Abnormal ownership concentration and the value-relevance of accounting information

Agency theory reflects an inherent conflict between shareholders (the principals) and managers (the agents). In a seminal paper, Jensen and Meckling (1976) use agency theory to explain the extent to which large shareholders dominate the ownership structure of firms (ownership concentration). For example, large shareholders can use private information to monitor managers, reducing agency cost to the benefit of all shareholders (Fan and Wong, 2002). However, concentrated ownership can also create a conflict of interest between shareholders, if large shareholders use their informational advantage to expropriate outside investors (Zhao and Millet-Reyes, 2007; Tigero et al., 2023). Large shareholders therefore have access to private information and incentives to limit the extent to which this information becomes public (Fan and Wong, 2002). Consequently, different ownership structures could result in divergent financial reporting outcomes (Firth et al., 2007).

Investor perceptions of accounting information reflect as value-relevance, namely the association between accounting information and market value of equity (Barth et al., 2001). Theoretically, if ownership concentration results in accounting information that primarily serves the needs of large shareholders, its value-relevance should decrease because the information needs of other shareholders are ignored (Firth et al., 2007). Alternatively, large shareholders might influence firms to improve value-relevance through higher financial reporting quality, as this maximises the value of their own shareholding (Fan and Wong, 2002; Edmans, 2009).

Existing empirical evidence is inconclusive. Several prior studies reflect that high ownership concentration is associated with lower earnings value-relevance in different countries (Fan and Wong, 2002; Zhao and Millet-Reyes, 2007). However, there is also evidence that high ownership concentration supports pricing during crisis periods (Mitton, 2002; da Cunha and Bortolon, 2016; Rao et al., 2022) and that high ownership concentration is associated with higher value-relevance in China (Ma et al., 2010) and the United Kingdom (Donnelly and Lynch, 2002). It is also possible that high ownership concentration does not moderate the value-relevance of all accounting information in the same way. For example, high ownership concentration increases the use of private communication, reducing demand for quality earnings with high value-relevance (Firth et al., 2007). Therefore, concentrated ownership could be associated with a decrease in earnings value-relevance. Less decision-useful earnings cause investors to make greater use of other information, such as book values (Zhao and Millet-Reyes, 2007; Barth et al., 2023). High ownership concentration could therefore moderate the value-relevance of earnings and book values in opposite directions. The objective of this study is to deepen our understanding of the relationship between ownership concentration and the value-relevance of earnings and book values.

Laws and regulations offering legal redress to minority shareholders lower expropriation risk and the agency cost associated with high ownership concentration (Tigero et al., 2023). Lower expropriation risk and reduced agency cost increase firm value and the value-relevance of accounting information (Firth et al., 2007). Our South African research setting reflects strong protection of minority shareholders during our sample period of 2010–2019 (World Bank, n.d.), mitigating expropriation opportunities of high ownership concentration ^[1]. Furthermore, although sophisticated, the South African equity market comprises a relatively small number of firms, increasing the probability of common shareholdings between investors (Titley, 2019; Hendrickse, 2022). Common shareholdings reduce the potential for monopolistic profits (Münster and Walther, 2021) and decrease the returns of concentrated ownership. These characteristics weaken the role of ownership concentration, and we find that ownership concentration does not moderate the value-relevance of earnings. However, both concentrated and dispersed ownership are associated with higher value-relevance of book values, suggesting that extreme ownership structures are associated with lower decision-usefulness of earnings (i.e. lower earnings quality).

We further investigate an earnings quality explanation by using unique South African regulatory requirements, which require listed firms to report “headline earnings” in addition to complying with International Financial Reporting Standards (IFRS). Headline earnings exhibit the characteristics of high-quality earnings, being smoother, more persistent, more value-relevant and more predictable than IFRS earnings (Stainbank and Harrod, 2007; Venter et al., 2013, 2014). However, unlike other non-GAAP earnings measures, headline earnings are prescribed and audited (Badenhorst and von Well, 2023a). Headline earnings are therefore less subject to management discretion than other non-GAAP earnings measures (Badenhorst and von Well, 2023a) ^[2]. Using headline earnings to separate IFRS earnings into high-quality and low-quality components reveals that concentrated ownership is associated with lower value-relevance of high-quality earnings but that dispersed ownership has no moderation effect. Therefore, an earnings quality explanation does not fully explain the higher value-relevance of book values associated with both ownership structures.

However, Jensen and Meckling (1976) argue that an optimal level of ownership concentration minimises agency costs for a specific firm. We therefore posit that investors have broad consensus on an optimal level of ownership concentration for a firm and that deviations from this level (“abnormal ownership concentration”) convey decision-useful information. For example, large investors might use their informational advantage to avoid firms with a weak performance outlook or low reporting quality, influencing decisions of investors with fewer resources. Although prior research identifies firm characteristics associated with ownership concentration (Callen et al., 2010; Wang, 2013), we are the first to our knowledge to employ firm characteristics to derive a measure of abnormal ownership concentration and investigate its association with the value-relevance of accounting information.

Abnormal ownership concentration is conceptually akin to abnormal (discretionary) accruals pioneered by Jones (1991). We apply a similar methodological approach and measure abnormal ownership concentration as the residual from an initial regression. Results show that abnormal ownership concentration is generally associated with higher value-relevance of book values. Although abnormal high ownership concentration does not moderate the value-relevance of IFRS earnings, both abnormally high and abnormally low ownership concentration are associated with lower value-relevance of high-quality earnings.

We contribute to the existing literature in several ways. Firstly, we present evidence that deviations of ownership concentration from expected levels reveal more about its association with value-relevance than absolute levels thereof. This reconciles prior research findings that both concentrated ownership (Fan and Wong, 2002; Firth et al., 2007) and dispersed ownership (Mitton, 2002; Ma et al., 2010; da Cunha and Bortolon, 2016) could be associated with lower value-relevance of accounting information, by revealing that any level of ownership concentration can be detrimental if it deviates from what firm characteristics dictate that it should be. Secondly, we find evidence that abnormal ownership concentration is associated with lower value-relevance of high-quality earnings. This explains why prior research finds that concentrated ownership can both support reporting quality (Dou et al., 2018) and detract from it (Saha et al., 2019). The relationship between ownership concentration and financial reporting quality is therefore nuanced and deviations from expectations matter more than actual ownership structures. Thirdly, our findings collectively provide empirical support for the theoretical construct of Jensen and Meckling (1976), namely that optimal ownership concentration, which minimises agency costs, is unique to every firm. This provides a potential framework to better understand the role of ownership concentration in other contexts, such as operational performance.

Our findings have implications for markets with similar institutional characteristics, such as high investor protection and a low number of listed firms. They suggest, for example, that abnormal ownership concentration hampers the ability of higher financial reporting quality to improve capital market outcomes. Investors and preparers aiming to maximise firm value will therefore be interested in our findings, which show that optimising ownership structure deserves attention equal to increasing financial reporting quality. Value-relevance and ownership concentration researchers will also be interested in our findings which introduce the concept of abnormal ownership concentration and deepens our understanding of the association between ownership concentration and decision-useful information.

The next section of this paper contains the background, literature review and hypotheses development. This is followed by a discussion of the research methodology and data in two separate sections. Detailed findings are presented thereafter, while the final section summarises and concludes the paper.

In various theoretical models, agency theory explains differences in ownership structures (Jensen and Meckling, 1976) and relationships between shareholders (Shleifer and Vishny, 1986). Large shareholders have access to private information and could choose to use this in their own interest, thus expropriating outside investors (Zhao and Millet-Reyes, 2007; Tigero et al., 2023). They therefore have incentives to keep certain information out of the public domain (Fan and Wong, 2002). Concentrated ownership could cause financial reporting to focus on the needs of a few shareholders, with the result that the information needs of other shareholders are ignored and value-relevance decreases (Firth et al., 2007). Several studies present empirical evidence consistent with this theoretical expectation for countries in Europe (Clark and Wójcik, 2005), Asia (Fan and Wong, 2002; Firth et al., 2007) and North America (Ghosh and Moon, 2010).

However, large shareholders also use private information to monitor managers (Shleifer and Vishny, 1986). The theoretical model of Shleifer and Vishny (1986) predicts that large shareholders could act to maximise firm value to compensate for bearing monitoring costs, which actions could include making information public to increase the attractiveness of the firm as a takeover target. Therefore, large shareholders could pressure firms to release more value-relevant information to maximise the value of their own holdings (Fan and Wong, 2002; Hu and Izumida, 2008). Consistent with this counterargument, prior research finds that concentrated ownership is associated with higher firm value in China (Ma et al., 2010) and higher earnings response coefficients in the UK (Donnelly and Lynch, 2002). Moreover, high ownership concentration supports value-relevance during crisis periods (Mitton, 2002), including macroeconomic shocks (da Cunha and Bortolon, 2016) and extreme rainfall patterns (Rao et al., 2022). Consequently, both theoretical models and empirical evidence lacks consensus on whether large shareholders use their influence to benefit only themselves or to benefit all shareholders.

When large shareholders exercise their influence for their own benefit, they limit the amount of information that becomes public (Fan and Wong, 2002). Information asymmetry increases and the value-relevance of accounting information decreases (Firth et al., 2007). In contrast, when large shareholders pressure firms to make information public to increase firm value, information asymmetry decreases and the value-relevance of accounting information increases (Rao et al., 2022; Barth et al., 2023).

The characteristics of a research setting determine research expectations (Firth et al., 2007; Konijn et al., 2011), as ownership concentration varies based on country-level institutional factors (La Porta et al., 1999; Richter and Weiss, 2013). Some characteristics of our South African setting suggest limited benefits from high ownership concentration, irrespective of large shareholders’ motivations. Firstly, strong protection of minority shareholder rights in South Africa during our sample period (World Bank, n.d.) reduces the likelihood that large shareholders can expropriate minorities for their own benefit. Secondly, ownership concentration affects value-relevance through information asymmetry (Shleifer and Vishny, 1986; Rao et al., 2022). Firms headquartered in main financial centres (i.e. with head offices in main financial centres) have greater analyst coverage, resulting in lower information asymmetry (Farooq and Zarouali, 2016). This places smaller and larger shareholders on a more equal footing, and Farooq and Zarouali (2016) find that ownership concentration is not associated with firm value in India for firms headquartered in main financial centres. As a far smaller country, South African firms are all headquartered in or close to the two main financial centres (Cape Town and Johannesburg), potentially negating any moderating effect of ownership concentration on the value-relevance of accounting information ^[3]. Thirdly, the South African equity market comprises a small number of firms. Large, sophisticated investors reduce firm-specific risk by investing in multiple firms, even if these firms operate in the same industry (Titley, 2019; Hendrickse, 2022). Common shareholdings in the same industry decrease the potential for monopolistic profits (Münster and Walther, 2021), reducing the returns (benefits) of high ownership concentration.

However, although these characteristics imply that the ownership of many South African firms will be concentrated amongst shareholders with larger holdings (due to the low number of investable firms), they also result in ownership structures without dominating shareholders (due to attempts to diversify). Greater dispersion of large shareholders is associated with lower firm value as a greater number of large shareholders imply more conflicting interests (Konijn et al., 2011). These conflicts also lower value-relevance of accounting information. Nel et al. (2021) find that high ownership concentration for South African firms is associated with lower dividends, which suggests that large shareholders prefer to extract private benefits of control rather than following a policy which would benefit all shareholders equally.

Therefore, whether and to which extent ownership concentration might moderate the value-relevance of accounting information in South Africa is unclear. While most prior research on the moderating effect of ownership concentration focuses on earnings value-relevance (Fan and Wong, 2002; Firth et al., 2007; Ghosh and Moon, 2010), book values represent alternative summary accounting information. In related research, Zhao and Millet-Reyes (2007) conclude that family control decreases the demand for timely information about firm performance. They find that the value-relevance of earnings, a timelier measure, decreases in this context, but that the value-relevance of book values increases. This is consistent with findings that other sources of information compensate when the decision-usefulness of earnings declines (Callao et al., 2016; Barth et al., 2023). Therefore, we state our first (related) hypotheses in the null form as follows:

While agency theory potentially explains the volume of information that firms with concentrated ownership make public, it can also explain reporting quality differences. When larger investors use private communication for their own benefit, their need for high-quality public information reduces (Firth et al., 2007). For example, Velury and Jenkins (2006) find that high ownership concentration detracts from earnings quality, Saha et al. (2019) show that the financial reports of firms with high ownership concentration frequently lack important disclosures and Grassa et al. (2021) conclude that higher ownership concentration reduces the quality of risk disclosures for banks in Islamic countries. The quality of financial reporting affects the value-relevance of accounting information (Stainbank and Harrod, 2007; Venter et al., 2014). Relatedly, Callao et al. (2016) find that high ownership concentration exacerbates weak value-relevance of low-quality earnings.

In contrast, large shareholders may seek to enforce high reporting quality to maximise the value of their own interest (Fan and Wong, 2002; Hu and Izumida, 2008). Furthermore, prior research posits that dispersed owners find it more difficult to enforce high reporting quality (Edmans, 2009; Dou et al., 2018). These studies present evidence that large shareholders are associated with higher financial reporting quality, provided that the threat of exiting their shareholding is realistic. Arguably, the small number of listed firms in South Africa limits alternatives and reduces the likelihood of a complete exit by large shareholders. Therefore, it appears less probable that high ownership concentration will be associated with more decision-useful earnings. Consequently, our second hypothesis (in alternative form) is:

Jensen and Meckling (1976) argue that an optimal level of ownership concentration exists at which agency costs for a specific firm is minimised. Moreover, several researchers identify firm characteristics associated with cross-sectional differences in ownership concentration (Bushee, 2001; Callen et al., 2005, 2010; Banghøj and Plenborg, 2008; Wang, 2013). Similar to the manner in which specific firm characteristics are associated with a normal (expected) level of accruals (Jones, 1991), we argue that firm characteristics predict what the normal (expected) level of ownership concentration of a specific firm should be. Deviations from this expectation (abnormal ownership concentration) could contain decision-useful information. For example, lower than expected ownership concentration indicates a lack of interest from large investors, reflecting concerns such as misleading disclosures, lack of growth opportunities or financial distress. Other investors might value the opinions of arguably the most knowledgeable investors, weakening the relationship between accounting information and firm value (i.e. resulting in lower value-relevance). However, although abnormally high ownership concentration could reflect enthusiasm from large investors, it also increases the risk of expropriation for minorities (Bae and Jeong, 2007). Therefore, the direction in which abnormal ownership concentration could moderate value-relevance is unclear and our third (related) hypotheses are stated in null form:

We follow prior research (Venter et al., 2014; Rainsbury et al., 2015) and base our value-relevance investigations on a simplified Ohlson (1995) model. After including variables for ownership concentration (full definitions of variables are in the  Appendix), the model is as follows (firm and year subscripts have been suppressed) ^[4]:

where P is share price, three months after reporting date to allow for the information dissemination process ^[5]; BPS is book value per share and EPS is basic earnings per share. Variables are determined as per-share variables, as this reliably compensates for scale effects in financial data (Barth and Clinch, 2009; Aledo Martinez et al., 2020). As firms with negative earnings are priced differently from firms with positive earnings (Hayn, 1995), we follow prior research (Rainsbury et al., 2015; Badenhorst and von Well, 2023b) and include NEG as a control variable.

Following prior research (Banghøj and Plenborg, 2008; Elshandidy, 2014; Münster and Walther, 2021), we measure ownership concentration as the total percentage strategic shareholding (NOSHST) on LSEG Workspace ^[6]. NOSHST accumulates the shareholding of every party (including management, institutional investors, other firms and individuals) who holds more than 5% of a firm’s shares. As the impact of ownership concentration inflects at very high or low levels (Fan and Wong, 2002; da Cunha and Bortolon, 2016), we sort sample firm-years into quintiles based on the ranking of NOSHST. We first analyse quintiles one to four, where OWN is set to one if ownership concentration falls into quintile one and zero otherwise. Thereafter, we analyse quintiles two to five, where OWN is set to one if ownership concentration falls into quintile five and zero otherwise. This ensures a consistent base sample (quintiles two to four) against which the impact of dispersed ownership concentration (quintile one) and concentrated ownership (quintile five) is evaluated. The variables of interest are the interactions of OWN with BPS and EPS, which reflect the moderating effect of ownership concentration on the value-relevance of accounting information.

Outsiders to a firm use proxies to measure high-quality earnings, which are fraught with measurement error (Dou et al., 2018). Furthermore, earnings quality is a complex construct, as high-quality earnings should be smooth, persistent, have predictive power and be value-relevant (Dichev and Tang, 2008, 2009; Ribeiro et al., 2019). However, firms listed on the JSE are required to report headline earnings in addition to complying with IFRS (Venter et al., 2014; Howard et al., 2019). The calculation and disclosure requirements for headline earnings adjustments are prescribed in a circular issued by the South African Institute of Chartered Accountants (SAICA, 2019), which is regularly updated for changes in IFRS ^[7]. Unlike other non-GAAP earnings measures, the content of headline earnings is therefore prescribed (firms are not permitted to deviate from the circular) and audited (Venter et al., 2014; Howard et al., 2019). Audited numbers are of higher quality (Lennox and Pittman, 2011; Ball et al., 2012) and, combined with regulation, this ensures that headline earnings are calculated consistently over time and between firms (Venter et al., 2014). Consequently, headline earnings do not reflect the unique reporting incentives of an individual firm. Although all earnings measures incorporate management discretion, headline earnings, as a regulated earnings measure, do not permit greater management discretion than the IFRS earnings from which headline earnings are derived.

Headline earnings are high-quality earnings, being smoother, more persistent, more value-relevant and more predictable than IFRS earnings (Stainbank and Harrod, 2007; Venter et al., 2014). As a disclosure prepared by the firm (in accordance with the prescriptions of the circular), headline earnings are also free from confounding effects that arise when investors attach greater weight to measures that they calculate themselves (Nelson and Tayler, 2007). Therefore, headline earnings provide a meaningful measure of high-quality earnings.

To investigate the association between ownership concentration and high-quality earnings, we decompose IFRS earnings into two components, namely headline earnings (high-quality earnings) and the difference between IFRS earnings and headline earnings (low-quality earnings). The resultant model reflects as follows:

where HEPS is headline earnings per share and APS is the adjustment (difference) between EPS and HEPS. We follow prior value-relevance research (Venter et al., 2014; Rainsbury et al., 2015) and specify HEPS and APS so that the sum of the two variables adds up to EPS. All other variables are as previously defined.

Abnormal ownership concentration is conceptually similar to the abnormal accruals of Jones (1991). Therefore, we similarly measure abnormal ownership concentration as a residual from the following regression (full definitions of variables are included in the  Appendix):

where NOSHST is the total percentage strategic shareholding; LIQ is the liquidity of shares traded during the reporting period (Bushee, 2001; Callen et al., 2005); VOL controls for information asymmetry (Firth et al., 2007); LEV is financial gearing (Bushee, 2001; Callen et al., 2005; Banghøj and Plenborg, 2008); MTB controls for intangible intensiveness and growth opportunities (Donnelly and Lynch, 2002; Firth et al., 2007) ^[8]; BETA is a measure of risk (Callen et al., 2005, 2010; Banghøj and Plenborg, 2008); AGE is firm age (Wang, 2013); AR is the analyst rating for the firm if available (Callen et al., 2005); AF reflects whether a firm has attracted an analyst following; DY is the dividend yield (Callen et al., 2005); SIZE is firm size (Firth et al., 2007; da Cunha and Bortolon, 2016); MAR controls for price momentum (Bushee, 2001; Callen et al., 2005); ΔEPS controls for firm performance (Bushee, 2001; Callen et al., 2010); and NEG is as earlier defined.

As actual ownership concentration (NOSHST) is the dependent variable, the residual of model (3) is the ownership concentration which is not predicted by these firm characteristics and therefore represents unexpected (i.e. abnormal) ownership concentration. By construction, the residual of a regression does not correlate with the dependent variable. Therefore, abnormal ownership concentration does not correlate with ownership concentration.

Ownership concentration is affected by institutional characteristics of the research setting (Konijn et al., 2011), which endure over long periods of time. Furthermore, the number of firms per industry is low. We therefore run model (3) on the whole sample to derive firm-year residuals. To control for differences across firms and years, we include firm and year fixed effects in models (1) and (2) ^[9]. As the effect of ownership concentration inflects at very high or low levels (Fan and Wong, 2002; da Cunha and Bortolon, 2016), we again rank the residuals into quintiles to measure abnormal ownership concentration (OWN_AB). Thereafter, we substitute OWN with OWN_AB in models (1) and (2) for analyses.

The preliminary sample for this study is all JSE tickers (dead and live) on LSEG Workspace with reports from 1 January 2010 to 31 December 2019. Recession events confound value-relevance results (Kane et al., 2015). Therefore, this sample period simultaneously avoids the 2007–2008 financial crisis and the global pandemic that started in 2020 (Badenhorst and von Well, 2023b). Data for all variables are obtained from LSEG Workspace, apart from basic earnings and headline earnings which are hand-collected (as reported) from results announcements. Where applicable, data is downloaded in South African rand (ZAR). Where the reporting currency is not ZAR, we follow Howard et al. (2019) and convert hand-collected data using the exchange rate on the database.

Following prior research (Choi et al., 2007; Barton et al., 2010; Venter et al., 2014), we reduce the impact of outliers by trimming observations at the 1% and 99% levels ^[10]. The sample reconciliation in Panel A of Table 1 shows that, after trimming, the final sample consists of 2026 firm-years (361 unique firms). Panel B reveals that no single industry dominates the sample, while Panel C reflects an even distribution across sample years. Table 1 also includes general descriptive statistics in Panel D. With a mean (median) of 2.817 (0.926) per share, headline earnings per share (HEPS) is higher than basic earnings per share (EPS), which has a mean (median) of 2.767 (0.902) per share. Notably, the mean (median) adjustment from IFRS earnings to headline earnings (APS) is small at 0.050 (0.000) per share. This arguably reflects the effectiveness of prescribing the headline earnings calculation. Furthermore, the distribution of APS is comparable to statistics reported in Venter et al. (2014). Finally, sample firms are generally profitable as only 14.8% of firm-years reflect a basic loss per share. Strategic ownership appears to be evenly distributed, as the mean (48.6%) and median (48.0%) of NOSHST are close together.

A Hausman test is significant and we therefore include firm and year fixed effects (Onali et al., 2017). We evaluate significance in regressions with reference to robust standard errors clustered by firm and year, corrected for the nestedness of fixed effects (Petersen, 2009; Cameron et al., 2011).

We start by evaluating whether our sample exhibits expected value-relevance characteristics. The first column in Panel A of Table 2 shows that book value and earnings are both positive and significant at the 1% level. Consistent with prior research (Hayn, 1995; Venter et al., 2014), this panel also reveals differences between firms with positive and negative earnings, where earnings are not significant when a firm reports a loss (p = 0.172). In the interest of brevity, the text that follows discusses results which include all sample firms (the “full sample”), although results are also separately tabulated for firm-years with positive earnings and those with negative earnings. Panel A shows that the coefficient of book value is reasonably close to one as predicted by theory and prior research (Barth et al., 2001; Aledo Martinez et al., 2020). The coefficient of earnings is far greater than one, which is similarly consistent with the expectation that earnings represent a proxy for unrecognised assets (Barth et al., 2001; Aledo Martinez et al., 2020).

Thereafter, we also consider the value-relevance of headline earnings. We code the variables so that headline earnings (HEPS) plus the adjustment (APS) equal IFRS earnings (EPS). If headline earnings represent the most value-relevant earnings measure, any incremental adjustment to this amount should not be value-relevant (Venter et al., 2014; Rainsbury et al., 2015). Consistent with this expectation, APS is insignificant in Panel B of Table 2. Similar to Rainsbury et al. (2015), we find that book value is insignificant when a non-GAAP earnings measure is included. However, as expected, HEPS is positive, has a coefficient far greater than one and is significant at the 1% level. These findings are consistent with Ball et al. (2020), who find that book values have explanatory power for firm value (returns) when they compensate for earnings volatility. As headline earnings are smoother and more persistent than IFRS earnings (Venter et al., 2014), book values no longer play a compensatory role and their value-relevance declines.

We also use a Vuong (1989) test to evaluate the relative value-relevance of headline earnings and IFRS earnings. Results are presented in Panel C of Table 2 ^[11]. Consistent with Venter et al. (2014), these results indicate a preference for headline earnings (p < 0.001) and this preference remains (p < 0.001) if only the competing earnings measures are included in the models (i.e. APS is omitted from the second model).

From these investigations, we conclude that our sample reflects expected value-relevance characteristics. Higher quality earnings are expected to have higher value-relevance (Venter et al., 2014; Ribeiro et al., 2019) and these results therefore add further evidence that headline earnings are an appropriate measure of high-quality earnings.

We first sort firms into quintiles of ownership concentration based on actual ownership concentration. Panel A of Table 3 shows that, on average, strategic shareholders own 18.9% of the bottom quintile of firms, which increases to 80.8% for the top quintile. All the firm characteristics associated with ownership concentration included in model (3) differ significantly between firms in the bottom and top quintiles.

The impact of ownership concentration on value-relevance is further explored in Panel B. If large shareholders use superior access to private information to expropriate outside shareholders, high ownership concentration should be associated with lower earnings value-relevance (Fan and Wong, 2002; Tigero et al., 2023). However, agency theory suggests that widely dispersed shareholders struggle to monitor management, which could also detract from the decision-usefulness of accounting information (Shleifer and Vishny, 1986; Konijn et al., 2011). Therefore, the variables of interest are the interaction variables, which alternatively reflect the association of low (bottom quintile) and high (top quintile) ownership concentration with the value-relevance of accounting information. Both dispersed ownership (p = 0.015) and concentrated ownership (p = 0.027) are associated with higher value-relevance of book values. However, the earnings interactions are insignificant. When the value-relevance of book values increases, this suggests that the decision-usefulness of earnings has decreased (Zhao and Millet-Reyes, 2007; Barth et al., 2023). The results could therefore imply that extreme ownership structures reflect less decision-useful (lower quality) earnings.

Deeper investigations into a potential earnings quality explanation are reflected in Panel C. These show that, when IFRS earnings are replaced with headline earnings, inferences for book value interactions are unchanged. However, while concentrated ownership significantly detracts from the value-relevance of headline earnings (p = 0.005), dispersed ownership concentration does not impact on the value-relevance of headline earnings (p = 0.303). This is consistent with prior research findings that large shareholders use private information for their own benefit, reducing demand for high-quality public information (Firth et al., 2007; Callao et al., 2016). Concentrated ownership therefore does not moderate overall earnings value-relevance, but is associated with lower value-relevance of high-quality earnings. As the high-quality component of earnings is more decision-useful, this increases the possibility that investors use accounting fundamentals suboptimally.

However, agency theory also suggests that an optimal ownership structure exists for each firm whereby large shareholders are compensated for the additional monitoring costs that they bear (Jensen and Meckling, 1976; Shleifer and Vishny, 1986). Therefore, concentrated ownership (and greater use of private information) could be optimal for specific firms. We first measure abnormal ownership concentration as the residual from model (3). In untabulated results we find significance for several of these firm characteristics with an adjusted R² of 25.9% (F = 55.53, p < 0.001). While R² should be compared across studies with caution (Gu, 2007), it is noteworthy that the explanatory power falls within the 4.8%–30.7% range reported by Bushee (2001) for different types of large shareholders. It is also higher than the 6.1% documented by Richter and Weiss (2013) for the explanatory power of firm-level characteristics for ownership concentration ^[12].

After ranking firms into quintiles based on abnormal ownership concentration, Panel A of Table 4 shows that strategic shareholders own an average of 22.7% of bottom quintile firms (where ownership concentration is much lower than expected) and 76.4% of top quintile firms (where ownership concentration is much higher than expected). However, this is the only significant difference between the bottom and top quintile, implying that abnormal ownership concentration is randomly distributed through the sample (i.e. abnormal ownership concentration is uncorrelated with the firm characteristics from which it is derived). This also suggests that any ownership structure (i.e. dispersed or concentrated) can deviate substantially from the theoretically optimal level for a specific firm.

Panel B of Table 4 shows that abnormal concentrated ownership is associated with higher value-relevance of book values, but the association for abnormal dispersed ownership is insignificant. Abnormal dispersed ownership is mildly associated with lower earnings value-relevance (p = 0.102), but the interaction for abnormal concentrated ownership is not significant (p = 0.253). Panel C contains results when high-quality and low-quality earnings components are distinguished. It shows that, when the earnings coefficient is permitted to vary, abnormal dispersed ownership and abnormal concentrated ownership are both associated with higher value-relevance of book values (p < 0.001). Furthermore, in both cases, the interaction with high-quality earnings is significantly negative (p < 0.010). Consequently, abnormal ownership concentration is associated with lower value-relevance of high-quality earnings and higher value-relevance of book values. As Jensen and Meckling (1976) predict in their agency theory model, deviations from the optimal ownership structure for a firm are detrimental for decision-making irrespective of whether the structure is concentrated or dispersed. In other words, abnormal ownership concentration (suboptimal ownership structures) hinder market participants’ ability to evaluate the quality of earnings, which is associated with greater likelihood that accounting fundamentals are perceived incorrectly.

As abnormal ownership concentration is associated with lower value-relevance of high-quality earnings, we expect that headline earnings will have the highest value-relevance for firms where ownership concentration is closest to being optimal (i.e. where abnormal concentration is lowest) ^[13]. Consistent with this prediction, when we code OWN_AB to equal one for firm-years in the middle quintile and zero otherwise, Table 5 reflects higher value-relevance of headline earnings for these firm-years (p = 0.005). Interestingly, the value-relevance of IFRS earnings is also higher (p = 0.052). Finally, investors appear to rely less on book values for these firms, with OWN_AB*BPS negative and significant (p < 0.050) in both analyses. This is consistent with theoretical expectations that an optimal ownership structure distributes the agency cost fairly amongst shareholders (Jensen and Meckling, 1976; Shleifer and Vishny, 1986).

In the main analyses, we ranked the signed (i.e. positive/negative) residual into abnormal dispersed ownership (the bottom quintile) and abnormal concentrated ownership (the top quintile). However, it might be that the size of abnormal ownership deviation matters more than the sign. We therefore also present results in Table 6 where the absolute value of the residual is ranked into quintiles. Here, the bottom quintile represents firms with the lowest abnormal ownership concentration. Consistent with the main regression results, these firms reflect lower value-relevance for book values (p < 0.050) and higher value-relevance for both IFRS (p = 0.025) and high-quality earnings (p < 0.001). For firms with the highest absolute abnormal ownership concentration, most interactions are insignificant. However, we detect an increase in the value-relevance of book values (p = 0.088) in Panel B, but also an increase in value-relevance for low quality earnings (p = 0.036). These findings suggest that the sign of abnormal ownership concentration matters more than its size.

In untabulated results, we also consider whether smaller differences in both actual and abnormal ownership concentration could explain value-relevance differences, by ranking firm-years above and below the median. Although we continue to note some cross-sectional differences, results are generally much weaker and often insignificant. It therefore appears, consistent with prior research (Fan and Wong, 2002; da Cunha and Bortolon, 2016), that the impact of ownership concentration inflects at very high or very low levels.

All empirical papers are affected by endogeneity concerns (Chenhall and Moers, 2007). Endogeneity should be primarily addressed through theory and logic, and econometric techniques should only be implemented when their use is clearly appropriate (Chenhall and Moers, 2007; Larcker and Rusticus, 2007; Gow and Ding, 2024). Importantly, no econometric approach exists which can eliminate endogeneity concerns (Chenhall and Moers, 2007; Larcker and Rusticus, 2007). However, research findings still have value even when endogeneity limits causal inferences (Chenhall and Moers, 2007).

In respect of one type of endogeneity, namely omitted variable bias, we include firm and year fixed effects which control, to some extent, for unobserved (omitted) firm characteristics (Breuer and deHaan, 2024). In robustness tests, we also include control variables for additional firm characteristics, namely gearing (LEV), intangible intensiveness and growth opportunities (MTB), firm size (SIZE), dividend yield (DY), firm age (AGE) and whether a firm has an analyst following (AF) in all regressions. For brevity, only the interaction terms of the various investigations are displayed in Table 7 and reflect that, despite small differences, inferences remain qualitatively unchanged.

Another type of endogeneity that could affect our results is simultaneity bias, as firm value (our dependent variable) is a potential determinant of ownership concentration levels (Donnelly and Lynch, 2002). Simultaneity bias is sometimes addressed through an instrumental variable approach. One such instrumental variable technique is generalised method of moments (GMM). GMM requires no formal identification of instruments but uses a dynamic estimation approach to include lags of the dependent variable as instruments. We therefore reperform earlier regressions using GMM ^[14]. For brevity, only the interaction terms of the various investigations are reported in Table 8. Although we note some differences across the various panels, our broader inferences remain qualitatively unchanged. In brief, high actual ownership concentration is associated with lower earnings value-relevance, while this is also true for both low and high abnormal ownership concentration. Findings for absolute abnormal ownership concentration are stronger than those reported earlier. We omit results for subsamples of firms reporting negative earnings from Table 8, as the number of instruments render these estimations infeasible in many instances.

Although formal tests show that most specifications in Table 8 are not significantly affected by endogeneity, the tests are inconclusive in limited instances. However, endogeneity tests frequently produce unreliable results (Gow and Ding, 2024), which include both false positives and false negatives. Furthermore, improving the GMM specifications would require the identification of an explicit exogeneous instrument. This is highly unlikely when the dependent variable is firm value (Hentschel and Kothari, 2001; Larcker and Rusticus, 2007; Cahan et al., 2009; Gow and Ding, 2024). In addition, lagged variables are frequently invalid instruments outside of the dynamic estimation approach of GMM (Larcker and Rusticus, 2010; Chen et al., 2023). Therefore, given the lack of plausible theory to identify explicit exogeneous instruments for further GMM analysis, we do not extend the use of an instrumental variable approach further ^[15].

However, several other research design elements reduce simultaneity concerns around ownership concentration. Firstly, we use ranked variables throughout. Ranked variables control for simultaneity bias, as crude cut-offs capture the level of a variable, but not the firm-induced variations (simultaneity) around that level (Hentschel and Kothari, 2001; Cahan et al., 2009) ^[16]. Secondly, abnormal ownership concentration reflects the error term which, by construction, represents the ownership concentration that is not explained by the firm characteristics of model (3). By extension, abnormal ownership concentration is independent of these firm characteristics and their association with market value of equity. Panel A of Table 4 supports this conclusion, showing that firm characteristics are randomly distributed when stratifying based on abnormal ownership concentration. Therefore, although no empirical paper can completely rule out endogeneity concerns (Chenhall and Moers, 2007; Larcker and Rusticus, 2007), it appears unlikely that endogeneity significantly impacts our inferences.

Our sample period excludes distress periods as these confound value-relevance results (Kane et al., 2015). However, prior research shows that distress periods can often provide additional insights about longitudinal and cross-sectional variation in the role of ownership concentration (Mitton, 2002; da Cunha and Bortolon, 2016; Rao et al., 2022). It is also possible that distress at firm level could induce cross-sectional variation in the role of ownership concentration. Therefore, different levels of distress, both at macroeconomic level and firm level, could cause cross-sectional variations in inferences around abnormal ownership concentration. However, these and other research questions we leave for future research.

There is a lack of consensus in existing literature on the association between ownership concentration and the value-relevance of accounting information, with evidence that both concentrated ownership (Ghosh and Moon, 2010) and dispersed ownership (Donnelly and Lynch, 2002) are associated with lower value-relevance. We show that high ownership concentration is associated with weakened earnings value-relevance, particularly in the case of high-quality earnings. However, the key contribution of this paper is the evidence that ownership concentration that deviates abnormally from expected levels is associated with increased value-relevance of book values and lower value-relevance of high-quality earnings. In other words, as the theoretical model of Jensen and Meckling (1976) predicts, any level of ownership concentration can detract from the value-relevance of high-quality earnings if it deviates from what firm characteristics dictate that it should be.

Prior research findings on the relationship between ownership concentration and reporting quality are also inconsistent (Dou et al., 2018; Saha et al., 2019). Our findings show that the moderation effect of abnormal ownership concentration differs between earnings and book values. Combined with our findings for high-quality earnings, we therefore conclude that earnings quality is the main channel through which abnormal ownership concentration moderates the pricing of accounting information. Therefore, our study offers a unifying theory and evidence that abnormal ownership structure is associated with lower value-relevance of high-quality earnings and higher value-relevance of book values of equity.

Our research setting in South Africa reflects strong protection of minority shareholders (World Bank, n.d.) in a deep market represented by a relatively low number of firms. While ownership concentration findings can differ significantly between countries, especially emerging markets (Wang and Shailer, 2015; Iwasaki and Mizobata, 2020), our findings could have implications for many markets across the globe with similar institutional characteristics. Our findings suggest, for example, that abnormal ownership concentration could hamper the ability of higher financial reporting quality to improve capital market outcomes. Investors and preparers that aim to maximise firm value will therefore be interested in our findings, which show that optimising ownership structure deserves attention equal to increasing financial reporting quality.

Thanks to Elmar Venter, Phillip de Jager, the associate editor and two anonymous reviewers for helpful comments and suggestions.