taking place between randomly selected firms. Indeed, the firm indicator variables reported in Table 4 show that the merged firms have practices that are more diversified than those of the average firm despite the fact that the mergers have taken place between firms that have more concentrated practices than the representative (contemporaneous) firm. In particular, EW and AY have the sixth and eighth most concentrated practices among the Big Eight auditors. Similarly, DHS and TR have the fourth and seventh most concentrated practices among the Big Eight auditors. Finally, PW and CL have the sixth and fourth most concentrated practices among the Big Six firms. The fact that the outcome of these mergers leads to firms that have very diversified practices, relative to the other firms, implies either that practice diversification effect of agglomeration is substantial relative to the inter-firm differences in the degree of practice concentration, or that the post-merger strategic repositioning leads to increased diversification, or both. In the next subsection, we measure the effect due to strategic repositioning after the Big Eight to Big Six mergers and show that, if anything, it leads to increased practice concentration. In Model (2) we try to explain the inter-firm differences in the degree of practice concentration as a function of key firm organizational and strategic attributes. We select Leveragejt (the staff to partner ratio) as a measure of the firm’s ability to exploit economies of scale. Partners of firms which have concentrated practices are more likely to be repositories of specialized knowledge and, therefore, be able to supervise a larger staff (Doogar and Easley 1998, Beckmann 1987). In other words, because audit engagements in a firm that has a concentrated practice are more likely to be 29

similar to each other relative to those of a firm that is fully diversified, the former type of firm requires fewer partners per staff member. Therefore, we expect a positive association between Leveragejt and log(σit) . An audit firm’s diversification strategy may also be a function of its product mix. We, therefore, include the proportion of firm revenues earned from providing management consulting services ( %MCSfeejt ) as another potential influence on firm diversification. A number of observers have suggested that audit firms’ business strategies have been significantly influenced by their focus on providing management consulting services (e.g. SEC 2000). If this hypothesis were correct, one would expect to find a negative association between %MCSfeejt and the standard deviation of departures of firm market shares from the market shares consistent with perfect practice diversification. The underlying intuition is that if an auditor were focused on pursuing profitable non-audit engagements independent of the economics of auditing, the impact of economies of scale would be diluted.29 Table 4 shows that Leveragejt is statistically significant and positively associated with log(σit) (as expected). Further, the relationship is also quantitatively important, with a normalized coefficient of 0.11. On the other hand, the coefficient of %MCSfeejt , though of the expected sign, is not statistically significant. In brief, our evidence is consistent with leverage increasing, and the mix of audit fees not affecting, audit practice concentration.

29

If, however, the provision of consulting services is subject to the same type of returns to scale as is auditing, i.e., if the variables we used in our regression also explain economies of scale in consulting, then we would not find a significant effect of consulting. Unfortunately, we cannot test this hypothesis because we do not have fees for consulting services broken down by industry. 30

Sensitivity Analyses Our first robustness check is to estimate Model (1b) which has the same specification as Model (1a), except that it is estimated via a median regression. The reported coefficients now show the effect of the regressors on the median value of log(σit) rather than their effect on the mean of log(σit) . The two sets of estimates will generally differ if the error distribution is not symmetric. If the error distribution is symmetric, then the median regression estimates are more robust than the linear regression estimates to measurement error in the dependent variable, but less efficient than the linear regression estimates under the null of normally distributed errors terms.30 The median estimation results are qualitatively and quantitatively similar to those of Model (1a). As a second robustness check of Model (1a), we investigate the stability of the parameter estimates when we relax the assumption of independent and identically distributed error terms (implied by the OLS and median regression procedures). For this dataset, the most important departure from independence is likely to be due to serial correlation. Generalized Least Squares estimation of Model (1a) assuming, say, an AR1 error process, is hampered by the fact that the industry time-series often has breaks (in part because when the number of clients in an industry drops below 30, that industry is dropped from our sample for that period). We take a conservative approach and assume that there is a single disturbance draw for each auditor-industry history (equivalent to assuming an intertemporal correlation of 1). The standard errors are calculated via bootstrapping by resampling with replacement entire auditor-industry histories. The standard errors are approximately twice the size of those obtained using OLS, but even under this (conservative) assumption of perfect 30

The relative efficiency of the two estimation approaches depends on the distribution of the error term. For example, if the error term is Laplace distributed, the median regression is efficient. 31

temporal dependence most of the statistically significant regressors remain statistically significant. Further, when we limit ourselves to regressions in which we use a truncated set of proxies for industry and client riskiness (by dropping ATurnit , ROAit , log(CRit) , and CFTLit ) and also eliminate RelSlsjt and log(MTBit) which are not statistically significant, statistical significance of the remaining variables is not affected even when standard errors are computed via bootstrapping. We next investigate whether the results are sensitive to our measure of auditor market shares. The results described above are based on the assumption that audit effort is proportional to the logarithm of the client’s assets. In particular, we re-estimate Models (1a) and (2) assuming that audit effort is the same for all clients (unweighted models). We also re-estimate Model (1a) assuming that audit effort is proportional to the square root of the client’s sales (square-root-of-sales weighted model). The results, reported in Table 5, are qualitatively and quantitatively similar to those of Table 4. Therefore, we do not discuss them in detail, except to note that Leveragejt is no longer statistically significant when auditor market shares are based on the client count. The R-squared of the squareroot-of-sales weighted model is much lower than that of the unweighted and log-of-assets-weighted models, suggesting that the square-root-of-sales is not a good measure of audit effort (assuming that the model is otherwise correctly specified).



32

Counter-factual Merger Analysis In this sub-section we discuss the results from the regressions in which we combine the merging firms’ clienteles retroactively since the beginning of our sample period and examine whether the degree of practice concentration is affected by the merger. More concretely, we recompute equations (1) through (6) assuming that the DT and EY mergers took place at the beginning of our sample period. By retroactively combining the practices from the beginning of the sample, we eliminate the agglomeration effect of the mergers. Therefore, any change in the extent of practice concentration after the merger will arise from a change in the diversification strategy of the merged firm. We have only one year of data for PricewaterhouseCoopers, which is not sufficient to estimate the effects of merger on practice diversification given the possible time lags in changing the merged firm’s practice. For this reason, we omit the last year of data and investigate only the strategic effects of the Big Eight to Big Six mergers. Price Waterhouse, which does not undergo any merger activity during the 1980 to 1987 sample period, is the omitted auditor indicator variable. This allows us to combine the practices of EY and AY under the label EY and those of DT and TR under the label DT. We divide the post-merger period into two sub-periods (1989 to 1993 and 1994 to 1997) and include indicator variables for each merged firm for each of the two post-merger periods. The estimated regression, Model (1c), is given below.

33

log(σijt) '

γ1 % γ2 AFixit% γ3 REGi % γ4 log(MTBit) % γ5 AdjGini it % γ6 log(nSIC2it) % γ7 LITIGi % γ8 log(Acqit) % γ9 Bankruptcy % γ10 RelSlsit % γ11 ROAit % γ12 ATurnit % γ13 log(CRit) % γ14 CFTLit % γ15 PERIOD2t% γ16 PERIOD3 t % γ17 PERIOD4t % γ18 AAjt % γ19 CLjt % γ20 EY jt % γ21 DT jt % γ22 KPjt % γ23 EY jt PERIOD3t % γ24 EY jt PERIOD4t % γ25 DTjt PERIOD3 t % γ26 DTjt PERIOD4 jt % εijt

Model (1c) has been estimated using log-asset-weighted, unweighted, and square-root-of-sales weighted measures of firm market shares. Table 6 reports the estimation results using OLS. The results of the median regressions and splines yield results very similar to the OLS specification and are omitted for brevity.



Overall, the signs and significance levels of coefficients of variables common to Models (1c) and (1a) are very similar across models. Table 6 shows that EY’s overall diversification strategy has not changed significantly as a result of the merger. On the other hand, DT’s practice seems to have become more concentrated after the merger. This difference in strategic behavior might possibly be driven by the relative size of DT. Table 3 shows that DT is by far the smallest of the Big Six firms. While no audit firm as large as DT (which audits more than 12% of our sample of large U.S. publicly traded clients) can be called a niche or boutique firm, it is nevertheless interesting to see from Table 3 that over our sample period DT’s share has also shown the most significant erosion. Consequently,

34

DT’s post-merger strategy of increasing its practice concentration seems to be consistent with that of a smaller competitor trying to select a few strong areas to build upon. We conclude the discussion of the results by drawing some implications from the comparison of the results of the standard and counter-factual analyses. Recall from the discussion of Table 3 that merged firms are amongst those with the most diversified practices, despite having emerged from constituent firms that had some of the most concentrated practices. This change in concentration can be decomposed into a practice agglomeration effect and a strategic merger effect. The results reported in Table 5 show that the strategic effect is neutral or towards greater practice concentration. Therefore, the increase in practice diversification among the merged firms is due to the practice agglomeration effect, and this effect is quantitatively large relative to the differences in the degree of practice concentration among the set of existing (pre-merger) firms.31 To summarize, mergers appear to be a vehicle through which auditors increase diversification, but the increase in diversification arises from the practice agglomeration effect and not because mergers facilitate the strategic repositioning of the merged firm’s practice.

31

Roughly speaking, the agglomeration effect of the EY merger is the difference between the average of the coefficients of AY (0.025) and EW (-0.051) in Table 4 minus the coefficient of EY in Table 4 (-0.374) plus the average of the coefficients on EYPERIOD3 (-0.056) and EYPERIOD4 (-0.074) in Table 6. This equals 0.296, or, 82% of the total effect of 0.361 documented in Table 4. The corresponding figures for the agglomeration effects of the DT merger are 0.412, or, 312% of the total effect of 0.132 documented in Table 4. The agglomeration effect is quantitatively large because the coefficients of pre-merger constituent firms in Table 4 differ substantially from those of the corresponding post-merger firm relative to the average difference in the coefficients between two randomly selected firms. We cannot determine whether the agglomeration effect is larger than it would have been in a merger between two randomly selected firms. Such determination requires a different analytical approach than that used in this paper and is left for future research. 35

VIII. SUMMARY AND CONCLUSIONS Both theory and intuition suggest that audit firms’ clienteles are likely to reflect a tradeoff between economies of scale and scope and diseconomies of concentrated risk-bearing as well as client preferences and audit firm characteristics. In this study we develop a framework based on a portfolio view of a firm’s audit practice which can be used to analyze large audit firm clienteles. We use this framework to formulate and test predictions on four important questions that have not been addressed in prior research on audit firm clienteles and audit firm conduct. First, we investigate the importance of factors related to economies of scale and client industry risk in determining the composition of large audit firm practices. Second, we examine whether an auditor’s production technology (audit structure, audit firm leverage) or its product mix (proportion of non-audit-service revenues) affect diversification strategies. Third, we investigate whether diversification itself may potentially have been a motive for large audit firm mergers. Finally, we investigate whether these mergers resulted in changes in the merged firm’s practice diversification strategy. We find that on average, departures from perfect diversification in an audit firm’s practice are related positively to measures of scale economies and negatively to various client-industry risk factors. Firms differ in their practice diversification strategies. In particular, a higher staff-to-partner ratio (leverage) tends to be associated with an increase in the weight assigned to some industries, i.e., it tends to lead to greater practice concentration. This result is consistent with higher leverage translating into greater ability to exploit economies of scale in audit production. The fraction of nonaudit service revenues, however, does not appear to be systematically related to the degree of audit practice diversification. In other words, unless the factors that drive scale economies in auditing and consulting services are the same, the evidence is not consistent with a search for non-audit revenues 36

driving audit practice diversification strategies. Our merger analysis shows that the Big Eight to Big Six and Big Six to Big Five audit firm mergers were between firms with higher than average degree of practice concentration and that all three mergers led to substantial increases in practice diversification. We show that for the Big Six to Big Eight mergers this effect was not driven by strategic repositioning of the merged firm, but rather due to the effect of agglomerating the clienteles of the constituent firms. Finally, the post-merger strategies of EY and DT appear to differ significantly: EY’s degree of practice concentration does not appear to have changed after the merger. DT’s post-merger practice, on the other hand, appears to become more concentrated. While the specific findings of this study are interesting in their own right, our methods, measures and approach are quite general. They should be useful in further research on audit firm conduct and competition in audit markets. In particular it will be interesting to revisit some of the results on auditor specialization arrived at using market-share-based measures of specialization using the practice concentration concept. Moreover, the analysis of the impact of mergers on practice diversification offers a perspective quite different from that offered by share-based measures in that it allows us to distinguish between the agglomeration and strategic repositioning effects.

37

References Altman, E. 1968. Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance 23(4): 589-609. Beckmann, M. 1987. Tinbergen Lectures on Organization Theory. Springer. New York. T. Bell, J. Bedard and K Johnstone. 2002. K-RISK: A computerized decision aid for client acceptance and continuance risk assessments. Auditing a Journal of Practice and Theory forthcoming. T. Bell, F. Marrs, I. Solomon and H. Thomas. 1997. Auditing Organizations Through a StrategicSystems Lens: The KPMG Business Measurement Process. Montvale, New Jersey: KPMG. Bernheim, B.D. and M.D. Whinston. 1985. Common Marketing Agency as a Device for Facilitating Collusion. Rand Journal of Economics 16(2): 269-81 Brealey, R.A., S.C. Myers and A.J. Marcus. 1999. Fundamentals of corporate finance. 2nd edition, Irwin, Boston, MA. Chan, L.K.C., N. Jegadeesh, and J. Lakonishok. 1995. Evaluating the Performance of Value Versus Glamour Stocks: The Impact of Selection Bias. Journal of Financial Economics 38 (3): 269-296. Chan, L.K.C., J. Lakonishok, T. Sougiannis. 2001. The Stock Market Valuation of Research and Development Expenditures. Journal of Finance 56 (6): 2431-56. Chatterjee, S. and B. Price. 1977. Regression Analysis by Example. Wiley, N.Y. Choi, J-H., R. Doogar and A.R. Ganguly. 2002. Audit firm client portfolios and changes in audit liability regimes: Evidence from the US audit market. Working paper. University of Illinois at Urbana-Champaign. Copley, P. 1992. An assessment of the potential effect of big eight mergers on competition in the market for audit services. Advances in Accounting 1: 185-205. Craswell, A.T., J.R. Francis and S.L. Taylor. 1995. Auditor brand name reputations and industry specializations. Journal of Accounting and Economics 20(3): 297-322. Cushing, B.E. and J.K. Loebbecke. 1986. Comparison of Audit Methodologies of Large Accounting Firms. Studies in Accounting Research 26. American Accounting Association, Fla. Danos, P. and J.W. Eichenseher. 1982. Audit Industry Dynamics: Factors Affecting Changes in Client-Industry Market Shares. Journal of Accounting Research 20 (2): 604-616. 38

Danos, P., and J.W. Eichenseher. 1986. Long Term Trends Toward Seller Concentration in the U.S. Audit Market. The Accounting Review 61 (4): 633-650. Datar, S.M., G.A. Feltham and J.S. Hughes. 1991. The role of audits and audit quality in valuing new issues. Journal of Accounting and Economics 14 (1): 3-49. Deltas G. 2003. The small sample bias of the Gini Coefficient: Results and Implications for Empirical work. The Review of Economics and Statistics. 85: 226-234. Doogar, R. and R.F. Easley, 1998. Concentration without differentiation: A new look at the determinants of audit market concentration. Journal of Accounting and Economics 25: 235-253. Dopuch, N. and D. Simunic, 1980. Competition in auditing: an assessment. Symposium on Auditing Research IV (University of Illinois, Urbana-Champaign): 401-450. Dunn, K., B.W. Mayhew and S.G. Morsefield. 2000. Auditor Industry Specialization and Client Disclosure Quality. Manuscript, University of Wisconsin. Dye, R.A. 1995. Incorporation and the audit market. Journal of Accounting and Economics 19 (1): 75-114. Eichenseher, J. W. and P. Danos. 1981. The analysis of industry specific auditor concentration: Towards an explanatory model. The Accounting Review 56(3): 479-492. Fama, E. F. and K. R. French. 1992. The Cross-Section of Expected Shock Returns. Journal of Finance 47 (2): 427-465. Ferguson, A. and D. Stokes. 2002. Brand name audit pricing, industry specialization and leadership premiums post Big 8 and Big 6 mergers. Contemporary Accounting Research 19(1): 77-110 GAO (United States General Accounting Office). 2002. Call for Research. Communication by Abraham Akresh, Assistant Director U.S. General Accounting Office to the Auditing Section of the American Accounting Association, e-mail dated September 18, 2002. Hall, R. 2001. The Stock Market and Capital Accumulation. American Economic Review 91 (5): 1185-1202. Hogan, C.E. and D.C. Jeter. 1999. Industry specialization by auditors. Auditing: A Journal of Practice an Theory 18(1): 1-17. Ivancevich, S.H., and A. Zardkoohi, 2000. An Exploratory Analysis of the 1989 Accounting Firm Megamergers. Accounting Horizons (December): 380-402.

39

Johnston, J, 1984. Econometric Methods, 3rd Edition. McGraw Hill, New York. Krishnan, J. 2001. A comparison of auditors’ self-reported industry expertise and alternative measure of industry specialization. Manuscript, Temple University. Kwon, S.Y., 1996. The impact of competition within the client’s industry on the auditor selection decision. Auditing: A Journal of Practice an Theory 15(1): 53-70. Lakonishok, J., A. Shleifer, and R. W. Vishny. 1994. Contrarian Investment, Extrapolation, and Risk. Journal of Finance 49 (5): 1541-1578. Lev, B. 2001. Intangibles. Brookings Institute, Washington D.C. Lev, B. and S.R. Thiagarajan. 1993. Fundamental Information Analysis. Journal of Accounting Research 31 (2): 190-215. Maddala, G.S. 1977. Econometrics. McGraw-Hill, N.Y. Mintzberg, Henry. 1987. The Strategy Concept I: Five Ps for Strategy. California Management Review 30(1): 11-24. Mintzberg, H. and J.A. Waters. 1985. Of Strategies, Deliberate and Emergent. Strategic Management Journal 6(3): 257-272. Minyard, D., and R. Tabor. 1991. The effect of Big Eight mergers on auditor concentration. Accounting Horizons (December): 79-90. Palmrose, Z-V., 1986. Audit fees and auditor size: Further evidence. Journal of Accounting Research 24(1): 97-100. Palmrose, Z-V., 1991. Trial of legal disputes involving independent auditors: Some empirical evidence. Journal of Accounting Research 29 (Supplement): 149-185. Penman, S.H. 2001. Financial Statement Analysis & Security Valuation. New York: McGrawHill. Porter, M. 1980. Competitive Strategy: Techniques for Analyzing Industries and Competitors. New York: The Free Press., 1980. Prahalad, C.K. and G. Hamel 1990. The core competence of the corporation. Harvard Business Review68(3); 79-91.

40

Revsine, L., D.W. Collins and W.B. Johnson. 2001. Financial Reporting and Analysis. 2nd ed. Upper Saddle River, NJ: Prentice Hall. Ricchiute, D.N. 1998. Auditing and Assurance Services. 5th ed. Cincinnati, OH: Southwestern. SEC (Securities and Exchange Commission). 2000. Revision of the Commission’s Auditor Independence Requirements: Final Rule. 17CFR Parts 210 and 240. File Number S7-13-00. Securities and Exchange Commission, Washington, D.C. Simunic, D. 1980. The pricing of audit services: theory and evidence, Journal of Accounting Research (18): 161-190. Simunic, D.A. and M.T. Stein. 1990. Audit risk in a client portfolio context. Contemporary Accounting Research 6(2): 329-343. Sullivan, M. 2002. The effect of the Big Eight audit firm mergers on the market for audit services. Journal of Law and Economics, 45(2): Part 1, forthcoming. Willenborg, M. 2002. Discussion of “Brand name audit pricing, industry specialization and leadership premiums post Big 8 and Big 6 mergers”. Contemporary Accounting Research 19(1): 111-116. Wooten, C., S. Tonge and C. Wolk. 1994. Pre and post Big Eight mergers: Comparisons of auditor concentration. Accounting Horizons (September): 58-74. Yardley J.A., N.L. Kaufmann, T.D. Cairney, and W.D. Albrecht. 1992. Supplier behavior in the U.S. Audit Market. Journal of Accounting Literature 11: 151-184.

41

Table 1. List of Regressors and Expected Signs.(a) Expected signs shown in column 2. All numbers within parentheses are annual Compustat item numbers. Auditor related variables. AA to PWC Firm dummies (TR is the omitted firm) Leverage + Staff per partnerb %MCS -Fraction of firm-wide revenues earned from Management Consulting Servicesb MERGER 0 if no merger or pre-merger, 1 if post-merger. Time related variables Period1 1 if year=1980-1984, 0 otherwise (omitted, base, case). Period2 1 if year=1985-1988, 0 otherwise. Period3 1 if year=1989-1993, 0 otherwise. Period4 1 if year=1994-1997, 0 otherwise. Period5 1 if year=1998, 0 otherwise. Measures of Returns to Scale in Auditing AFix + Non-current to Total Assets [1 - {Current Assets (14) ÷ Total Assets (6)}]. REG + 1 if 2-digit SIC code is regulatedc log(MTB) + Natural log of [Market Value of Equity (24*25) ÷ Book Value of Equity (computed as TA (6) - TL (181))]. Measures of Client Industry Size and Competitiveness adjGINI -The adjusted Gini based on client sales. LOGnSIC2 -log of the number of audited clients. Measures of Client Industry Risk LITIG -1 if 2-digit SIC code has high litigation riskc Bankruptcy -Fraction of clients in industry with value of Altman’s (1968) z-score less than 3 if client SIC code is less than 6000, 0 else. Altman’s Z score is computed as: [1.2*((CA(14)-CL(5)) ÷ TA(6)) + 1.4*(RE(36) ÷ TA(6)) + 0.6*(MVEQ(24)*(25) ÷TL(181)) + 0.998*(Sales(12) ÷ TA(6))]. log(Acq) -Natural log of Acquisition expenditures (129). RelSls + Average client sales (normalized by the size of the economy). ATurn -Asset Turnover [Sales (12)÷Total Assets(6)]. ROA -Return on Assets [Earnings before interest & taxes(170+15)÷Total Assets (6)]. log(CR) + natural log of Current Ratio [Current Assets (14) ÷ Current Liabilities (5)]. CFTL -Ratio of Cash Flow (123+125) to Total Liabilities (181). Notes: (a) All segment level variables are client averages by SIC-code and year. Interactions with period dummies are not listed as separate variables. Expected sign refers to the sign of the coefficients of equation (1). (b) data obtained from Public Accounting Report. (c) As reported in Hogan and Jeter (1999).

42

Table 2: Variable Distributions This table reports values of independent variables used in the analysis at the individual client level. The total sample consists of all clients for which financial information for some subset of variables listed in Table 1 was available in the 1999 Compustat Database. Number of Standard Variable Observations Mean Deviation Minimum Maximum Sales ($ millions) 142035 926.4359 4896.737 -107.639 195805 Total Assets ($ millions) 142811 1301.069 8632.857 0 668641 Current Assets/ Total Assets 129112 0.525854 0.270572 -0.256 26.19 Current Ratio 128910 3.401758 18.22441 -33 3104 EBIT/Total Assets 136504 -0.09786 7.949306 -922 1625 Sales/Total Assets 141872 1.215445 5.171538 -6.538 1625 Cash Flow/Total Liabilities 138803 0.010895 13.64885 -1711.2 2588.401 Acquisitions ($ millions) 134155 17.26792 197.6297 -4971 18610 Z Score (Altman) 126343 238.9376 85074.39 -73227.5 30200000 Market to Book ratio 100541 2.963133 72.85899 -6424.081 9532.632 Regulation 142871 0.275122 0.4465775 0 1 Litigation 142871 0.371265 0.4831448 0 1

43

Table 3 Audit Firm Market Shares Panel A: 1980 to 1998 Big Eight/Big Six/Big Five audit firm market shares based on number of clients audited, by auditor. Year 1980 1981 1982 1983 1984 1985 1986 1987 1988* 1989 1990 1991 1992 1993 1994 1995 1996 1997** 1998

AA 14.5% 14.1% 14.6% 14.7% 14.8% 14.8% 14.1% 13.7% 14.2% 14.3% 14.7% 15.4% 15.2% 15.6% 15.8% 16.1% 16.7% 17.0% 17.2%

AY/EY EW 9.5% 6.3% 9.5% 6.3% 9.3% 6.3% 9.6% 6.5% 9.5% 6.8% 9.5% 7.0% 9.7% 7.4% 9.9% 7.6% 9.9% 7.5% 17.3% 17.6% 17.6% 17.7% 17.4% 17.5% 17.8% 18.1% 18.2% 18.3%

Auditor DHS/DT TR 6.8% 7.7% 6.1% 8.2% 6.8% 8.2% 7.1% 8.0% 7.0% 8.1% 7.2% 8.0% 7.3% 7.6% 7.7% 7.3% 7.7% 7.4% 15.1% 15.0% 14.5% 13.5% 13.2% 12.5% 11.9% 12.2% 12.5% 12.4%

CL PW/PwC 9.6% 9.6% 9.1% 8.8% 9.8% 8.8% 9.8% 8.7% 9.7% 8.7% 9.7% 8.5% 9.5% 8.3% 10.0% 8.4% 10.3% 8.3% 10.4% 8.3% 10.4% 8.2% 10.4% 8.8% 11.0% 9.1% 11.5% 9.5% 11.9% 9.8% 12.1% 10.0% 11.7% 9.9% 10.8% 10.2% 20.9%

KPMG 10.2% 10.2% 10.0% 9.9% 9.9% 10.0% 11.0% 13.2% 13.0% 13.3% 13.8% 14.1% 14.1% 13.8% 14.1% 14.5% 14.5% 14.5% 14.0%

Other 25.2% 26.8% 25.5% 25.1% 24.7% 24.7% 23.6% 21.1% 20.6% 20.1% 18.8% 18.0% 18.0% 18.1% 17.4% 16.8% 16.4% 16.3% 16.9%

Unaudited 0.6% 0.9% 0.7% 0.6% 0.8% 0.7% 1.4% 1.0% 1.2% 1.2% 1.3% 1.4% 1.4% 1.0% 0.9% 0.7% 0.6% 0.4% 0.3%

Total 4480 5686 6081 6346 6462 6790 7199 7331 7197 7023 7011 7173 7393 8040 8528 9390 9645 9350 8923

Legend: AA=Arthur Andersen; AY=Arthur Young, EY=Ernst & Young, EW=Ernst & Whinney; CL=Coopers & Lybrand, PW=Price Waterhouse, PwC=PricewaterhouseCoopers; DHS=Deloitte, Haskins & Sells, DT=Deloitte & Touche, TR=Touche Ross; KPMG=KPMG Peat Marwick. * In 1988, Arthur Young (AY) and Ernst & Whinney (EW) merged into Ernst & Young (EY) and Deloitte, Haskins & Sells (DHS) and Touche Ross (TR) merged into Deloitte & Touche (DT). ** In 1997, Coopers & Lybrand (CL)and Price Waterhouse merged into PricewaterhouseCoopers (PwC).

44

Table 3 (continued) Audit Firm Market Shares Panel B: 1980 to 1998 Big Eight/Big Six/Big Five audit firm market shares based on log of total assets audited, by auditor. Year 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998

AA 15.0% 14.6% 15.3% 15.5% 15.5% 15.4% 14.8% 14.4% 14.9% 15.1% 15.5% 16.0% 15.8% 16.1% 16.3% 16.5% 17.0% 17.4% 17.6%

AY/EY EW 9.8% 6.5% 9.8% 6.5% 9.6% 6.5% 9.8% 6.6% 9.7% 6.9% 9.8% 7.1% 9.9% 7.6% 10.1% 7.7% 10.2% 7.6% 17.7% 18.0% 18.0% 18.2% 17.8% 17.9% 18.3% 18.5% 18.5% 18.6%

Auditor DHS/DT TR 7.1% 7.7% 6.5% 8.2% 7.2% 8.1% 7.4% 8.0% 7.3% 8.1% 7.5% 8.0% 7.6% 7.6% 7.9% 7.4% 7.9% 7.4% 15.3% 15.3% 14.7% 13.8% 13.5% 13.0% 12.3% 12.6% 12.8% 12.8%

CL PW/PwC 9.8% 10.2% 9.4% 9.4% 10.0% 9.3% 10.1% 9.2% 10.0% 9.2% 10.0% 9.0% 9.9% 8.8% 10.4% 8.9% 10.6% 8.8% 10.7% 8.8% 10.9% 8.8% 10.8% 9.3% 11.3% 9.6% 11.8% 9.9% 12.2% 10.2% 12.4% 10.4% 12.1% 10.2% 11.2% 10.6% 21.6%

KPMG 10.6% 10.5% 10.3% 10.1% 10.1% 10.3% 11.2% 13.4% 13.1% 13.5% 14.1% 14.3% 14.4% 14.1% 14.4% 14.8% 14.7% 14.8% 14.3%

Other 22.8% 24.3% 23.1% 22.7% 22.5% 22.4% 21.4% 18.9% 18.5% 17.9% 16.5% 15.7% 15.8% 15.9% 15.4% 14.8% 14.4% 14.3% 14.9%

Unaudited Total 0.6% 78743 0.8% 99350 0.6% 106617 0.5% 111437 0.7% 113690 0.5% 118896 1.2% 125990 0.8% 128969 0.9% 127215 0.9% 124721 1.0% 124855 1.1% 127917 1.1% 131825 0.8% 144619 0.6% 154543 0.5% 170132 0.4% 176274 0.3% 172337 0.2% 165453

Legend: AA=Arthur Andersen; AY=Arthur Young, EY=Ernst & Young, EW=Ernst & Whinney; CL=Coopers & Lybrand, PW=Price Waterhouse, PwC=PricewaterhouseCoopers; DHS=Deloitte, Haskins & Sells, DT=Deloitte & Touche, TR=Touche Ross; KPMG=KPMG Peat Marwick. * In 1988, Arthur Young (AY) and Ernst & Whinney (EW) merged into Ernst & Young (EY) and Deloitte, Haskins & Sells (DHS) and Touche Ross (TR) merged into Deloitte & Touche (DT). ** In 1997, Coopers & Lybrand (CL)and Price Waterhouse merged into PricewaterhouseCoopers (PwC).

45

Table 4 Determinants of Audit Firm Diversification 1980-1998 – Log(Assets)-weighted Models Regression coefficients and standard errors (SE) for models explained in the text. Significance levels are as follows: *** = significant at 1%, (** = at 5%, * = at 10%). Model (1a) Model (2) Model (1b) Coefficient SE Coefficient SE Coefficient SE AFix 0.496*** 0.116 0.336* 0.187 0.594*** 0.117 Returns to Scale REG 0.182*** 0.040 0.340*** 0.062 0.211*** 0.040 Measures log(MTB) 0.041 0.026 0.016 0.043 0.030 0.026 Client Industry adjGINI -1.143*** 0.237 -0.920*** 0.354 -1.006*** 0.239 Competitiveness -0.279*** 0.048 -0.317*** 0.031 and Size Measures log(nSIC2) -0.273*** 0.031 LITIG -0.212*** 0.063 -0.260** 0.104 -0.125** 0.063 log(Acq) -0.028** 0.012 -0.045** 0.021 -0.038*** 0.013 Bankruptcy -0.048 0.068 -0.097 0.102 0.002 0.068 RelSls -0.006 0.006 -0.012 0.011 -0.014** 0.006 Client Industry Risk Measures ROA -0.024 0.016 0.018 0.024 -0.039*** 0.015 ATurn -0.017 0.019 0.057 0.039 0.001 0.019 Log(CR) -0.005 0.034 0.049 0.055 -0.023 0.034 CFTL -0.002 0.015 -0.004 0.020 -0.003 0.014 PERIOD2 0.013 0.041 0.037 0.042 PERIOD3 -0.074 0.045 -0.104** 0.046 Time Indicators PERIOD4 0.033 0.050 0.179*** 0.065 0.018 0.050 PERIOD5 0.048 0.091 0.092 0.097 -0.095 0.091 AA -0.315*** 0.069 -0.224*** 0.070 AY 0.025 0.077 0.126 0.078 CL -0.252*** 0.069 -0.111 0.070 DHS -0.084 0.075 0.000 0.076 DT -0.174** 0.085 -0.057 0.086 Auditor Indicators EW -0.051 0.075 0.050 0.076 EY -0.374*** 0.085 -0.247*** 0.086 KP -0.355*** 0.069 -0.267*** 0.070 PW -0.072 0.069 -0.025 0.070 PWC -0.503*** 0.188 -0.260 0.189 Leverage 0.040** 0.018 Auditor % MCS -0.007 0.005 Attributes MERGED 0.062 0.057 Regression N 6013 2488 6013 2 Statistics adjusted R 0.11 0.11 0.08 See Table 1 for variable definitions. Legend: AA=Arthur Andersen; AY=Arthur Young, EY=Ernst & Young, EW=Ernst & Whinney; CL=Coopers & Lybrand, PW=Price Waterhouse, PwC=PricewaterhouseCoopers; DHS=Deloitte, Haskins & Sells, DT=Deloitte & Touche, TR=Touche Ross (omitted, base case); KPMG=KPMG Peat Marwick. # Results based on 1990-1998 data. Leverage and fee data not available for earlier years.

46

Table 5 Determinants of Audit Firm Diversification 1980-1998 – Unweighted and Sales-weighted Models Regression coefficients and standard errors (SE) for models explained in the text. Significance levels are as follows: *** = significant at 1%, (** = at 5%, * = at 10%). Unweighted Models Square-root-of-Sales # Weighted Model (1a) Model (1a) Model (2) Coefficient SE Coefficient SE Coefficient SE AFix 0.506*** 0.117 0.356* 0.185 0.640*** 0.112 Returns to Scale REG 0.201*** 0.040 0.326*** 0.061 0.065* 0.039 Measures log(MTB) 0.044* 0.026 0.033 0.042 0.043* 0.025 Client Industry adjGINI -1.050*** 0.238 -1.012*** 0.349 -0.284 0.230 Competitiveness -0.299*** 0.047 -0.273*** 0.030 and Size Measures log(nSIC2) -0.297*** 0.031 LITIG -0.202*** 0.063 -0.248** 0.103 -0.065 0.061 log(Acq) -0.040*** 0.013 -0.055*** 0.021 -0.006 0.012 Bankruptcy -0.068 0.068 -0.081 0.101 0.079 0.066 RelSls -0.007 0.006 -0.005 0.011 -0.012** 0.006 Client Industry Risk Measures ROA -0.017 0.016 0.015 0.024 -0.031** 0.015 ATurn -0.014 0.020 0.032 0.038 0.011 0.019 Log(CR) 0.001 0.034 0.045 0.055 -0.038 0.033 CFTL -0.005 0.015 -0.011 0.020 -0.001 0.014 PERIOD2 0.039 0.042 -0.004 0.040 PERIOD3 -0.071 0.045 -0.045 0.044 Time Indicators PERIOD4 0.067 0.050 0.180*** 0.064 -0.086* 0.048 PERIOD5 0.068 0.091 0.086 0.096 -0.133 0.088 AA -0.291*** 0.069 -0.288*** 0.067 AY 0.055 0.078 -0.106 0.075 CL -0.193*** 0.070 -0.226*** 0.067 DHS -0.017 0.076 -0.124* 0.073 DT -0.126 0.085 -0.207** 0.082 Auditor Indicators EW 0.015 0.076 -0.275*** 0.073 EY -0.284*** 0.085 -0.472*** 0.082 KP -0.279*** 0.069 -0.490*** 0.067 PW 0.005 0.070 -0.132** 0.067 PWC -0.452** 0.189 -0.471*** 0.182 Leverage 0.019 0.018 Auditor % MCS -0.003 0.005 Attributes MERGED 0.031 0.056 Regression N 6013 2488 6013 Statistics adjusted R2 0.11 0.12 0.08 See Table 1 for variable definitions. Legend: AA=Arthur Andersen; AY=Arthur Young, EY=Ernst & Young, EW=Ernst & Whinney; CL=Coopers & Lybrand, PW=Price Waterhouse, PwC=PricewaterhouseCoopers; DHS=Deloitte, Haskins & Sells, DT=Deloitte & Touche, TR=Touche Ross (omitted, base case); KPMG=KPMG Peat Marwick. # Results based on 1990-1998 data. Leverage and fee data not available for earlier years.

47

Table 6 Effects of Big Eight to Big Six mergers on diversification based on comparisons of the post-merger practice with the combined pre-merger practices of the constituent firms. Regression coefficients and their standard deviations for model (1c) explained in the text. Significance levels are as follows: * = significant at 1%, † = significant at 5%, ‡ = significant at 10%. Log assets Weighted Unweighted Sales weighted Coefficient SE Coefficient SE Coefficient SE AFix 0.569*** 0.127 0.568*** 0.128 0.629*** 0.123 Returns to Scale REG 0.268*** 0.043 0.278*** 0.044 0.147*** 0.042 Measures Log(MTB) 0.029 0.028 0.041 0.028 0.022 0.027 Client Industry adjGINI -0.853*** 0.259 -0.895*** 0.261 -0.245 0.250 Competitiveness LOGnSIC2 -0.310*** 0.034 -0.316*** 0.035 -0.245*** 0.033 and Size Measures LITIG -0.240*** 0.069 -0.214*** 0.069 -0.165** 0.067 Log(Acq) -0.014 0.014 -0.026* 0.014 0.005 0.013 Bankruptcy 0.024 0.074 0.050 0.075 0.066 0.072 RelSls -0.017*** 0.007 -0.023*** 0.007 -0.006 0.006 Client Industry Risk Measures ROA -0.014 0.017 -0.012 0.017 -0.019 0.016 ATurn 0.011 0.021 0.009 0.021 0.024 0.021 Log(CR) 0.021 0.037 0.020 0.037 -0.002 0.036 CFTL -0.012 0.016 -0.011 0.016 -0.007 0.015 PERIOD2 -0.009 0.047 0.007 0.048 -0.076* 0.046 Time Indicators PERIOD3 -0.052 0.050 -0.066 0.051 -0.090* 0.049 PERIOD4 0.006 0.054 0.040 0.054 -0.080 0.052 AA -0.250*** 0.054 -0.306*** 0.054 -0.153*** 0.052 CL -0.180*** 0.054 -0.198*** 0.054 -0.094* 0.052 Pre-merger EY -0.196*** 0.069 -0.195*** 0.070 -0.278*** 0.067 Auditor Indicators DT -0.372*** 0.069 -0.511*** 0.070 -0.262*** 0.067 KP -0.286*** 0.054 -0.284*** 0.054 -0.363*** 0.052 EYPERIOD3 -0.056 0.101 -0.012 0.102 0.042 0.098 EYPERIOD4 -0.074 0.107 -0.115 0.108 -0.203* 0.104 Post-merger Auditor Indicators DTPERIOD3 0.151 0.101 0.273*** 0.102 0.256*** 0.098 DTPERIOD4 0.408*** 0.107 0.505*** 0.108 0.106 0.104 Regression N 4956 4956 4956 Statistics adjusted R2 0.11 0.12 0.08 See Table 1 for variable definitions. Legend: AA=Arthur Andersen, EY=Ernst & Young, CL=Coopers & Lybrand, PW=Price Waterhouse (omitted, base case), DT=Deloitte & Touche, KPMG=KPMG Peat Marwick.

48

Figure 1. Market shares in industry A, a “risky” industry with no economies of scale.

49

Figure 2a. Market shares in industry B1, a low-risk industry with substantial economies of scale.

Figure 2b. Market shares in industry B2, another low-risk industry with substantial economies of scale.

50

Suggest Documents