Passive Investors and Audit Quality: Evidence from the U.S.

Abstract

The rise of index funds, or passive investing, in recent decades has caused heated debates over the efficacy of passive investors' stewardship role in corporate governance. Our study adds to this emerging line of literature by examining whether passive investors enhance the quality of financial statement audits, a key aspect of corporate governance mechanisms. We exploit the yearly Russell index reassignment, which provides us with an ideal setting to study the causal relation between passive institutional investors (i.e., index trackers) and firms' audit quality. Examining firms closely surrounding the Russell 1000/2000 cutoff line, we find that higher passive ownership leads to higher audit quality proxied by audit fees. To investigate the channel through which passive investors influence audit-related governance issues, our evidence from auditor ratification voting records suggests that passive investors do ‘voice’ their opinion on low-quality audits. Such effort also leads to a higher likelihood of auditor turnover in the following year. In the cross-sectional analysis, we also find that the positive effect of passive investors on audit quality is more pronounced in firms with higher agency costs. Thus, our study supports the view that passive investors play an active role in improving corporate governance.

Keywords:

• Auditing
• Passive ownership
• Russel index
• Audit fee
• Audit quality

JEL Codes:

• M41
• G10
• G30

Because the funds’ holdings tend to be long term in nature (in the case of index funds, we’re essentially permanent shareholders), it’s crucial that we demand the highest standards of stewardship from the companies in which our funds invest. (Vanguard, 2016)

It makes sense for us to try to raise the ocean in order to lift our boat. (Richard Koppes, former chief counsel, CalPERS, 21 March 1996)

Introduction

The rise of passive investing in recent years has transformed ownership patterns in the U.S. stock market. In 2019, the size of assets managed by U.S. index funds and ETFs surpassed that of assets in active stock funds for the first time in history (Gittelsohn, 2019). Such a dramatic change has provoked heated debates over how companies are governed when the stock market is run by passive managers, who may lack the incentive and resources to actively exercise their fiduciary duties. On the one hand, the permanent nature of passive investors may lead to higher incentives for monitoring, active voting, and behind-the-scenes engagements (Appel et al., 2016; Crane et al., 2016; Gomtsian, 2019; Jahnke, 2019). On the other hand, some scholars and commentators have questioned whether index funds deliver on their stewardship promises.¹ While the debate over the efficacy of passive investors’ stewardship role is still in its early stages, in this study, we attempt to add further empirical evidence to this emerging line of literature.

We examine the impact of passive investors on corporate governance through the lens of financial statement audit quality. Identifying the causal impact of passive ownership on firms’ audit quality is a challenging task. This is because factors that directly affect audit quality – such as managerial capacity and firm transparency – may affect passive ownership. As Hermalin and Weisbach (2001) note, endogeneity issues are inherent in most governance-related archival research. To our knowledge, prior literature in auditing has not examined the impact of passive ownership per se, whereas studies focusing on whether institutional ownership affects audit/financial reporting quality are often subject to the same endogeneity concerns (Kane & Velury, 2004; Velury et al., 2003; Velury & Jenkins, 2006).

To overcome this challenge and to examine whether passive ownership causes any change in audit quality, we follow Appel et al. (2016) to exploit variations in ownership by passive mutual funds around the Russell 1000 and Russell 2000 indexes. Every year, Russell ranks all U.S. listed companies by size. The Russell 1000 and 2000 indexes include the largest 1,000 and next largest 2,000 firms, respectively. Russell assigns a portfolio weight that is value-weighted according to each stock’s market capitalization. For passively managed mutual funds that are benchmarked on the Russell 1000/2000 index, there is a sharp shift in fund ownership between the smallest firms in the Russell 1000 index and the largest firms in the Russell 2000 index, even though these firms are of similar market value. In this study, we rely on the exogenous change in ownership around the Russell 1000/2000 cutoff to assess whether passive funds affect audit quality.

Specifically, we use the Russell 2000 index membership to instrument for passive ownership. Our instrumental variable (IV) estimation relies on the assumption that, for firms close to the cutoff, after conditioning on stocks’ market capitalization, inclusion in the Russell 2000 index does not directly affect audit quality, except through its impact on passive ownership. Similar to Appel et al. (2016), we find that firms at the top of the Russell 2000 index have substantially higher passive ownership than those at the bottom of the Russell 1000 index.

We use audit fees as the primary measure of audit quality, and we use index fund ownership to proxy for passive ownership (Appel et al., 2016). Our IV estimation includes a battery of standard control variables used in the audit fee literature, firms’ end-of-May market capitalization (up to the third polynomial order), the float-adjusted market cap (used by Russell to determine index weights), and year fixed effects. Our results – based on samples of firms within the 100, 150, and 200 bandwidths around the cutoff – consistently show that passive ownership leads to higher audit fees. Specifically, a one percentage point increase in passive ownership leads to approximately 18% higher audit fees. This suggests that passive investors play an active role in audit-related governance issues.

We next examine the channels/mechanisms through which passive investors exert their influence on audit-related governance issues. Some anecdotal evidence suggests that passive investors actively engage with firm management to improve corporate governance,² but archival evidence on this notion is rare due to data limitations.³ Since we cannot observe behind-the-scenes engagements between investors and their portfolio companies, we use proxy voting – an important channel through which institutions express their (dis)satisfaction with the company (Garcia Osma & Grande-Herrera, 2021) – to examine whether passive investors pay attention to audit-related governance issues. Our IV estimation shows that passive ownership is significantly and negatively associated with the percentage of shareholders who vote with management on auditor ratification. Further, we find that passive ownership leads to a subsequently higher likelihood of auditor turnover. Since the ‘control over hiring and firing the auditor serves as the core incentive for auditors to maintain or compromise their independence’ (Mayhew & Pike, 2004, p. 798), our results provide strong evidence that passive investors play an active and effective role in improving audit-related governance issues through the ‘voice’ channel.

As ample auditing studies have shown, both supply- and demand-side forces can drive changes in audit outcomes (e.g., DeFond & Zhang, 2014; Hay et al., 2006). Our results on auditor ratification votes and subsequent turnover suggest that the demand-side pressure from passive investors is at least an important driver in the audit fee change around the Russell 1000/2000 cutoff.

In the cross-sectional analysis, we examine whether the impact of passive institutional ownership on audit quality varies across firms. The demand for higher auditing quality originates from agency costs. Therefore, shareholders’ monitoring efforts should vary between firms with high and low agency costs. To examine whether the impact of passive ownership on audit quality varies across firms, we split the sample by (1) auditor independence, (2) information complexity of the firms, (3) pressure from short-term investors, (4) CEO incentive to conduct earnings management, and (5) board independence. We find that the impact of passive investors on audit quality is more pronounced for firms with higher agency costs. This result further strengthens our argument that the observed changes in audit quality around the Russell 1000/2000 cutoff are at least partly due to demand-side drivers imposed by passive investors.

Finally, we conduct three sets of robustness checks. First, we use alternative measures of audit quality to provide a more comprehensive picture of how passive investors affect audit quality. We find that passive ownership is negatively associated with the likelihood of reporting small profits and the subsequent restatements of the current financial statements. We also find some evidence that passive ownership is positively associated with the number of reported internal control weaknesses, but the result is not statistically significant at the conventional level of 5%. Combined, our study using both input (audit fees) and output (restatements and small profits) measures of audit quality shows that passive investors have a positive impact on audit quality.

Second, we re-examine our main results using three alternative definitions of passive ownership. Our results provide consistent and strong evidence that higher passive ownership leads to higher audit fees. Third, we address some recent concerns regarding various empirical methods applied to the Russell index setting (Appel et al., 2020; Wei & Young, 2019). More specifically, we rank stocks according to their end-of-May market cap, calculated from the Center for Research in Security Prices (CRSP) data. Based on this ranking, we select our sample firms within the 100, 150, and 200 bandwidths. Our results, based on this sample, are also largely in line with our main results.

This study contributes to the literature in at least two ways. First, we provide new insights into the literature that investigates the drivers of client demand for audit quality. Since audit quality, corporate governance system, and agency cost proxies are all choice variables, it is difficult to establish causal inference between these variables (DeFond & Zhang, 2014). This study contributes to the literature by addressing the endogeneity problem in this line of research. Unlike previous studies, our work looks at a relatively new type of owners, namely passive investors, who play an increasingly important role in the capital market.

Second, this study adds new evidence to the ongoing debate over the activism of passive institutional investors. We show that passive ownership leads to higher audit quality, and that the positive effect is not merely transmitted from improvements in other governance aspects. Rather, we find direct evidence that passive owners pay attention specifically to audit-related governance issues. Although we do not exclude the possibility that passive owners may underinvest in some aspects of (especially high-cost) stewardship, our study highlights that the positive impact of passive owners should not be ignored. The implications of this study are, therefore, important for investors and regulators.

Related Literature and Hypotheses Development

The Debate Over Passive Investors’ Impact on Corporate Governance

The recent debate over the consequences of passive ownership on corporate governance centers on whether passive investors have the incentive to exert any effort to improve governance. Many commentators⁴ and scholars have expressed negative views in this regard. For passive owners to influence firm governance, the ‘exit’ option is not available because the (ex-)/inclusion of a firm is determined by the index composition. However, according to Schmidt and Fahlenbrach (2017), the ‘voice’ channel is not feasible either, because it seems too expensive for low-cost, low-overhead index funds. Bebchuk and Hirst (2019) and Heath et al. (2022) share this view; they show that passive investors underinvest in stewardship, are more likely to vote in a pro-management manner, and are less likely to monitor governance.⁵

On the positive side, Carleton et al. (1998) have shown early evidence (through a case study) that (1) large institutional investors who follow a passive investment style have to adopt activist positions due to their inability to sell, and (2) their concentrated ownership facilitates the success of behind-the-scenes engagements initiated by such investors.⁶ This is consistent with the recent empirical study by Appel et al. (2016), who find that passive ownership positively affects various aspects of corporate governance. Passive institutional investors may exert their influence through active voting (Iliev & Lowry, 2015) and supporting activist campaigns (Appel et al., 2019; Jahnke, 2019). Another potential incentive for index funds’ activism is that engagements with certain firms often have important spillover effects (e.g., through publicity) on other firms that proactively improve their corporate governance and avert conflict and public scrutiny (Del Guercio & Hawkins, 1999; Jahnke, 2019).⁷

Combined, the extant literature on the consequences of increased passive ownership is inconclusive. Thus, we aim to provide further evidence from the financial statement audit perspective. Unlike related studies – such as Bird and Karolyi (2016), which use the Russell setting to examine the impact of total institutional ownership on disclosure quality–our study focuses on index funds that are central to the current debate on the efficacy of passive investors’ stewardship. More importantly, we conduct our investigation through the lens of financial statement auditing, a key mechanism of corporate governance that has been neglected in the emerging literature that questions the incentives and behavior of index funds.

Corporate Governance and the Quality of Financial Statement Audits

Agency theory predicts that equity holders, especially outsiders, demand information about a firm because of conflicts of interest and information asymmetry (Jensen & Meckling, 1976). Financial reports that reflect a firm’s fundamentals are vital sources of information for shareholders. Since managers’ interests may deviate from shareholders’ interests, high-quality audits have become an important element in the corporate governance sphere, and shareholders have the incentive to employ high-quality auditors to help control agency conflicts and signal high-quality financial reporting (DeAngelo, 1981; Han et al., 2012; Kausar et al., 2016; Lennox & Pittman, 2011).

Several studies show that ownership structure affects the demand for audit services.⁸ Institutional investors are powerful and sophisticated participants in the stock market (Chakravarty, 2001; Walther, 1997). Previous research provides empirical evidence on institutional owners’ monitoring behavior and its effect on corporate governance, firm performance, earnings quality, and similar topics. In terms of auditing, Kane and Velury (2004), Velury et al. (2003), Mitra et al. (2007), and Lim et al. (2013) provide evidence that institutional ownership is positively associated with higher audit quality, which is proxied by audit firm size, choice of industry-specialist auditors, audit fees, and so on.

However, the current debate over passive owners’ incentives to monitor casts doubt on whether the aforementioned findings would be directly applicable to passive institutional investors’ impact on auditing-related issues. Furthermore, even if there are observable effects, we still lack an understanding of the underlying mechanisms. In other words, how do passive institutional owners influence companies’ auditors?

Audit outcomes are driven by the supply- and demand-side forces, and passive owners may affect both. On the supply side, auditors may choose to supply additional audit effort based on a perception of increased risk due to heightened scrutiny by passive shareholders or the activism campaigns they facilitate (Appel et al., 2019; Guo et al., 2021). On the demand side, passive investors may directly raise specific audit-related issues or indirectly improve audit quality through changes that enhance corporate governance in general (such as board independence and equal voting rights). Empirical evidence on institutional investors’ demand-side (both direct and indirect) impact on audit quality has been shown in studies such as Chan et al. (2021), Beasley and Petroni (2001), and Niu (2008), but there has been no research on the effect of passive investors per se.

Since we cannot directly observe audit input (such as hours)⁹ and how passive investors interact with firms behind the scenes, it is empirically challenging to rule out either the supply-side or demand-side explanation. Rather, we believe that both the demand- and supply-side factors are most likely operating in concert. In this study, we examine proxy voting, a demand-side mechanism through which passive owners may affect auditor choice, and leave other potential channels for future research.

Taken together, given the ongoing discussion about whether passive investors deliver their stewardship promise, we aim to provide further empirical evidence from the audit quality perspective and examine the mechanism through which passive investors exert their influence on financial statement audits. We formulate the following null hypothesis:

H1: Passive institutional investors have no impact on firms’ audit quality.

Identification Strategy and Research Design

The key challenge in establishing causal inference on the relationship between passive ownership and audit quality is identifying an exogenous instrument. Firms that hire high-quality auditors, present true and fair financial reports, and adopt sound governance policies may attract institutional owners. Omitted variables may also confound the results of an ordinary least squares (OLS) regression (DeFond & Zhang, 2014; Hay et al., 2006). Hence, we follow Appel et al. (2016, 2020) and use the Russell index assignment as an exogenous shock to passive mutual fund ownership. In this section, we describe our identification strategy and data in detail.

Russell 1000/2000 Index Construction

Our identification strategy relies on a special mechanism that Russell Investments uses to assign stocks to the Russell 1000 and 2000 indexes. Russell Investments uses the last traded price on the final trading day of May each year to calculate each stock’s total market capitalization. The largest 1,000 stocks comprise the Russell 1000 index, and the next 2,000 stocks comprise the Russell 2000 index. The final reconstitution date is the last Friday of June. After this date, firms remain within their allocated index for a year unless certain events occur, such as mergers, acquisitions, or delisting. The cutoff line between the Russell 1000 and 2000 indexes depends on the 1,000th largest stock size, which varies annually. Every year, the 1,000th-largest stock on the U.S. market is classified in the Russell 1000 index, whereas the 1,001st-largest stock falls into the Russell 2000 index. Thus, the fundamental characteristics of firms around the cutoff point are continuous, and the assignment around the threshold is somewhat random. Similar to Appel et al. (2016),¹⁰ we restrict our sample to a bandwidth close to this threshold, which resembles a randomized sample, in line with the regression discontinuity research design.¹¹

We obtain the Russell 1000 and 2000 index member firms from Russell Investments for the period from 2000 to 2006. The sample starts from the year 2000, because this was when Audit Analytics started to have relatively good coverage of U.S. firms. Our sample ends in 2006 because Russell Investments introduced a banding policy that does not strictly use market capitalization to reconstitute the indexes; this change invalidates our identification strategy without further assumptions. We exclude firms in the finance industry and utility companies because of their special business model and the model’s reflection on the firm’s accounting.

Russell Investments also assigns a portfolio weight to each stock once the index membership list is reconstituted and uses this portfolio weight to rank all stocks within an index. The weight is determined by the stock’s float-adjusted market capitalization, which differs from the market capitalization used to determine index assignment. Float-adjusted market capitalization excludes the value of low investible shares (e.g., shares owned by blockholders, employees’ stock ownership plans, other Russell index members, or the government). As described by Boone and White (2015), there is a large discontinuity in the portfolio weight for stocks around the Russell 1000/2000 cutoff line. They report that the top 20 Russell 2000 firms receive portfolio weight, on average, 46 times greater than stocks at the bottom of the Russell 1000.

Passive Mutual Fund Ownership

In this section, we explain why the index-weighting mechanism described above is important for index fund investment. Index funds are mutual funds that track the performance of a specific index. In the attempt to replicate the target index, index funds invest all or substantially all their assets in the stocks that make up the index. For the same purpose, each stock in the fund portfolio is value-weighted according to its index weight. In contrast to active fund managers selecting stocks into their portfolio, index funds passively invest in (nearly) all stocks in a market index that they use as benchmarks. Since index funds are the most visible type of passive funds (Appel et al., 2016), we follow Appel et al. (2016) and use index fund ownership to proxy for passive institutional ownership in our main test. We note, however, that not all passive investors are index funds.¹² Passive investing could also be in the form of allocating a stock portfolio across various index funds.¹³ Moreover, as Appel et al. (2016) pointed out, using index funds as a proxy for passive ownership cannot capture the total number of passive ownership stakes because ‘passive investment[s] by some institutions, like pension funds, are not reported in the S12 mutual fund database’ (p. 118). Nevertheless, since index funds follow the most visible passive investment style and there is no superior way to measure passive ownership stake, we follow the literature and resort to the S12 mutual fund file.

Because of the value-weighting mechanism implemented by Russell Investments, stocks ranked at the bottom of the Russell 1000 index receive significantly less attention from index funds than stocks at the top of the Russell 2000 index. The 1,000th-largest stock in its May capitalization is likely ranked at the bottom of the Russell 1000 index with small index weights and, hence, is of minor importance to a fund benchmarked to the Russell 1000 index. In stark contrast, the 1,001st-largest stock will most likely have a top ranking in the Russell 2000 index, attracting much more index fund attention owing to its heavy index weight.¹⁴ Hence, it is obvious that index assignment has a significant impact on passive ownership for firms around the Russell 1000/2000 threshold. Russell Investments uses a proprietary method of adjusting stocks placed in their 1000 and 2000 indexes after classifying stocks to the former or latter. This proprietary method is not observable by researchers, and we follow Appel et al.’s (2016) identification strategy, as detailed in the following subsection.

Identification Strategy

First-stage estimation

The Russell 1000/2000 index assignment mechanism provides an exogenous instrument for examining the relationship between passive ownership and audit quality. Stocks close to the threshold resemble a random sample, in which those included in the Russell 2000 index receive significantly heavier ‘treatment’ from passive investors. Therefore, we restrict our sample to a close band around the index cutoff line and use inclusion in the Russell 2000 index to instrument for index fund ownership. In doing so, we attempt to exclude other endogenous factors that could correlate with both index fund ownership and audit quality. This is equivalent to the IV setting. Specifically, we estimate the following (first-stage) equation and use the estimated value of $\widehat{\mathrm{PMF}{O}_{\mathrm{it}}}$ from Equation (1): (1) ${\mathit{\text{PMFO}}}_{\mathrm{it}}=a+\mathrm{bR}{2000}_{\mathrm{it}}+\sum _{n=1}^N{\tau }_n(\ln (\mathrm{MktCa}{p}_{\mathrm{it}}){)}^n+\mathrm{\phi ln}(\mathrm{Floa}{t}_{\mathrm{it}})+\varphi X_{\mathrm{it}}+{\rho }_t+{ϵ}_{\mathrm{it}}$(1) where $\mathrm{PMF}{O}_{\mathrm{it}}$ is passive mutual fund (index fund) ownership and $R{2000}_{\mathrm{it}}$ is an indicator variable equal to one when a sample firm is assigned to the Russell 2000 index and zero if the firm is a Russell 1000 firm in year $t$. We control for (1) the first-, second-, and third-order polynomials of the natural logarithm of the end-of-May market capitalization ($\ln (\mathrm{MktCa}{p}_{\mathrm{it}})$, based on the CRSP data), and (2) end-of-June float-adjusted market capitalization ($\mathrm{Floa}{t}_{\mathrm{it}}$, provided by Russell). This is because, within a certain index, the impact of passive ownership on audit quality may be endogenous: end-of-May market capitalization and float-adjusted market value of firms after reclassification by Russell could correlate with both passive ownership and audit quality; that is, these are likely determinants of the assignment to the treatment group. However, after conditioning on these two factors, the exclusion restriction seems quite reasonable: being included at the top of the Russell 2000 significantly increases passive ownership, but there is no obvious reason why index inclusion could significantly affect audit quality, given that our sample is restricted to a tight band around the index threshold after controlling for the functional forms on either side of the threshold.¹⁵ Finally, $X_{\mathrm{it}}$ is a set of control variables for firm $i$ in year $t$. We apply the same set of control variables in our first-stage estimation as in the second-stage estimation described below, and our tests for each outcome variable include different sets of control variables. ${\rho }_t$ are year indicators.

Second-stage estimation

Once we have isolated the variation in index fund ownership stemming from the index assignment, we gauge the impact of passive institutional investment on audit fees and other audit outcomes. Specifically, we estimate the causal effect of index fund ownership on audit quality from the following (second-stage) equation: (2) $Y_{\mathrm{it}}=\alpha +\beta {\widehat{\mathrm{PMFO}\mathrm{\%}}}_{\mathrm{it}}+\sum _{n=1}^N{\tau }_n(\ln (\mathrm{MktCa}{p}_{\mathrm{it}}){)}^n+\mathrm{\phi ln}(\mathrm{Floa}{t}_{\mathrm{it}})+\varphi X_{\mathrm{it}}+{\rho }_t+{ϵ}_{\mathrm{it}}$(2) where $Y_{\mathrm{it}}$ is the audit quality measure for firm $i$ in year $t$ unless otherwise specified. ${\widehat{\mathrm{PMFO}}}_{\mathrm{it}}$ is the predicted passive mutual fund (index fund) ownership. Our main proxy for audit quality $Y_{\mathrm{it}}$ is the natural log of audit fees (LN_AF). We posit that if passive investors play an active role in audit-related governance issues, their demand for higher audit quality may increase audit fees (Cassell et al., 2012; Eilifsen et al., 2001; Niemi, 2005; O’Sullivan, 2000). According to DeFond and Zhang (2014), audit fees is one of the main measures of audit quality from the input perspective.

Audit risk can also be a source of higher audit fees. However, our sample firms’ market value is continuous around the index cutoff, where the Russell index assignment is close to random. Therefore, we argue that there should not be a significant shift in audit risk among our sample firms across the index threshold. Nevertheless, we employ a vector of audit-related control variables ($X_{\mathrm{it}}$) to consider possible confounding factors. These variables include the level of transient institutional ownership (TRA), natural log of end-of-year total assets (LN_TA), natural log of end-of-year revenue (LN_REV), number of geographic segments (NGS), number of business segments (NBS), leverage (LEV), inventory scaled by total assets (INVT_SCALED), receivables scaled by total assets (RECEIVABLES_SCALED), return on assets (ROA), an indicator variable equal to one if the firm reports special items and zero otherwise (SPECIAL_ITEM), an indicator variable equal to one if the firm reports a negative net income and zero otherwise (LOSS_NI), and the number of analysts following a given firm (LN(1 + NUM_ANALYST)).

In essence, our model is an extended audit fee model based on Simunic’s (1980) framework, plus a control for transient institutional investors (Bushee, 2001) and the number of analysts following a firm.¹⁶ We add index fund ownership as our key variable to the equation and take care of the endogeneity problem using an IV approach, as described above. We include year-fixed effects to take care of the time trend for all estimations.¹⁷ We control for the value of transient ownership measured at the end of September in all specifications.

We also investigate the potential mechanism by which passive institutional investors can affect audit quality. We use the percentages of support (‘for’ minus ‘against’ shareholder votes) for auditor ratification proposals (AUD_RATIF_SUPPORT) to inspect whether passive institutional ownership impacts these proposals’ voting outcomes. Furthermore, we examine whether passive ownership affects the likelihood of auditor dismissal in the following year (AUDITOR_TURNOVER). We conduct cross-sectional tests in the subsection ‘Cross-sectional analysis based on agency cost’ to examine whether the effect of passive ownership varies among firms with different levels of agency costs. We also check the robustness of our results using alternative measures of audit quality proxied by (1) meeting or beating the zero earnings threshold (SMALL_PROFITS), (2) restatements (RESTATEMENT_AO), and (3) internal control weaknesses (LN(1 + NUM_IC_W)). The subsection ‘Robustness check’ describes these in detail.

Data Sources

Following Appel et al. (2016), we use passive mutual fund (index fund) holdings data to measure passive institutional investor ownership for the main tests. Mutual fund holdings data come from the Thomson Reuters S12 file. We use the complementary method by Iliev and Lowry (2015) and Appel et al. (2016) to extract passively managed funds. Hence, a fund is identified as a passively managed index fund if the fund name includes one of the following strings: ‘Index, Idx, Indx, Ind_ (where _ indicates a space), Russell, S & P, S and P, S&P, SandP, SP, DOW, Dow, DJ, MSCI, Bloomberg, KBW, NASDAQ, NYSE, STOXX, FTSE, Wilshire, Morningstar, 100, 400, 500, 600, 900, 1000, 1500, 2000, and 5000.’ In this paper, we denote index ownership calculated by this approach as PMF_A%.

We use funds’ end-of-September holding as a proxy for passive investor ownership stake, because the Russell reconstitution happens in June. Before May 2004, funds were only required to report twice a year regarding their holdings’ information, although some voluntarily filed quarterly reports to the U.S. Securities and Exchange Commission (SEC). Therefore, we populate missing September holdings by taking the average of the reported holdings before and after this missing date if the gap is within one year. If the gap is over one year, the Thomson Reuters S12 file description states that it is very likely that the fund does not hold any shares of the firm, or the fund has undergone significant changes. Our method arrives at a passive mutual fund ownership percentage level similar to Appel et al. (2016).

Our data for audit fees, restatements, internal control weaknesses, and auditor turnover come from the Audit Analytics database. We obtain data on auditor ratification voting from the Institutional Shareholder Services (ISS) database. Additionally, accounting numbers come from the Compustat Annual file, and end-of-May market capitalization information comes from the CRSP file. Finally, we obtain data on the number of analysts following firms from the Institutional Brokers’ Estimate System (I/B/E/S) database. We follow Bergstresser and Philippon (2006) in constructing our measure of the CEO’s incentive ratio using data from Execucomp in our cross-sectional analysis.¹⁸ In the online appendix, OA2, we present the details of our data source for all the variables used in this study.

We initially identify, on average, around 2,997 firms’ PERMNO each year in the Russell 3000 (Russell 1000 and 2000 combined) firm list for the period from 2000 to 2006, as shown in Table 1, column (2). We then merge the Russell index list with CRSP, S12, 13F, COMPUSTAT, and Audit Analytics (A.A.). Finally, we require each firm-year observation to have non-missing values for all necessary test variables, and delete firms in the financial and utility industries, thereby arriving at around 1,531 firms each year on average (see column (4) of Table 1) from the Russell 3000 firm list. The lower number in the year 2000 is due to the relatively poor coverage of firms by Audit Analytics in that year.

Table 1. Sample description.

	Initial Russell 3000 firm list	Identifiable with PERMNO	After merging with other databases	Observations with non-missing data	Final main sample	Final main sample (R1000)	Final main sample (R2000)
Year	(1)	(2)	(3)	(4)	(5)	(6)	(7)
2000	3,000	2,999	2,733	914	55	24	31
2001	3,000	3,000	2,788	1,526	70	35	35
2002	3,000	3,000	2,862	1,650	84	40	44
2003	2,999	2,999	2,861	1,691	88	45	43
2004	2,998	2,998	2,852	1,683	83	43	40
2005	3,000	2,999	2,831	1,631	82	40	42
2006	2,987	2,987	2,807	1,622	96	48	48
Total	20,984	20,982	19,734	10,717	558	275	283

Note: This table shows the process used to generate the final sample. Column (1) shows the number of firms initially included in the Russell 3000 list of firms during the period from 2000 to 2006. We obtain this information from Russell Investments. From this initial sample, we identify 20,982 firm-year observations by PERMNO (shown in column (2)). Subsequently, we merge the list of firms with the S12, 13F, CRSP, COMPUSTAT, and Audit Analytics database and arrive at column (3). We further delete firms in the finance and utility industry (SIC 5999–7000 and SIC 4899–5000) and firms that have missing values for variables in the main (audit fee) model, which reduces the sample to 10,717 observations (shown in column (4)). Further, we restrict our sample to ±100 bandwidth around the Russell 1000/2000 cutoff with the available data to run our main regression for audit fees; thus, the final sample is as shown in column (5). Finally, we split our final sample into firms at the bottom 100 places of the Russell 1000 index (column (6)) and those at the top 100 places of the Russell 2000 index (column (7)).

We restrict our main test sample to the bottom 100 firms in the original Russell 1000 list and the top 100 firms in the original Russell 2000 index member list, as determined by the portfolio weights assigned by Russell for each index. We also present the results obtained using bandwidths of ±150 and ±200. For the audit fee tests, we are left with 558 firm-year observations based on the ±100 bandwidth (column (5) of Table 1). Nevertheless, for some of the other tests, we are restricted to a smaller sample; for instance, the internal control weakness data are only available from 2004. Finally, we winsorize all continuous variables at 1% and 99% by year, although the results (not tabulated) are consistent with our main results when we do not winsorize our variables.

Descriptive Statistics and Empirical Results

Summary Statistics

Table 2 reports the summary statistics for the main test variables partitioned into two panels. Panel A shows the descriptive statistics for the bottom 100 firms in the Russell 1000 index, whereas Panel B shows descriptive statistics for the top 100 firms in the Russell 2000 index. Our main sample has 558 firm-year observations in our audit fee test sample (our primary test), which includes firms in the ±100 bandwidth across the Russell 1000/2000 cutoff line for the period from 2000 to 2006. Of the 558 observations, 275 are from the Russell 1000 index and 283 from the Russell 2000 index. For brevity, we present the descriptive statistics for the control variables in the online appendix, Table A1.

Table 2. Descriptive statistics.

Variable	Obs.	Mean	S.D.	P25	P50	P75
Panel A: R1000
Passive ownership measures
PMF_A%	275	0.019	0.014	0.007	0.015	0.028
PMF_P%	275	0.015	0.009	0.008	0.013	0.019
PMF_Big3%	275	0.046	0.029	0.024	0.04	0.062
QIX	275	0.346	0.176	0.205	0.337	0.467
Audit quality measures
LN_AF	275	13.628	1.103	12.799	13.631	14.318
AUDITOR_TURNOVER	275	0.073	0.26	0	0	0
AUD_RATIFICATION_SUPPORT	117	0.983	0.024	0.978	0.992	0.998
LN(1 + NUM_IC_W)	109	0.106	0.342	0	0	0
SMALL_PROFITS	275	0.015	0.12	0	0	0
RESTATEMENT_AO	275	0.225	0.419	0	0	0
Panel B: R2000
Passive ownership measures
PMF_A%	283	0.037	0.018	0.027	0.038	0.047
PMF_P%	283	0.024	0.009	0.018	0.024	0.029
PMF_Big3%	283	0.078	0.027	0.061	0.077	0.091
QIX	283	0.504	0.129	0.419	0.5	0.583
Audit quality measures
LN_AF	283	13.562	1.053	12.854	13.549	14.23
AUDITOR_TURNOVER	283	0.053	0.224	0	0	0
AUD_RATIFICATION_SUPPORT	114	0.966	0.055	0.961	0.984	0.992
LN(1 + NUM_IC_W)	113	0.103	0.42	0	0	0
SMALL_PROFITS	283	0.007	0.084	0	0	0
RESTATEMENT_AO	283	0.205	0.404	0	0	0

Note: This table presents the summary statistics of the main test variables. Panel A shows the bottom 100 firms in the Russell 1000 index, whereas Panel B shows the top 100 firms in the Russell 2000 index. Variable definitions are presented in the Appendix. The sample sizes are slightly different for different tests because of the data availability for calculating audit quality proxies and control variables. Hence, in this table, we report most of the variables based on the audit fee test sample, which includes firms with no missing data within the ±100 bandwidth around the Russell 1000/2000 cutoff line for the period from 2000 to 2006. For each variable, we present the mean, median (p. 50), standard deviation, first quartile (p. 25), third quartile (p. 75), and number of observations. For brevity, we present the descriptive statistics for the control variables in the online appendix, Table A1.

For the pooled sample, the median value of index fund ownership (denoted as PMF_A%) is 2.8 percentage points in our sample (not tabulated for brevity), which is very close to the value of 2.6 percentage points reported in Appel et al. (2016).¹⁹ The descriptive statistics in Table 2 and Table A1 also show the difference between the Russell 1000 and 2000 firms. The Russell 1000 index comprises larger firms in terms of market value, total assets (LN_TA), and return on assets (ROA). Consistent with the larger company size, the Russell 1000 firms also pay higher audit fees.²⁰ Despite the larger size, the Russell 1000 firms exhibit lower index fund ownership (1.9 percentage points on average for PMF_A%, for example) than the Russell 2000 firms (3.7 percentage points on average for PMF_A%), in line with our expectation. Our observations also show that larger firms in the Russell 1000 index exhibit higher auditor turnover, larger percentages of auditor ratification support from shareholders, a higher likelihood of restatements and meeting or beating the zero earnings threshold, and more internal control weaknesses.

Main Results

First-stage regression

To demonstrate that index assignment has important implications for index fund ownership among firms around the cutoff line, we run the first-stage regression, as in Equation (1). The specification is estimated for the period from 2000 to 2006 using our main measure of passive institutional ownership. Table 3 shows the test results. We start with our main sample using the ±100 bandwidth and a polynomial of orders one, two, and three in columns (1), (2), and (3), respectively. The estimated coefficient on the indicator for the Russell 2000 firms (R2000) is statistically significant at the 1% level and is not sensitive to different polynomial orders and bandwidth selection. Overall, the explanatory power of our first-stage regression is between 49% and 51%, depending on the specification. Columns (4) to (9) demonstrate that the estimated relationship between the Russell 2000 index inclusion and passive ownership is robust to different bandwidth choices.²¹ Overall, our first-stage result is consistent with that of Appel et al. (2016), showing that index assignment significantly impacts passive ownership.

Table 3. First-stage regression – how index assignment affects index fund ownership.

Dependent variable:	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
PMF_A%
R2000	0.012***	0.012***	0.012***	0.010***	0.010***	0.010***	0.009***	0.009***	0.009***
	(6.455)	(6.581)	(6.288)	(7.314)	(7.325)	(6.715)	(8.282)	(8.272)	(7.388)
Observations	558	558	558	876	876	876	1,184	1,184	1,184
Polynomial order	1	2	3	1	2	3	1	2	3
Bandwidth	100	100	100	150	150	150	200	200	200
Float-adj M. Cap.?	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Controls?	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Fixed effects?	Year	Year	Year	Year	Year	Year	Year	Year	Year
Adj. R^2	51%	51%	51%	50%	50%	49%	49%	49%	49%

Note: This table shows the results for the first-stage estimation based on Equation (1). Specifically, we estimate. $\mathrm{PMF}\mathrm{\_}{A}_{\mathrm{it}}=a+\mathrm{bR}{2000}_{\mathrm{it}}+\sum _{n=1}^N{\tau }_n(\ln (\mathrm{MktCa}{p}_{\mathrm{it}}){)}^n+\mathrm{\phi ln}(\mathrm{Floa}{t}_{\mathrm{it}})+\varphi X_{\mathrm{it}}+{\rho }_t+{ϵ}_{\mathrm{it}},$

where $\mathrm{PMF}\mathrm{\_}{A}_{\mathrm{it}}$ is firm $i$’s end-of-September index fund ownership in year t based on Appel et al.’s (2016) approach. The indicator variable $R{2000}_{\mathrm{it}}$ is equal to one if firm $i$ is assigned to the Russell 2000 index in year t, and zero if the firm is assigned to the Russell 1000. We control for firms’ end-of-June float-adjusted market capitalization provided by Russell. We show the test results for the ±100 (columns (1), (2), and (3)), ± 150 (columns (4), (5), and (6)), and ±200 bandwidths (columns (7), (8), and (9)). For brevity, we suppress the regression coefficients on the control variables and constant, and present the results using the first-, second-, and third-order polynomials. Variable definitions are presented in the Appendix. T-statistics are in parentheses. Standard errors are robust and clustered at the firm level to account for serial correlation. ***, **, and * represent significance levels at the 1%, 5%, and 10% levels, respectively.

Passive ownership and audit fees

As discussed in our empirical framework, our main measure of audit quality is audit fees (LN_AF). We present our main results only using PMF_A% as a proxy for passive ownership.²² In Table 4, we report the results from our IV approach estimating Equations (1) and (2) using 2SLS. We start with a ± 100 bandwidth around the threshold in columns (1) to (3), with varying polynomial orders of controls for end-of-May market capitalization. Next, in columns (4) to (6) and columns (7) to (9), we present the results based on the ±150 and ±200 bandwidths, respectively. The sample firms are arguably more comparable when we use the ± 100 bandwidth. Thus, we consider the functional form that uses a first-order polynomial more reliable (Atanasov & Black, 2016; Roberts & Whited, 2013), and focus on interpreting the results in column (1). Column (1) shows that a one percentage point increase in index fund ownership increases audit fees by around 18.1%, which is statistically significant at the 5% significance level. The results in columns (2) to (9) are consistent with those in column (1).

Table 4. Index fund ownership and audit fees.

Dependent Variable:	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Audit fee (LN_AF)
PMF_A%	18.144**	19.255**	21.669**	12.319*	13.387**	16.840**	9.385	10.045*	13.660**
	(2.135)	(2.333)	(2.538)	(1.837)	(2.090)	(2.431)	(1.574)	(1.748)	(2.137)
TRA	0.799**	0.847***	0.865***	0.972***	1.010***	1.026***	0.864***	0.888***	0.903***
	(2.560)	(2.729)	(2.745)	(3.948)	(4.123)	(4.114)	(4.075)	(4.209)	(4.227)
NBS	0.017*	0.016*	0.015	0.011	0.010	0.009	0.009	0.008	0.007
	(1.816)	(1.734)	(1.630)	(1.562)	(1.480)	(1.295)	(1.322)	(1.276)	(1.089)
NGS	0.036***	0.036***	0.036***	0.035***	0.035***	0.035***	0.034***	0.034***	0.034***
	(5.654)	(5.811)	(5.668)	(6.797)	(6.944)	(6.800)	(6.967)	(7.062)	(7.044)
ROA	−0.006	−0.003	0.001	−0.003	0.000	0.004	−0.017	−0.016	−0.014
	(−0.204)	(−0.112)	(0.037)	(−0.087)	(0.013)	(0.114)	(−0.474)	(−0.445)	(−0.350)
LN_TA	0.497***	0.505***	0.511***	0.476***	0.482***	0.491***	0.435***	0.440***	0.448***
	(7.497)	(7.763)	(7.627)	(8.854)	(9.158)	(9.060)	(8.542)	(8.668)	(8.578)
LN_REV	0.079**	0.080**	0.075**	0.103***	0.102***	0.093**	0.123***	0.121***	0.110***
	(2.196)	(2.228)	(2.077)	(2.761)	(2.801)	(2.506)	(3.510)	(3.503)	(3.109)
LEV	0.113	0.091	0.101	0.069	0.052	0.055	0.183	0.176	0.181
	(0.575)	(0.456)	(0.497)	(0.456)	(0.346)	(0.359)	(1.375)	(1.308)	(1.318)
INVT_SCALED	−0.175	−0.193	−0.203	−0.203	−0.223	−0.236	−0.322	−0.332*	−0.331
	(−0.700)	(−0.771)	(−0.797)	(−0.952)	(−1.046)	(−1.085)	(−1.623)	(−1.671)	(−1.630)
RECEIVABLES_SCALED	1.666***	1.686***	1.670***	1.219***	1.227***	1.233***	1.076***	1.091***	1.101***
	(3.773)	(3.783)	(3.677)	(3.694)	(3.692)	(3.631)	(3.763)	(3.793)	(3.752)
SPECIAL_ITEM	0.163**	0.167**	0.167**	0.203***	0.207***	0.207***	0.230***	0.231***	0.232***
	(2.521)	(2.565)	(2.527)	(4.088)	(4.158)	(4.085)	(5.177)	(5.199)	(5.148)
LOSS_NI	0.193**	0.201**	0.190**	0.205***	0.211***	0.203***	0.221***	0.222***	0.213***
	(2.288)	(2.372)	(2.188)	(2.918)	(2.993)	(2.799)	(3.442)	(3.461)	(3.257)
LN(1 + NUM_ANALYST)	−0.058	−0.060	−0.057	−0.069	−0.073	−0.070	−0.084*	−0.087*	−0.083
	(−0.860)	(−0.873)	(−0.815)	(−1.247)	(−1.294)	(−1.222)	(−1.669)	(−1.710)	(−1.611)
Observations	558	558	558	876	876	876	1,184	1,184	1,184
Polynomial Order	1	2	3	1	2	3	1	2	3
Bandwidth	100	100	100	150	150	150	200	200	200
Float-adj M. Cap.?	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Fixed Effects?	Year	Year	Year	Year	Year	Year	Year	Year	Year
Adj. R^2	43%	42%	39%	48%	48%	45%	49%	48%	46%
First Stage F	41.66	43.32	39.53	53.50	53.65	45.09	68.59	68.43	54.58

Note: This table presents the IV estimation results to identify the effect of index fund ownership on audit fees based on Equation (2): $\mathrm{LN}\mathrm{\_A}{F}_{\mathrm{it}}=\alpha +\beta {\widehat{\text{PMF}\mathrm{\_}\text{A\% }}}_{\mathrm{it}}+\sum _{n=1}^N{\tau }_n(\text{ln}(\mathrm{MktCa}{p}_{\mathrm{it}}){)}^n+\mathrm{\phi ln}(\mathrm{Floa}{t}_{\mathrm{it}})+\varphi X_{\mathrm{it}}+{\rho }_t+{ϵ}_{\mathrm{it}},$

where variable LN_AF is firm $i$’s audit fee in year t and $\mathrm{PMF}\mathrm{\_}{A}_{\mathrm{it}}$ is firm $i$’s end-of-September index fund ownership based on Appel et al.’s (2016) approach in year t predicted from the first-stage regression. We control for firms’ end-of-June float-adjusted market capitalization. We show the results using the ±100 (columns (1), (2), and (3)), ± 150 (columns (4), (5), and (6)), and ±200 bandwidths (columns (7), (8), and (9)). Depending on the specification, our estimations use the first-order polynomial of the natural logarithm of end-of-May market capitalization from CRSP (columns (1), (4), and (7)), second-order polynomial (columns (2), (5), and (8)), and third-order polynomial (columns (3), (6), and (9)). The sample period is from 2000 to 2006. The control variables include TRA, NBS, NGS, ROA, LN_TA, LN_REV, LEV, INVT_SCALED, RECEIVABLES_SCALED, SPECIAL_ITEM, LOSS_NI, and LN(1 + NUM_ANALYST). Variable definitions are presented in the Appendix. T-statistics are in parentheses. Standard errors are robust and clustered at the firm level to account for serial correlation. ***, **, and * represent significance levels at the 1%, 5%, and 10% levels, respectively.

Note that in column (7), when we use the first-order polynomial for the larger bandwidth of ±200 stocks, the coefficient on PMF_A% is not statistically significant. This may be due to the need for more complex functional forms to account for firms that are less comparable as we broaden the bandwidth (Atanasov & Black, 2016; Roberts & Whited, 2013). In column (9), where we use a broader bandwidth (±200 stocks) and a polynomial order of three, we recover a positive and statistically significant result. The magnitude of the effect decreases as we broaden the bandwidth; this is partially consistent with index assignment having a stronger impact where it is closer to the cutoff line. The estimated coefficients in column (9) show that a one percentage point increase in index fund ownership increases audit fees by approximately 13.7%, and the effect is statistically significant at the 5% significance level. The magnitude of economic significance is also sizable: a pooled standard deviation (pooled SD = 1.88% for the ±200 stocks bandwidth) increase in index fund ownership (PMF_A%) increases the audit fees by 0.2568 standard deviations.²³ Equally important, the first-stage F-statistic supports the relevance condition of our IV approach, as it ranges from 41.66 to 68.59. Therefore, the Cragg-Donald Wald F-statistics exceed their Stock-Yogo weak instrument critical values in all specifications, which alleviates concerns about weak instruments (Larcker & Rusticus, 2010). Overall, our estimations indicate that index fund ownership causes firms to pay more for audit services.²⁴ Therefore, we reject the null hypothesis H1.²⁵

The mechanism: auditor ratification proposals and auditor turnover

In this subsection, we identify a potential mechanism by which passive institutional investors can impact audit quality. We do this in two ways. First, we examine the impact of passive ownership on the voting outcomes from auditor ratification proposals. In our descriptive statistics, we can see that for firms in both indexes, the support for auditor ratification is consistently well over 95%. We investigate whether these proposals receive less support from firms with more passive ownership. Table 5 presents the results of these analyses.²⁶ We test our IV specification from Equations (1) and (2) using AUD_RATIF_SUPPORT as the outcome variable, which is measured as the proportion of shareholders who voted ‘for’ on auditor ratification proposals²⁷ minus the proportion of shareholders who voted ‘against.’

Table 5. Index fund ownership and auditor ratification support.

Dependent variable:	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
AUD_RATIF_SUPPORT
PMF_A%	−2.366**	−1.371*	−1.460	−2.618**	−1.889*	−1.689*	−2.506**	−1.745**	−1.411**
	(−2.078)	(−1.699)	(−1.637)	(−2.205)	(−1.832)	(−1.914)	(−2.227)	(−2.000)	(−2.019)
Observations	254	254	254	371	371	371	498	498	498
Polynomial order	1	2	3	1	2	3	1	2	3
Bandwidth	100	100	100	150	150	150	200	200	200
Float-adj M. Cap.?	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Controls?	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Fixed effects?	Year	Year	Year	Year	Year	Year	Year	Year	Year
Adj. R^2	3%	5%	6%	5%	2%	2%	6%	2%	1%
First stage F	13.10	24.97	23.01	12.81	17.19	21.03	13.93	21.44	27.80

Note: This table reports the IV estimation results to identify the effect of index fund ownership on auditor ratification support from Equation (2): $\mathrm{AUD}\mathrm{\_RATIF}\mathrm{\_SUPPOR}{T}_{\mathrm{it}}=\alpha +\beta {\widehat{\mathrm{PMF}\mathrm{\_A\%}}}_{\mathrm{it}}+\sum _{n=1}^N{\tau }_n(\text{ln}(\mathrm{MktCa}{p}_{\mathrm{it}}){)}^n+\mathrm{\phi ln}(\mathrm{Floa}{t}_{\mathrm{it}})+\varphi X_{\mathrm{it}}+{\rho }_t+{ϵ}_{\mathrm{it}},$

where variable AUD_RATIF_SUPPORT is the percentage of votes for ‘for’ minus the percentage of votes for ‘against’ for auditor ratification, and $\mathrm{PMF}\mathrm{\_}{A}_{\mathrm{it}}$ is firm $i$’s end-of-September index fund ownership based on Appel et al.’s (2016) approach in year t predicted from the first-stage regression. We control for firms’ end-of-June float-adjusted market capitalization. We show the test results for the ±100 (columns (1), (2), and (3)), ± 150 (columns (4), (5), and (6)), and ±200 bandwidths (columns (7), (8), and (9)). For each bandwidth, we also report the results controlling for the first-order polynomial of the natural logarithm of end-of-May market capitalization from CRSP (columns (1), (4), and (7)), second-order polynomial (columns (2), (5), and (8)), and third-order polynomial (columns (3), (6), and (9)). The sample period is from 2000 to 2006. The control variables include TRA, LN_TA, LEV, LN(AUDIT_TENURE), LAG_GOING_CON, ROA, SALES_GROWTH, ISSUANCE, LN(1 + NUM_ANALST), and LOSS_NI. For brevity, we do not tabulate the estimated coefficients on the control variables in this table. We present the full table for this estimation in the online appendix, Table A2. Variable definitions are presented in the Appendix. T-statistics are in parentheses. Standard errors are robust and clustered at the firm level to account for serial correlation. ***, **, and * represent significance levels at the 1%, 5%, and 10% levels, respectively.

Further, we use a set of controls different from those in the main audit fee analysis because our outcome variables are different.²⁸ Following Hennes et al. (2014), we include firm-level characteristics and determinants of auditor turnover: TRA, LN_TA, LEV, ROA, LN(1 + NUM_ANALYST), LOSS_NI (indicator that takes the value of one if the firm reports a negative net income, and zero otherwise), LN(AUDIT_TENURE) (auditor tenure), LAG_GOING_CON (indicator that takes the value of one if the auditor issues a going-concern opinion in year t–1, and zero otherwise), ISSUANCE (indicator that takes the value of one if the firm has issued either debt or equity securities in year t, and zero otherwise), and SALES_GROWTH (growth rate of sales). The negative and significant coefficient on PMF_A% in Table 5 suggests that higher passive ownership leads to less support for auditor ratification proposals.²⁹ An increase of one percentage point in passive ownership leads to a 2.37% decrease in net votes for the current auditor. This is evidence supporting the view that passive investors are not passive owners (Appel et al., 2016). This finding also represents a direct channel that connects passive investors and audit quality, one that is separate from other corporate governance mechanisms such as board independence (Appel et al., 2016) or payout policy (Crane et al., 2016).

Furthermore, we study the impact of passive ownership on subsequent auditor turnover. As Table 5 shows, passive ownership leads to less support for auditor ratification. However, such a reduction in support does not often translate into the rejection of auditor ratification proposals. On average, the percentage of votes for auditor ratification is over 95% for both firms in the Russell 1000 and 2000 indexes. The remaining votes are abstentions. Given that these proposals get passed very often, we study the impact of passive ownership on auditor turnover. We use our IV estimation procedure from Equations (1) and (2) to examine whether passive ownership affects auditor turnover. The outcome variable, Auditor_Turnover, is an indicator equal to one if the firm changes its auditor in year t + 1 and zero otherwise.

We report these results in Table 6 and use the same bandwidths as in our main analyses of audit fees.³⁰ The positive and statistically significant coefficients on PMF_A% in columns (1) to (5) suggest that higher index fund ownership increases firms’ likelihood of replacing their auditors in the following fiscal year.³¹ The results become statistically insignificant in columns (6) to (9) but remain in the same direction. This is probably because firms are less comparable when the bandwidth is wider (Atanasov & Black, 2016; Roberts & Whited, 2013).

Table 6. Index fund ownership and auditor turnover.

Dependent variable:	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
AUDITOR_TURNOVER
PMF_A%	6.193**	6.259**	6.388*	3.662*	3.750*	3.266	2.291	2.303	2.018
	(1.993)	(2.036)	(1.817)	(1.713)	(1.719)	(1.366)	(1.202)	(1.205)	(0.945)
Observations	513	513	513	803	803	803	1,093	1,093	1,093
Polynomial order	1	2	3	1	2	3	1	2	3
Bandwidth	100	100	100	150	150	150	200	200	200
Float-adj M. Cap.?	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Control variables?	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Fixed Effects?	Year	Year	Year	Year	Year	Year	Year	Year	Year
Adj. R^2	8%	9%	10%	6%	6%	7%	14%	13%	14%
First stage F	39.02	42.82	34.71	53.42	54.11	46.14	62.40	64.99	56

Note: This table reports the IV estimation results to identify the effect of index fund ownership on auditor turnover from Equation (2): $\mathrm{AUDITOR}\mathrm{\_TURNOVE}{R}_{\mathrm{it}+1}=\alpha +\beta {\widehat{\mathit{\text{PMF}}\mathrm{\_}\mathit{\text{A}}\mathrm{\%}}}_{\mathrm{it}}+\sum _{n=1}^N{\tau }_n(\text{ln}(\mathrm{MktCa}{p}_{\mathrm{it}}){)}^n+\mathrm{\phi ln}(\mathrm{Floa}{t}_{\mathrm{it}})+\varphi X_{\mathrm{it}}+{\rho }_t+{ϵ}_{\mathrm{it}},$

where variable AUDITOR_TURNOVER is an indicator that takes the value of one when firm i’s auditing firm changes in year t + 1 and zero otherwise, and $\mathrm{PMF}\mathrm{\_}{A}_{\mathrm{it}}$ is firm $i$’s end-of-September index fund ownership in year t, as predicted by Appel et al. (2016). We control for firms’ end-of-June float-adjusted market capitalization. We show the test results using the ±100 (columns (1), (2), and (3)), ± 150 (columns (4), (5), and (6)), and ±200 bandwidths (columns (7), (8), and (9)). For each bandwidth, we also report the results controlling for the first-order polynomial of the natural logarithm of end-of-May market capitalization from CRSP (columns (1), (4), and (7)), second-order polynomial (columns (2), (5), and (8)), and third-order polynomial (columns (3), (6), and (9)). The sample period is from 2000 to 2006. The control variables include TRA, LN_TA, LEV, LN(AUDIT_TENURE), LAG_GOING_CON, ROA, SALES_GROWTH, ISSUANCE, LN(1 + NUM_ANALST), and LOSS_NI. For brevity, we do not tabulate the estimated coefficients on the control variables in this table. We present the full results for this estimation, including all coefficients on the control variables, in the online appendix, Table A3. Variable definitions are presented in the Appendix. T-statistics are in parentheses. Standard errors are robust and clustered at the firm level to account for serial correlation. ***, **, and * represent significance levels at the 1%, 5%, and 10% levels, respectively.

Taken together, our results in Tables 5 and 6 consistently suggest that active voting is at least one channel through which passive owners directly influence audit-related corporate governance issues. Given the large ownership stake of passive investors, we expect that they may combine voting with other, probably more effective, mechanisms such as behind-the-curtain tactics (Carleton et al., 1998; Del Guercio & Hawkins, 1999; McCahery et al., 2016).

Cross-sectional analysis based on agency cost

In this subsection, we examine whether the impact of passive ownership on audit quality varies with firms’ agency costs. We split our sample by five broad indicators of agency costs and conduct the same tests as our main test for audit fees. We find that the positive effect of passive ownership on audit fees is more pronounced in firms with higher non-audit fee ratio, more complex asset structure (proxied by the intangibility of assets), higher ownership by transient investors, and where the CEO has higher incentives to conduct earnings management due to their compensation. However, we find no difference between firms when we split the sample by board independence: this may be due to a very similar level of board independence between the two subsamples or a power issue when the sample size is relatively small. For brevity, we present the details of this cross-sectional analysis in the online appendix, OA1. Overall, our cross-sectional analysis provides evidence that passive investors’ monitoring incentives depend on the agency cost of a firm, consistent with (1) institutional investors varying their monitoring efforts according to firm characteristics and (2) passive investors being active owners who pay attention to the audit quality of their portfolio firms.

Robustness Check

Alternative measures of audit quality

In Table 7, we report the estimation from our tests that use alternative measures of auditing quality: (1) propensity to report small profits, (2) restatements, and (3) internal control weakness.

Table 7. Index fund ownership and alternative measures of audit quality.

	SMALL_PROFITS			RESTATEMENT_AO			LN(1 + NUM_IC_W)
Dep. Var:	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
PMF_A%	−2.524*	−2.912*	−3.313*	−11.481**	−11.498**	−8.109	4.943	6.980	8.223
	(−1.670)	(−1.805)	(−1.911)	(−2.161)	(−2.201)	(−1.237)	(0.132)	(0.557)	(0.484)
Observations	540	540	540	400	400	400	219	219	219
Polynomial order	1	2	3	1	2	3	1	2	3
Audit fee controls?	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Bandwidth	100	100	100	100	100	100	100	100	100
Float-adj M. Cap.?	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Fixed effects?	Year	Year	Year	Year	Year	Year	Year	Year	Year
Adj. R^2	0.014	0.003	0.024	0.007	0.008	0.046	0.082	0.073	0.063
First stage F	34.65	34.91	32.70	25.16	29.47	19.30	28.30	24.29	19.90

Note: This table reports the IV estimation results to identify the effect of index fund ownership on alternative measures of audit quality: meeting or beating the zero earnings threshold (SMALL_PROFITS), all occurrences of a restatement (RESTATEMENT_AO), and internal control weaknesses (LN(1 + NUM_IC_W)) from Equation (2): $\mathrm{AUDIT}\mathrm{\_QUALIT}{Y}_{\mathrm{it}}=\alpha +\beta {\widehat{\text{PMF}\mathrm{\_}\text{A}\mathrm{\%}}}_{\mathrm{it}}+\sum _{n=1}^N{\tau }_n(\ln (\mathrm{MktCa}{p}_{\mathrm{it}}){)}^n+\phi \text{ln}(\mathrm{Floa}{t}_{\mathrm{it}})+\varphi X_{\mathrm{it}}+{\rho }_t+{ϵ}_{\mathrm{it}},$

where the variable AUDIT_QUALITY is either SMALL_PROFITS, RESTATEMENT_AO, or LN(1 + NUM_IC_W); and $\mathrm{PMF}\mathrm{\_}{A}_{\mathrm{it}}$ is firm $i$’s end-of-September index fund ownership based on Appel et al.’s (2016) approach in year t predicted from the first-stage regression. We present the results using the ±100 bandwidth and for the first-, second-, and third-order polynomials. In each column of the tests, we also control for firm fundamentals and the determinants of each audit quality (described in the subsection ‘Alternative measures of audit quality’), which are somewhat different for each audit quality measure. The sample period is from 2000 to 2006. The control variables in columns (1), (2), and (3) include TRA, LN_TA, ISSUANCE, LEV, ROA, LN(AUDIT_TENURE), BM, CURRENT, SPECIAL_ITEM, and LN(1 + NUM_ANALYST); and those in columns (4), (5), and (6) include TRA, LN_TA, ISSUANCE, SOFT_ASSETS, DELTA_CSALES, DELTA_ROA, DELTA_EMP, LEASE, RETURN, DELTA_REC, DELTA_INV, and TOTAL_ACCRUAL. Finally, in columns (7), (8), and (9), they include TRA, LN_TA, LEV, ALTMAN_Z, NBS, NGS, LOSS_NI, FIRM_AGE, FOREIGN, ACQUISITIONS, SALES_GROWTH, and REST_CHARGE. Variable definitions are presented in the Appendix. T-statistics are in parentheses. Standard errors are robust and clustered at the firm level to account for serial correlation. ***, **, and * represent significance levels at the 1%, 5%, and 10% levels, respectively.

First, we test the impact of passive ownership on the propensity to meet or beat the zero earnings threshold (Aobdia, 2019). Following Aobdia (2019), we use the indicator variable SMALL_PROFIT, which is equal to one if the firm’s ROA is between 0 and 0.5% in year t, and zero otherwise.³² For brevity, we report results using the narrowest bandwidth, as in the main tests, and only tabulate the coefficients on PMF_A%. We control for firm-level characteristics and known determinants of meeting or beating the zero earnings thresholds (Caramanis & Lennox, 2008; Dechow et al., 2010; McVay, 2006; Zang, 2012). These factors are TRA, LN_TA, ISSUANCE, LEV, ROA, LN(AUDIT_TENURE), LN(1 + NUM_ANALYST), SPECIAL_ITEM, BM (book value of equity over its market value), and CURRENT (current ratio). We use the ±100 bandwidth with the first-, second-, and third-order polynomials in columns (1) to (3) of Table 7, respectively. The negative coefficient on PMF_A% in columns (1) to (3) suggests that higher index fund ownership reduces the likelihood of firms meeting or beating the zero earnings threshold, indicating higher audit quality. This finding is consistent with that in the main analysis of audit fees in Table 4.³³

Second, we turn our analysis to restatements. We set the indicator variable RESTATEMENT_AO to one if a firm’s financial statement in year t is subsequently restated, and zero otherwise. We also control for firm-level characteristics and known determinants of restatements (Lobo & Zhao, 2013). These factors are TRA, LN_TA, ISSUANCE, SOFT_ASSETS (total assets minus net PPE, cash, and cash equivalents, divided by total assets), DELTA_CSALES (percentage change in cash-based sales over total assets), DELTA_ROA (percentage change in ROA), DELTA_EMP (percentage change in the number of employees), LEASE (indicator variable that takes the value of one if the firm has a rental commitment, and zero otherwise), RETURN (annual market return), DELTA_REC (percentage change in accounts receivable), DELTA_INV (percentage change in total inventories), and TOTAL_ACCRUAL (the difference between income before extraordinary items minus operating cash flow over total assets). The negative coefficient on PMF_A% in columns (4) to (6) suggests that higher index fund ownership reduces the likelihood that financial statements are subsequently restated.³⁴ This result is statistically significant for the first- and second-order polynomials at the 5% level.³⁵

Finally, in Table 7, columns (7) to (9), we report the results of estimating the association between passive ownership and material internal control weaknesses reported by auditors based on SOX 302/404 disclosure rules. We control for firm factors and known determinants of internal control weaknesses (Doyle et al., 2007). These factors are TRA, LN_TA, LEV, NBS, NGS, LOSS_NI, FIRM_AGE, SALES_GROWTH, FOREIGN, ACQUISITIONS, REST_CHARGE (an indicator variable that takes the value of one if the firm has restructuring charges during the year, and zero otherwise), and ALTMAN_Z (Altman’s Z measure following Altman (1968, 2013)). Our results indicate that the coefficient on PMF_A% is not statistically significant at conventional levels.

Taken together, the results in Table 7 show that when we use output measures for audit quality, the results are consistent with our main tests using audit fees (an input measure) as a proxy for audit quality. One caveat is that we fail to find evidence that passive investors have an impact on internal control weaknesses. Since audit fees, small profits, and restatements are the three audit quality measures that are most aligned with practitioners’ assessments, we believe that our evidence lends strong support to passive owners’ positive influence on audit quality.

Alternative measures of passive ownership

We continue our robustness checks by presenting evidence that our main audit fee result is not sensitive to how we measure passive ownership. We use three additional measures to proxy for passive ownership. First, we follow Petajisto (2013) and use the data provided by Petajisto (2013) and Cremers and Petajisto (2009) to calculate the index fund ownership (denoted as PMF_P%). For the second measure, we follow Appel et al. (2016) and use the three largest passive institutional investors’ total ownership stakes during our sample period to proxy for passive institutional ownership as a robustness check (denoted as PMF_Big3%). Finally, we use Bushee’s (2001) institutional ownership definition of quasi-indexer institutional investors (denoted as QIX).³⁶ We present the results of this robustness check in the online appendix, Table A5. We use audit fees as the measure for audit quality and apply the same IV model as in the main test presented in Table 4. We find that the results hold when we use alternative measures of passive ownership.

Using end-of-May CRSP rankings

This study follows Appel et al.’s (2016) identification strategy to test the hypothesis. For their main analyses, Appel et al. select a sample based on the rankings by Russell after float adjustments. However, Russell’s float-adjusted reweighting of stocks within an index may affect our findings because stocks at the bottom of the Russell 1000 are likely different from those at the top of the Russell 2000 in terms of liquidity. Thus, Appel et al. (2016) propose an alternative sampling choice that ranks stocks based on their end-of-May CRSP market capitalization and selects firms within a close bandwidth surrounding the 1000 threshold. This alternative ranking method should closely resemble Russell’s assignment of stocks to either the Russell 1000 or 2000 index before making within-index changes for float adjustments (Appel et al., 2016), and the sample selected within the narrow bandwidth around the cutoff should not be affected by liquidity concerns. To check the robustness of our main results, we use this alternative sampling approach. We present the results of this analysis in the online appendix, Table A6. Our main results still hold when we use this alternative sampling method.

Discussion and Conclusion

Our study adds new evidence to the recent debate on the impact of passive investing on corporate governance. We find causal evidence suggesting that passive institutional investors play an active role in improving audit quality, and that active voting is at least one measure that passive investors use to express their dissatisfaction with the incumbent auditor. Thus, our findings support the view that passive investing has merit in improving governance. Our study also shows that proxy voting is an important channel through which index funds voice their (dis)satisfaction with financial statement audit quality.

Our study has at least three limitations. First, since our sample is restricted to the period from 2000 to 2006, it is difficult to generalize our conclusions to the current capital market. However, with the increased importance of passive investing (Sullivan & Xiong, 2012) and the strengthened corporate governance efforts of large passive investors in recent years (Gomtsian, 2019), we expect passive investors to play an even more crucial role in the corporate governance sphere. Second, because of the Russell index setting, our sample is confined to an arguably narrow bandwidth around the threshold; hence, we warn the readers to be careful while generalizing the results to a wider range of U.S. midcap firms. Our third concern is that our index fund ownership measure may suffer from a downward bias because we cannot capture the (potentially very large) number of indexed investments held by, for example, pension funds.

As many have argued, one limitation of passive investing is that fund managers cannot monitor every firm; thus, the one-size-fits-all approach to gauge governance practices may have its shortcomings. Prior literature suggests that index funds use a set of screening criteria to filter out target firms (Carleton et al., 1998; Gomtsian, 2019; Koppes & Reilly, 1994). Such an approach may lack focus on firm-specific strategy or performance. However, as Appel et al. (2019) have shown, index funds may facilitate activist campaigns to push for better governance. The 2017 landmark shareholder victory – in which indexers cast the key votes – that pushed Exxon Mobil to disclose the impact of climate change on its business is a good example.³⁷ Thus, the consequences of increased passive ownership may need to be examined in a broader perspective, and any regulatory intervention to index funds’ voting power should be considered carefully.

Acknowledgments

The authors thank the Editor Beatriz García Osma and the two referees for their valuable feedback. The authors wish to thank Jason Chao for the Russel inclusion data. Florian Eugster acknowledges the support of the Handelsbanken Wallander stipend. Ting Dong acknowledges financial support from the Handelsbanken Wallander stipend and the Torsten Söderbergs Foundation. Antonio B. Vazquez acknowledges financial support from the Tore Browaldh Foundation. We also acknowledge fruitful comments from Debarati Basu (discussant), Tuukka Järvinen (discussant), Mariassunta Giannetti, Henrik Nilsson, Kenth Skogsvik, Martin Walker and participants in the 2016 Swiss Economist Abroad Conference, 2016 Nordic Accounting Conference, FIRE workshops with the Stockholm School of Economics, Stockholm University and Uppsala University, 2017 EAA Annual Congress, 2017 Swedish National Accounting Conference (Nationella Redovisningskonferensen 2017), 2018 JAAF Symposium, and the University of St. Gallen. We thank Brian Bushee and Antti Petajisto for sharing their research data.

Disclosure statement

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

Supplemental Data and Research Materials

Supplemental data for this article can be accessed online at https://doi.org/10.1080/09638180.2022.2136227.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

This work was supported by Jan Wallanders och Tom Hedelius Stiftelse samt Tore Browaldhs Stiftelse; Torsten Söderbergs Stiftelse.

Notes

1 For example, Schmidt and Fahlenbrach (2017) find evidence that increases in passive ownership lead to higher CEO power and fewer independent director appointments.

2 Anecdotal evidence on the work routine of BlackRock’s governance team can be found in ‘The Giant of Shareholders, Quietly Stirring’ published in the New York Times on May 18, 2013. (https://www.nytimes.com/2013/05/19/business/blackrock-a-shareholding-giant-is-quietly-stirring.html).

3 The case study by Carleton et al. (1998) highlights the effectiveness and the powerful influence of behind-the-scenes engagements by large institutions. More than two decades later, passive investors’ efforts in such engagements seem to have only increased. For example, BlackRock reported in its 2020 Q2 Investment Stewardship Global Engagement Summary Report that, for the first half of the year 2020, it engaged with 1,365 companies worldwide (https://www.blackrock.com/corporate/literature/press-release/blk-engagement-summary-report-2020.pdf).

4 The Economist (2015), for example, commented: ‘But the past 15 years have cast a shadow over the public company. … Governance has been weakened by the rise of passive index funds, which means that many firms’ largest shareholders are software programs. … Index funds and ETFs mimic the market’s movements, and typically take little interest in how firms are run.’

5 Concerned with the threat from index funds on corporate governance, Lund (2017) even proposes restrictions on passive funds’ voting power to ‘preserve the role of informed investors as a force for managerial discipline’ (p. 493).

6 Carleton et al. (1998) demonstrate the case of Teachers Insurance Annuity Association–College Retirement Equities Fund (TIAA–CREF), one of the largest pension funds in the U.S. During their study period. TIAA–CREF was indexing 80% of its total assets. Using a private database consisting of correspondences between TIAA–CREF and 45 other firms, Carleton et al. (1998) show that the success rate of behind-the-scenes engagements initiated by the fund is very high (95%). Another early study that touches upon the effectiveness of passive investors’ activism is Del Guercio and Hawkins (1999), which examines shareholder proposals of the largest, ‘most active’ funds (on governance issues) from 1987 to 1993. Out of the five funds in their sample, four are heavily indexed. They find that proposals sponsored by the heavily indexed funds in their sample also have broader and more substantial consequences for target firms.

7 Jahnke (2019) also contends that the ‘low-cost’ argumentation is ill-suited because, given the massive scale of index funds, the cost of engagement is negligible compared to the potential return from governance investments. His interviews with activist investors also support Appel et al. (2019) in terms of how passive investors facilitate activist campaigns.

8 With regard to how managerial ownership affects the demand for auditing, see for example, Chow (1982), Niemi (2005), O’Sullivan (2000), Gul and Tsui (2001), and Lennox (2005).

9 Our empirical analyses do not find evidence that passive ownership affects other proxies for auditor input (such as audit report lag).

10 Some other studies that have used the sample closely surrounding the Russell 1000/2000 cutoff include, for example, Boone and White (2015), Crane et al. (2016), Khan et al. (2017), Lin et al. (2018), Baghdadi et al. (2018), and Chen et al. (2019).

11 Following the recommendations from Appel et al. (2020) and Glossner (2022), we shy away from exploiting a sharp regression discontinuity design. We follow the IV design by Appel et al. (2016), which makes use of a feature of the regression discontinuity design: the IV methodology employed by Appel et al. (2016) restricts the sample to firms around the cutoff where the assignment to the Russell 2000 index is as good as random conditional on the functional form of the end-of-May market capitalization and the end-of-June Russell float-adjusted market capitalization.We thank one anonymous reviewer for suggesting this approach.

12 For more details on passive investing, see http://www.etf.com/sections/index-investor-corner/24096-swedroe-how-to-define-passive-investment.html?nopaging=1.

13 For an example of such portfolio, see https://www.thebalance.com/passive-investing-and-index-funds-2388593.

14 The total amount of institutional assets benchmarked to each index also matters. Chang et al. (2015) report that in 2005, over $200 billion of assets were benchmarked against the Russell 2000 index, whereas only $90 billion were benchmarked against the Russell 1000 index.

15 For a discussion of the different approaches, please see Appel et al. (2020).

16 We thank one of the reviewers for this suggestion. Transient institutional investors, although highly diversified, maintain their holdings for a short period of time, which is not consistent with benchmarking against an index (Bushee, 2001).

17 In unreported results, we also make use of industry fixed effects (using multiple definitions of industry classification such as the Fama-French 5, 10, 12 and 48 industry classification, 1-digit SIC code and the 2-digit SIC code adjusted for historical changes) and arrive at the same conclusions for our main results.

18 The CEO incentive ratio is a measure developed by Bergstresser and Philippon (2006) and it assesses the relation between earnings manipulation and the power of CEO equity-based incentives. In calculating this measure, we strictly follow Bergstresser and Philippon (2006), section 3.2, page 520, Equation (8) and (10). We use Core and Guay’s (2002) approach (CG) to estimate the delta of out-of-the-money options.

19 The sample period in Appel et al. (2016) starts from 1998, and they use a larger bandwidth (±250) as their main test sample. Our sample is limited to 2000–2006; therefore, we choose a narrower bandwidth to obtain more comparable firms.

20 The average value of LN_AF (natural log of audit fees) for the Russell 1000 firms is 13.63, while the value is 13.56 for the Russell 2000 firms.

21 Column (2), for example, shows that, on average, firms at the top 100 band of the Russell 2000 have a value of index fund ownership (PMF_A%) of 1.2 percentage points higher than those in the bottom 100 band of the Russell 1000 index, controlling for firm characteristics and year fixed effects. Since the average value of PMF_A% in the whole sample is 2.8 percentage points with a standard deviation of 1.8 percentage points, the estimation from our first-stage regression indicates that index assignment significantly impacts passive ownership.

22 Please see the subsection ‘Alternative measures of passive ownership’ for a robustness check of our main results using three alternative measures of passive institutional ownership, that is, PMF_P%, PMF_Big3%, and QIX.

23 This interpretation has been calculated using the regression results from Table 4, column (9).

24 In untabulated results, we also test our main specification on two sub-samples, one pre-SOX and another one post-SOX. We find that the positive effect of passive ownership on audit fees is quantitative and qualitative robust in the pos-SOX period. This could be indicative that after the passage of the SOX, passive investors pay more attention to audit quality. We thank one of the reviewers for this suggestion.

25 In untabulated results, we also test whether passive ownership affects non-audit fees. We do not find any evidence that index ownership causes changes in non-audit fees. This finding indicates that passive investors may also help enhance auditor independence. Some scholars also argue that the inherent audit risk in the top Russell 2000 list firms might be higher, which would result in higher audit fees. To alleviate such concerns, we control for a full set of audit fee determinants widely used in the audit fee literature. Besides, given that our sample is arguably random across the cutoff line, especially when the bandwidth is as small as ±100, there should not be a systematic difference in audit risk among firms in our sample. Nevertheless, we conduct a set of robustness checks, which are presented in the subsection ‘Alternative measures of audit quality.’ The robustness tests support our main results on the positive impact of passive ownership on audit quality.

26 For brevity, we suppress the control variables in Table 5. We present the full results of this estimation in our online appendix, Table A2.

27 The proposals are all sponsored by management.

28 We thank one of the reviewers for raising this point.

29 In untabulated tests, we use the proportion of ‘for’ and ‘against’ votes separately as outcome variable and carry out the same tests as in Table 5. In line with our results in Table 5, we find that passive ownership leads to more ‘against’ votes and fewer ‘for’ votes.

30 For brevity, we suppress the control variables in Table 5. We present the full results of this estimation in our online appendix, Table A3.

31 When the control variables are at their means, a one percentage point increase in passive ownership leads to a 7% increase in the probability that the firm will replace its audit firm in the following fiscal year. To ascertain the economic magnitude of the results, we use a Probit IV approach (Cameron & Trivedi, 2010) to calculate the marginal effect of changes in PMF_A%, based on the results in Table 6, column (2). We find that increasing passive ownership by one percentage point will lead to an increase of 6.91% in the probability of having an auditor turnover in the following fiscal year, holding the control variables at their mean values.

32 We arrive at similar results if we use the threshold of ROA between 0 to 3%.

33 Based on the results in Table 7, column (2), a one percentage point increase in PMF_A% leads to a decrease of 5% in the probability of meeting or beating the zero earnings threshold when the control variables are at their mean values.

34 Based on the results in Table 7, column (5), increasing passive ownership by one percentage point will lead to a 3.98% decrease in the probability of a restatement.

35 In column (6), when we use the third-order polynomial, the coefficient on PMF_A% remains negative and of a similar magnitude but is not statistically significant. This outcome may be partly due to the relatively high polynomial order. As Atanasov and Black (2016) and Roberts and Whited (2013) note, lower polynomial order is required for smaller bandwidth to capture the differences in unobservable characteristics for firms around the threshold.

36 The data are available at https://accounting-faculty.wharton.upenn.edu/bushee/.

37 https://www.wsj.com/articles/index-funds-are-the-new-kings-of-wall-street-11568799004.

Appendix

Variable definition.

Variable	Description
Dependent variables
LN_AF	Natural log of audit fee in fiscal year t
AUDITOR_TURNOVER	Indicator that takes the value of one when firm $i$ changes its auditor in the following fiscal year (t + 1), and zero otherwise
AUD_RATIF_SUPPORT	The proportion of shareholders who voted ‘for’ on auditor ratification proposals minus the proportion of shareholders who voted ‘against’
RESTATEMENT_AO	Indicator that takes the value of one if a firm’s financial statements of fiscal year t are subsequently subjected to a restatement
SMALL_PROFITS	Indicator that takes the value of one if a firm’s ROA (income before extraordinary items scaled by beginning assets) is between 0% and 0.5% in fiscal year t, and zero otherwise
LN(1 + NUM_IC_W)	The natural log of one plus the number of material internal control weaknesses in fiscal year t
Key explanatory variables of research
PMF_A%	Index fund ownership (based on Appel et al.’s [2016] approach), calculated as the number of shares outstanding held in September of year t by index funds divided by the total number of shares of firm i
PMF_P%	Index fund ownership (based on Petajisto’s [2013] classification), calculated as the number of shares outstanding held in September of year t by index funds divided by the total number of shares of firm i
PMF_Big3%	Index fund ownership by the three largest passive institutional investors during our sample period, namely Vanguard, State Street, and Barclays Bank
QIX	Quasi-indexer institutional ownership based on Bushee’s (2001) classification
Instrument:
R2000	Indicator variable that is equal to one if the company belongs to the Russell 2000 indexes, and zero otherwise
Control variables – Market capitalization
MktCap	End-of-May market capitalization based on the CRSP data
Float	End-of-June float-adjusted market capitalization.
Control variables
TRA	Transient ownership based on Bushee’s (2001) classification.
LN_TA	Natural log of end-of-year total assets (ln[at])
NBS	Number of business segments
NGS	Number of geographic segments
LEV	Leverage as measured by total liabilities (lt) scaled by total assets (at)
ROA	Return on assets (net profit [ni] divided by the opening balance of total assets [at])
INVT_SCALED	Inventory (invt) scaled by the opening balance of total assets (at) at the beginning of the year
RECEIVABLES_SCALED	Receivables (rect) scaled by the opening balance of total assets (at)
SPECIAL_ITEM	Indicator variable equal to one if the firm reports a special item (spi), and zero otherwise
LOSS_NI	Indicator variable equal to one if the firm reports a loss in net income (ni), and zero otherwise
LN(1 + NUM_ANALYST)	Natural logarithm of one plus the number of analysts following a firm
LN(AUDIT_TENURE)	Natural logarithm of the auditor tenure with the client firm
LAG_GOING_CON	Lag value of an indicator variable that takes the value of one if the firm has been issued a going concern opinion in year t–1
SALES_GROWTH	Growth rate of sales (sale) compared to the previous year
ISSUANCE	Indicator variable that takes the value of one if the firm has issued either debt or equity securities in year t (dltis > 0 or sstk > 0), and zero otherwise
LEASE	Indicator variable that takes the value of 1 if the firm has rental commitments (mrct ≥ 0), and zero otherwise
CURRENT	The ratio of current assets (act) to current liabilities (lct)
BM	Book value of equity (ceq) divided by market value of equity (prcc_f*csho)
FIRM_AGE	Number of years since the firm first appeared in the CRSP database
ALTMAN_Z	Measure of financial distress following Altman (1968, 2013)
FOREIGN	Indicator variable that takes the value of one if there is a foreign currency adjustment in the financial statement, and zero otherwise
ACQUISITIONS	Indicator variable that takes the value of one if the firm has made an acquisition during year t, and zero otherwise
REST_CHARGE	Indicator variable that takes the value of one if the firm has restructuring charges during the year, and zero otherwise
SOFT_ASSETS	Total assets minus net property, plant, and equipment; and cash and cash equivalents, divided by total assets
DELTA_CSALES	Percentage change in cash-based sales over total assets (at). Cash sales are approximated as sales (sale) minus the change in accounts receivables (rect)
DELTA_ROA	Percentage change in ROA in year t relative to year t–1
DELTA_EMP	Percentage change in the number of employees (emp) in year t relative to year t–1
RETURN	Annual market return (prcc_f)
TOTAL_ACCRUALS	Income before extraordinary items (ib) minus operating cash flow (oancf), divided by total assets (at) in year t
DELTA_REC	Percentage change in accounts receivable (rect) in year t relative to year t – 1
DELTA_INV	Percentage change in total inventories (invt) in year t relative to year t – 1
Split variables
NON_AUDIT_FEE_RATIO	The ratio of non-audit fees divided by total fees in fiscal year t.
INTANGIBLE_INTENSITY	The ratio of R&D (xrd) over total assets (at) for fiscal year t.
TRAN_SHAREHOLDERS	Transient ownership based on Bushee’s (2001) classification.
CEO_INCENTIVE_RATIO	Calculated following Bergstresser and Philippon (2006). We also use Core and Guay’s (2002) approach to estimate the delta of out-of-the-money options. The variable is 0.01*share price*(shares held by the CEO + number of options held by the CEO) divided by (0.01*share price*[shares held by the CEO + number of options held by the CEO] + CEO fixed salary + CEO bonuses).
BOARD_INDEPENDENCE	Percentage of independent board members during fiscal year t.

Note: We provide the Compustat variable names in brackets.