Do Boards Shape REIT Performance? Evidence from the South African REIT Sector

Abstract

We examine whether board activity (B_ACTIV), board size (B_SIZE), board independence (BIND), and board tenure (BOARD_TEN) are associated with the performance of South African real estate investment trusts (REITs) over the period 2013 to 2025. The REIT framework provides a rigorous setting to evaluate corporate governance theory, as statutory distribution mandates constrain payout discretion and contracted-income business models limit managerial opportunism, suggesting that governance effects concentrate within specific performance channels. We estimate dynamic panel models using a two-step system GMM framework with collapsed instruments, year fixed effects, Windmeijer-corrected standard errors, and firm-level controls for firm size (SIZE), leverage (LEV), and asset growth (GROWTH) to address endogeneity, unobserved heterogeneity, and performance persistence. We evaluate robustness through an endogenous-regressor specification, a bootstrap bias-corrected LSDVC estimator, and outlier-adjusted estimations. The sample comprises 30 JSE-listed REITs. We evaluate performance across funds from operations per share (FFO_PS), dividend yield (DIV_YIELD), return on assets (ROA), return on equity (ROE), return on invested capital (ROIC), and earnings per share (EPS). Our findings reveal that B_SIZE exhibits a statistically significant negative association with accounting profitability, where each additional director corresponds to a 1.0 percentage point reduction in ROE and a 0.32 percentage point reduction in ROA. The ROE effect remains robust across every identification strategy, including specifications treating board composition as endogenous and estimations winsorizing the dependent variables. Because firm SIZE remains statistically insignificant while LEV and GROWTH display their expected theoretical signs, the B_SIZE effect is isolated from firm scale. BIND demonstrates a directionally positive but specification-sensitive association with returns and payouts, whereas BOARD_TEN shows no robust association with any performance metric, and B_ACTIV effects attenuate once endogeneity is addressed. Overall, governance effects concentrate in operating efficiency and payout measures while remaining absent from per-share metrics, reflecting the precise channels through which boards exercise authority. Our findings caution against board expansion in this sector, highlight board scale as a transparent governance screen for investors, and demonstrate that meeting frequency and tenure benchmarks offer no reliable performance signal.

1. Introduction

Board structure sits at the centre of corporate governance because it determines how effectively a firm is monitored, challenged, and guided. In listed property companies, this matters in a very practical way because boards approve strategy, capital allocation, acquisitions and disposals, debt and hedging decisions, related-party transactions, and distribution policies, all of which feed directly into cash flows and investor confidence (Jensen & Meckling, 1976; Pfeffer & Salancik, 1978). When board design is weak, agency costs rise, disclosure credibility can suffer, and market trust can unwind quickly (Ntim et al., 2019).

Within the South African REIT sector, the composition and structure of boards carry significant practical implications. Historical events have demonstrated that governance uncertainties can prompt pronounced market responses, evidenced by notable share-price declines among leading listed property companies and broader index volatility during periods of increased scrutiny (SAREIT, 2023; JSE, 2018). Capital is highly mobile, funding environments change rapidly, and governance signals shape both company valuation and access to financing. For boards and executive management teams, the relevant issue is not whether governance matters, but which particular board attributes correlate with improved outcomes in a sector where distributions, stable cash earnings, and prudent financial management are the considerations that investors price.

The question of whether board structure relates to firm performance has been studied extensively across markets and industries, and we do not claim the question itself is new. What the REIT setting adds is identification of the channels. Three institutional features of REITs generate predictions that differ from those for general corporate studies and that a general-corporate sample cannot test. For instance, the statutory distribution requirement places a floor under payouts but leaves discretion above the floor, so the payout ratio is precisely the margin on which board monitoring of cash retention should be visible (La Porta et al., 2000). Moreover, the business model is comparatively transparent: income derives from lease contracts on visible physical assets, which compresses the informational advantage of management and, with it, both the monitoring burden and the advisory value of additional directors. Furthermore, income persistence is high because rental escalations are contracted years ahead, which makes the dynamic specification a substantive matter rather than a technical nicety. Together, these features predict that governance effects should surface in efficiency measures (returns on assets and equity) and in the payout channel and should be weak or absent in per-share earnings constructs that are one step removed from board decisions. Testing that channel-specific prediction, rather than re-estimating a generic governance–performance association, is the economic contribution of this paper.

The South African context further strengthens the empirical relevance of this investigation. The establishment of the post-2013 SA REIT regulatory framework provided the sector with a common institutional baseline, while the study period encompasses several material market events, including the 2018 listed-property sell-off, the COVID-19 pandemic, and a complete interest-rate cycle. These episodes provide a particularly informative setting in which governance signals may have had substantial implications for market valuation and investor confidence (SAREIT, 2023). Existing South African evidence also indicates that governance quality, board oversight, and board composition remain salient determinants of REIT stability and operating performance within the JSE-listed property sector (Mofokeng, 2024; Kitulazzi et al., 2025). In addition, the broader REIT literature demonstrates that sectoral returns and profitability are highly exposed to macroeconomic conditions and investor-sentiment shocks (Lin et al., 2009; Bossman et al., 2022; Mensi et al., 2023), a concern addressed in the present empirical design through the inclusion of comprehensive time effects.

Consequently, this paper investigates the relationship between board structure and performance among REITs listed on the JSE. The research evaluates whether board activity (B_ACTIV), board size (B_SIZE), board independence (BIND), and board tenure (BOARD_TEN) are associated with funds from operations per share (FFO_PS), dividend yield (DIV_YIELD), ROA, ROE, ROIC, and EPS across 30 REITs over the period 2013 to 2025. To address pervasive endogeneity concerns, including reverse causality from firm performance to board composition, unobserved firm heterogeneity, and dynamic performance persistence, the analysis employs two-step system GMM (Blundell & Bond, 1998) with collapsed instruments (Roodman, 2009), year fixed effects, and Windmeijer-corrected standard errors (Windmeijer, 2005). This methodology is supported by specifications treating board variables as endogenous and by a bias-corrected LSDVC estimator (Kiviet, 1995; Bruno, 2005) as an independent robustness check.

This paper contributes to the literature across three fronts. Firstly, it provides channel-specific evidence by evaluating board attributes jointly against the performance measures that real estate investment trust boards directly govern, with results aligning with institutional predictions by demonstrating effects in operating efficiency and payouts while showing silence in per-share metrics. Secondly, it demonstrates within a single dataset how much of the apparent governance–performance relationship in a small panel represents a specification artifact, showing that three findings generated under difference GMM with proliferated instruments and omitted time effects do not survive a corrected design, offering a cautionary insight for the emerging-market governance literature. Thirdly, the surviving empirical findings translate into actionable guidance, demonstrating that board scale serves as an observable governance screen for investors and a restraint signal for nomination committees, whereas meeting frequency and tenure benchmarks provide no reliable performance signal in this sector (Ntim & Osei, 2011; Deloitte Southern Africa, 2020; IoDSA, 2025).

2. Problem Statement

Board structure is widely recognised as a central mechanism through which corporate governance may influence firm-level outcomes. However, within the South African REIT sector, empirical evidence capable of informing effective board design remains limited and fragmented. Recent international studies continue to indicate that board activity, board size, board independence, and board tenure are associated with organisational performance, although the direction and magnitude of these relationships vary across institutional contexts, governance systems, and performance measures (Petrova, 2023; Hasan et al., 2024; Noguera, 2024; Zureigat et al., 2024). The governance literature further suggests that the effectiveness of board structures is contingent on sector-specific characteristics and the strategic demands imposed on boards, particularly in capital-intensive industries such as REITs (Kitulazzi et al., 2025).

The gap addressed in this paper is not solely geographic, as the absence of a country-specific case would provide only a limited basis for scholarly motivation. Rather, the central gap concerns the extent to which theoretical predictions regarding board structure are contingent upon institutional features that are especially pronounced in the REIT sector, and whether these conditional predictions hold in a market characterised by a relatively recent, concentrated, and recently stress-tested REIT regime. If larger boards generate net coordination costs in contexts where managerial discretion is already institutionally constrained, such effects should be observable in REIT efficiency measures. Similarly, if independent directors influence performance through the cash-retention margin, their effect should be reflected in payouts above the statutory distribution floor. If board meeting frequency is predominantly reactive rather than causal, any apparent performance effects should dissipate once common shocks are adequately controlled. Each of these propositions constitutes a falsifiable, mechanism-level claim. The South African REIT panel over the 2013–2025 period, which encompasses a governance scandal, the COVID-19 pandemic, and a full interest-rate cycle, provides an appropriate empirical setting in which to test these claims.

For REITs, these issues have immediate practical significance. REIT boards operate within an environment characterised by recurring earnings expectations, high distribution obligations, debt-refinancing pressures, sensitivity to property valuations, and substantial dependence on investor confidence. Governance failures may therefore have direct consequences for financing conditions, market valuations, and the sustainability of distributions (Mofokeng, 2024; Noguera, 2024). Although governance frameworks such as the IoDSA King V Code emphasise board independence, effective oversight, and balanced board composition, policymakers, regulators, and investors continue to face a shortage of REIT-specific empirical evidence linking individual board characteristics to sector-relevant performance indicators. Existing South African studies largely focus on general listed companies or broader governance themes rather than the distinctive institutional and operational features of REITs (Viviers et al., 2023; Moloi et al., 2024; Kitulazzi et al., 2025). This paper addresses this limitation by systematically examining the association between board structure and multiple performance indicators for JSE-listed REITs over the 2013–2025 period, using an estimation design intended to withstand the endogeneity and small-sample concerns commonly associated with dynamic governance panels (Wintoki et al., 2012; Roodman, 2009; Hill et al., 2021).

3. Objectives and Hypothesis

The primary objective of this study is to examine the relationship between board structure and the financial and operating performance of JSE-listed REITs. Performance is assessed using both REIT-specific indicators that are central to industry practice, namely funds from operations per share and dividend yield, and conventional accounting-based measures, including ROA, ROE, ROIC, and EPS. This dual measurement approach reflects the manner in which performance is evaluated by boards, investors, lenders, and analysts within the listed property market (Feng et al., 2021; Ntuli & Akinsomi, 2017). It also enables the channel-specific predictions developed above to be tested empirically: if governance effects operate primarily through efficiency and payout channels, the estimated coefficients should vary systematically across these two categories of performance measures.

Instead of treating board characteristics as isolated mechanisms, this paper adopts an integrated view of board structure. Agency theory suggests that more active and independent boards enhance monitoring and reduce agency costs. Stewardship theory highlights the potential value of experience, continuity, and collective expertise, implying that board size and tenure may support performance up to a point. Resource dependence theory emphasises the role of boards in providing access to external resources, legitimacy, and market credibility, which matters in capital-intensive sectors such as listed property (Jensen & Meckling, 1976; Donaldson & Davis, 1991; Pfeffer & Salancik, 1978). Synthesising these perspectives, the study tests the proposition that board structure is systematically related to REIT performance, with the direction and strength of the relationship differing across board dimensions and performance measures. In particular, the analysis examines whether board activity and independence are associated with improved performance, whether board size reflects a trade-off between resource availability and coordination costs, and whether extended tenure enhances stability or weakens oversight through entrenchment (Hermalin & Weisbach, 2001; Guest, 2009).

The empirical analysis is guided by the following general hypothesis:

H₀. 

Board structure has no significant impact on the performance of South African REITs.

H₁. 

Board structure is significantly associated with the performance of South African REITs.

This hypothesis is evaluated by examining the individual and joint effects of board activity, board size, board independence, and board tenure on multiple performance measures within a dynamic panel framework.

4. Literature Review

4.1. Theoretical Framework

Analysing the relationship between board structure and firm performance requires engagement with three foundational theoretical perspectives: agency theory, stewardship theory, and resource dependence theory. Each framework offers a distinct explanation of how governance arrangements may influence organisational effectiveness. Rather than treating these perspectives as competing theoretical labels, this study draws on the specific mechanisms predicted by each framework and considers how these mechanisms are conditioned by the institutional characteristics of REITs. This approach is appropriate because it is the underlying mechanisms, rather than the theoretical labels themselves, that generate empirically testable differences across the six performance measures examined in this study.

4.1.1. Agency Theory and REIT Governance

Agency theory was first conceptualised by Berle and Means (1932) and formalised by Jensen and Meckling (1976) to address the conflict between principals (shareholders) and agents (directors and managers). Agents may pursue self-serving objectives, generating costs for shareholders unless proper oversight mechanisms are established. Independent directors and active boards mitigate these conflicts by enhancing monitoring and reducing managerial opportunism (Tosi & Gomez-Mejia, 1989). For South African REITs, this oversight is particularly important because investors are diversified and rely on professional management for asset allocation and valuation, and the 2018 Viceroy Research governance episode, in which allegations of malpractice triggered large share-price declines across leading REITs, demonstrated how quickly governance doubts translate into valuation losses (JSE, 2018). The REIT institutional design, though, changes where agency effects should appear. The distribution requirement removes much of the free-cash-flow problem that agency theory centres on (Jensen, 1993), but it leaves discretion over the payout ratio above the statutory floor and over the efficiency with which the asset base is managed. Agency theory in a REIT therefore predicts that monitoring quality, proxied by independence, should show up in payouts (La Porta et al., 2000), and that monitoring costs, proxied by board scale, should show up in returns on assets and equity.

4.1.2. Stewardship Theory

Stewardship theory proposes that managers and directors act as responsible stewards of shareholder wealth rather than opportunistic agents. Donaldson and Davis (1991) argue that executives are intrinsically motivated to achieve organisational success, so governance mechanisms should not overemphasise monitoring but instead empower directors with autonomy and discretion. Board size and tenure are particularly relevant: larger boards can enhance decision-making by incorporating diverse expertise (Dalton et al., 1999), while longer tenure allows directors to develop institutional knowledge and promote strategic stability (Jones, 1995). Within South African REITs, where real estate markets are cyclical and regulatory environments complex, continuity and accumulated expertise are plausibly valuable. Stewardship theory therefore predicts positive coefficients on board size and tenure, and it is the theory our results place under the greatest pressure.

4.1.3. Resource Dependence Theory

Resource dependence theory, as developed by Pfeffer and Salancik (1978), conceptualises boards as strategic mechanisms through which firms manage dependence on their external environment. From this perspective, boards provide access to networks, legitimacy, external expertise, and critical information, thereby reducing organisational exposure to uncertainty in resource and capital markets (Haniffa & Hudaib, 2006). Larger and more diverse boards may expand these external linkages, while independent directors may enhance organisational credibility and contribute specialised knowledge (Dalton et al., 1999). In the South African REIT market, these functions are particularly relevant because listed property companies depend heavily on capital raising, investor confidence, and regulatory legitimacy. However, the REIT setting also conditions the theoretical prediction. Because REIT operations are comparatively transparent and asset-backed, the informational and advisory premium associated with larger boards may be more limited than in more opaque corporate structures. Consequently, resource dependence theory’s positive prediction regarding board size is likely to be weakest in settings where the coordination costs of larger boards remain present while the marginal advisory benefits are reduced. Whether resource benefits outweigh coordination costs is therefore an empirical question addressed through the board-size tests in this study.

4.2. Empirical Literature

4.2.1. Board Activity and REIT Performance

Board activity, typically operationalised as the frequency of board meetings, reflects the extent of director engagement in monitoring, strategic decision-making, and financial oversight. Agency theory suggests that more frequent meetings may strengthen monitoring capacity, while resource dependence theory implies that active boards may enhance access to expertise and external knowledge (Jensen & Meckling, 1976; Dalton et al., 1999). However, the empirical evidence remains mixed, and this lack of consensus is analytically informative. Nguyen and Huynh (2023) report that meeting frequency positively affects financial performance among Vietnamese real estate and construction firms, while Noguera (2024) finds that stronger board and committee engagement improves US REIT performance through enhanced investment monitoring. Conversely, Venkatesh et al. (2025) identify a nonlinear relationship between board busyness and firm performance, Cao and Nguyen (2025) show that busy boards are associated with higher earnings management among Vietnamese real estate firms, and Alomair and Al Naim (2025) find that governance busyness may impair monitoring effectiveness.

The central unresolved issue in this literature concerns causal direction. Vafeas (1999) demonstrates that boards tend to meet more frequently following periods of weak performance; consequently, a negative association between meeting frequency and performance may arise without implying a causal effect, while a positive association may reflect the selection of already well-governed firms into higher levels of board engagement. This concern is compounded by the fact that relatively few studies adequately control for common time shocks, despite the likelihood that crisis periods simultaneously increase meeting frequency across the sector and depress firm performance. The empirical design adopted in this study addresses this issue through comprehensive year effects. Consistent with the reactive-meetings interpretation, the meeting-performance association in the sample appears only when time effects are omitted and disappears once they are included. In the South African context, the IoDSA (2025) King V Code recommends regular meetings and sufficient director engagement, and prior studies, including Mofokeng (2024), Moloi et al. (2024), and Zvinowanda and Odendaal (2025), report positive associations between meeting frequency and performance. Nevertheless, REIT-specific evidence remains limited.

4.2.2. Board Size and REIT Performance

Board size remains one of the most contested mechanisms in the corporate governance literature. The theoretical tension is substantive: as board size increases, monitoring capacity and access to strategic resources may improve (Jensen, 1993; Guest, 2009), yet larger boards may also generate higher coordination costs, slower decision-making processes, and greater opportunities for director free-riding (Lipton & Lorsch, 1992; Hermalin & Weisbach, 2001). Much of the classic firm-level evidence supports the cost-based interpretation. Yermack (1996) documents an inverse relationship between board size and firm value among large US corporations, Eisenberg et al. (1998) report similar evidence for small firms, and Guest (2009) reaches comparable conclusions using a large UK panel. By contrast, emerging-market studies more frequently identify positive associations between board size and firm outcomes, including Hasan et al. (2024) for Bangladeshi financial institutions, Zureigat et al. (2024) for Jordanian firms, and Satapathy et al. (2023) for Indian firms. Petrova (2023) further shows that the board-size-performance relationship varies substantially across national contexts, while K. M. Khan and Mahmood (2023) argue that leaner boards may enhance monitoring effectiveness.

This divergence in findings should not be interpreted as empirical noise; rather, it appears to reflect variation in institutional environments. In settings where firms are opaque and external resources are scarce, the advisory, reputational, and network benefits associated with larger boards may plausibly outweigh their coordination costs. Conversely, where operations are relatively transparent and cash flows are largely contracted, the marginal benefits of board expansion may diminish while the costs of coordination remain. REITs are situated closer to the transparent end of this spectrum, which underscores the need for sector-specific evidence rather than reliance on findings from general corporate samples. In South Africa, the Companies Act No. 71 of 2008 requires public companies to have at least three directors. Ajayi (2022) finds no significant relationship between board size and REIT performance volatility, whereas Scholtz and Kieviet (2017) report positive associations between board size, Tobin’s Q, and ROA among general listed companies. Mofokeng (2024) further emphasises the importance of board oversight quality in JSE-listed REITs. The present study therefore examines whether the cost-based interpretation of board size dominates in the REIT context, as suggested by the institutional argument developed above.

4.2.3. Board Independence and REIT Performance

Board independence is widely regarded as a central corporate governance mechanism through which accountability, monitoring effectiveness, and strategic oversight may be strengthened. In practice, independence is primarily achieved through the appointment of independent non-executive directors, who are expected to contribute specialised expertise, external networks, and objective judgement to board deliberations (Hoitash, 2010; Fuzi et al., 2016). Agency theory emphasises the monitoring role of independent directors in mitigating managerial opportunism and reducing agency conflicts (Mohamad & Sahul Hamid, 2022). By contrast, stewardship theory conceptualises directors as strategic partners who support organisational continuity and long-term value creation, while resource dependence theory stresses the contribution of independent directors to external legitimacy, knowledge acquisition, and access to critical resources (Donaldson & Davis, 1991; Pfeffer & Salancik, 1978).

Empirical evidence on the independence-performance relationship remains inconclusive, although the pattern of findings again suggests the importance of identifying the relevant performance channel. Yeon et al. (2021) find that outside directors positively moderate the relationship between environmental management and performance in lodging REITs, while Mariani et al. (2018) associate board independence and diversity with improved financial performance among European Green REITs. David et al. (2021) report positive effects for Malaysian REITs but insignificant effects for Hong Kong REITs, and Noguera (2024) shows that committee composition and director expertise are relevant to US REIT performance. In contrast, M. J. Khan et al. (2024) suggest that excessive independence may impair performance in emerging markets where directors maintain close ties to dominant shareholders. The most theoretically precise evidence concerns payout policy. La Porta et al. (2000) formulate the outcome model of dividends, under which stronger governance compels firms to distribute cash to shareholders, while Viviers et al. (2023) document a comparable association between board independence and dividend distributions in South Africa. In the REIT context, where the statutory distribution requirement establishes a payout floor but leaves discretion above that threshold, the payout ratio constitutes the margin on which independent monitoring should be most directly observable. This prediction is examined directly in the present study. Kitulazzi et al. (2025) further report that, among JSE-listed real estate firms, independent directors are negatively associated with short-term financial performance but positively associated with sustainability outcomes, which is consistent with the view that independent boards may prioritise risk management and compliance over immediate accounting returns. The IoDSA (2025) King V Code similarly recommends a board structure comprising a majority of non-executive directors, most of whom should be independent, together with an independent non-executive chairperson and a fully independent audit committee.

4.2.4. Board Tenure and REIT Performance

Board tenure remains theoretically contested because it may simultaneously enhance board capability and increase the risk of director entrenchment. From a resource dependence perspective, longer-serving directors accumulate firm-specific knowledge, property-market expertise, regulatory familiarity, and stakeholder relationships, all of which are particularly relevant in REITs, where investment decisions require an understanding of property cycles, asset valuation, tenant risk, and capital-market conditions. Noguera (2024) finds that longer-tenured directors with real estate expertise, particularly those serving on investment committees, are associated with stronger performance among US equity REITs, while Morri et al. (2023) show that shareholder-oriented governance improves real estate firm value. Conversely, agency theory cautions that excessive tenure may weaken independence and reduce critical oversight. Haider et al. (2025) argue that shorter tenure may reduce groupthink, Zeng et al. (2025) find that longer tenure increases misconduct risk, and Cashman et al. (2018) document both decision-making benefits and monitoring costs associated with extended board service. The IoDSA (2025) King V Code discourages rigid term limits and emphasises the value of retained expertise, whereas Shill and Strand (2021) caution against complacency among long-serving board members. Given these offsetting theoretical predictions and the compressed tenure range typical of a relatively young REIT regime, the null hypothesis warrants serious consideration for this variable rather than being interpreted merely as a failure to detect an expected effect.

4.2.5. Board Structure Among the Determinants of REIT Performance

A final strand of the literature situates governance within the broader set of determinants of REIT performance. Recent evidence indicates that REIT returns and profitability are strongly shaped by macroeconomic and market-wide forces, including investor sentiment (Lin et al., 2009; Chiang & Tsai, 2023), asymmetric pandemic-related shocks (Bossman et al., 2022; Tong et al., 2024), spillovers and connectedness across global REIT markets driven by uncertainty and macroeconomic shocks (Mensi et al., 2023; Hadad et al., 2024; Hadad & Choi, 2026), regional dependence in REIT pricing (Motegi & Iitsuka, 2023), and the information content of disclosure tone for institutional investment in the South African REIT market specifically (Carstens & Freybote, 2019). This literature has two implications for the present governance analysis. First, common time-varying forces constitute first-order determinants of REIT performance; therefore, firm-level governance estimates that omit time effects risk attributing sector-wide shocks to board characteristics. The empirical specification adopted in this study accordingly includes a full set of year effects, which absorb, within the single-country panel, the common component of interest-rate movements, policy uncertainty, sentiment shifts, and the COVID-19 pandemic. Second, governance coefficients should be interpreted as within-firm effects that remain after these dominant common forces have been removed. This constitutes a deliberately conservative identification strategy, as it reduces the likelihood of detecting governance effects and thereby strengthens the interpretation of those effects that persist. A natural extension would test this absorption claim directly by entering a standard economic policy uncertainty or geopolitical risk index into the estimation. In a single-country panel, however, any such country-level index varies only over time and is therefore perfectly collinear with the full set of year dummies; implementing the test would require replacing the nonparametric absorption of all common shocks with a parametric proxy for some of them, a strictly weaker specification. The year effects retained here already absorb whatever common component uncertainty, sentiment, and risk indices carry, while interactions of firm-level exposures with such indices, which survive the collinearity constraint, are identified as an avenue for future work in the Conclusion.

5. Methodology

5.1. Sample and Data

We adopt a quantitative research methodology to examine the empirical relationship between board structure and REIT performance. The analysis focuses on South African REITs listed on the JSE from 2013 to 2025, a timeframe that encompasses the entirety of the post-South African REIT regulatory framework from its inception. The final dataset comprises 30 REITs, effectively representing the entire target population of the JSE REIT sector rather than a sample drawn from it. Financial and corporate governance metrics were extracted from Iress^® and Morningstar™, with inclusion restricted to entities possessing adequate governance and financial data throughout the observation period.

The resulting panel dataset consists of 360 firm-year observations. Variable-level data availability yields between 320 and 326 usable observations for the core performance and board structure measures, while balance sheet metrics for firm-level controls are available for 309 observations regarding total assets and 300 observations regarding total debt. Constructing the lagged dependent variable and the asset growth control variable, in combination with the listwise deletion of incomplete firm-years, reduces the final estimation samples to between 233 and 240 firm-year observations across the six statistical models. These three distinct metrics are articulated explicitly to differentiate the total assembled panel size, individual variable coverage, and the effective estimation sample size, thereby preventing misinterpretation of the dataset composition.

5.2. Empirical Model

To examine the relationship between board structure and REIT performance, we estimate the following dynamic panel model for each performance measure:

$$\begin{array}{c}{y}_{it}=\alpha +\rho y_{i,t-1}+{\beta }_1 BACTI{V}_{it}+{\beta }_2 BSIZ{E}_{it}+{\beta }_3 BIN{D}_{it}+{\beta }_4 BOARDTE{N}_{it}+{\gamma }_1 SIZ{E}_{it}+{\gamma }_2 LE{V}_{it}+{\gamma }_3 GROWT{H}_{it}+{\lambda }_t+{\eta }_i+{\epsilon }_{it}\end{array}$$

(1)

where $y_{it}$ is the performance measure of firm $i$ in year $t$ (one of FFO_PS, DIV_YIELD, ROA, ROE, ROIC, EPS, each estimated in a separate equation), $y_{i,t-1}$ is its first lag, ${\lambda }_t$ is a full set of year dummies, ${\eta }_i$ is an unobserved firm-specific effect, and ${\epsilon }_{it}$ is an idiosyncratic error assumed serially uncorrelated. The vector of firm-level controls comprises firm size ($\mathrm{S}\mathrm{I}\mathrm{Z}\mathrm{E}$, the natural logarithm of total assets), leverage ($\mathrm{L}\mathrm{E}\mathrm{V}$, the ratio of total debt to total assets), and growth ($\mathrm{G}\mathrm{R}\mathrm{O}\mathrm{W}\mathrm{T}\mathrm{H}$, the annual change in the logarithm of total assets), the three time-varying firm characteristics most consistently identified as determinants of REIT returns and risk. Their inclusion addresses the concern that board composition covaries with firm scale, capital structure, or expansion activity, and allows the board size coefficient to be identified separately from firm size itself. The coefficient $\rho$ measures performance persistence. The year dummies are estimated in every specification, not merely defined: they absorb the common component of the 2018 listed-property selloff, the COVID-19 disruption, the interest-rate cycle, and sector-wide sentiment and valuation movements documented in the REIT literature (Bossman et al., 2022; Mensi et al., 2023). The firm effect ${\eta }_i$ absorbs time-invariant heterogeneity such as property sector focus, listing history, and persistent scale differences between REITs.

The lagged dependent variable belongs in the model on theoretical and statistical grounds. REIT income derives from multi-year lease contracts with contracted escalations, so performance persistence is a structural feature of the sector, and omitting the lag would push that persistence into the error term and bias the board coefficients. The estimated persistence coefficients of up to 0.74 in the controlled models, and 0.83 without controls, confirm the dynamic term is not ignorable.

5.3. The Endogeneity Problem

Three econometric problems make ordinary estimation of Equation (1) unreliable, each documented in the governance literature (Wintoki et al., 2012; Hill et al., 2021).

The first is unobserved heterogeneity: firm-specific factors such as managerial ability influence both board composition and performance, so if these factors sit in ${\eta }_i$ and correlate with the regressors, pooled OLS estimates are inconsistent.

The second is simultaneity and dynamic endogeneity. Board structure is not assigned at random: poorly performing firms may enlarge or refresh their boards, and past performance shapes current composition. Formally, the strict exogeneity condition $E\left[{\epsilon }_{it}\mid X_{i1},\dots ,X_{iT},{\eta }_i\right]=0$ fails because $X_{it}$ responds to past values of $\epsilon$. We take this problem seriously at the level of the board variables themselves: our main specification instruments the lagged dependent variable and enters the board variables as standard instruments, and the Endogenous Board Variables subsection of the Methodology then re-estimates every model treating the four board variables as endogenous, instrumented GMM-style by their own lags, so that no result rests on an exogeneity assumption for board structure.

The third is the Nickell (1981) bias. The within (fixed effects) transformation demeans each variable, so the transformed lagged dependent variable correlates with the transformed error by construction. The bias on $\rho$ is negative and of order $1/T$:

$$\begin{array}{c}\mathrm{p}\mathrm{l}\mathrm{i}\mathrm{m} \left({\widehat{\rho }}_{FE}-\rho \right)\approx -{\displaystyle \frac{1+\rho }{T-1}}\end{array}$$

(2)

With the time dimension of this panel the bias is material, and it contaminates the board coefficients through their correlation with the lag. Pooled OLS biases $\widehat{\rho }$ upward because the lag correlates positively with ${\eta }_i$, while fixed effects biases it downward through Equation (2); consistent estimates should fall between these bounds, a property we use as a specification check on the GMM results.

5.4. Difference GMM

Arellano and Bond (1991) address these problems by first-differencing Equation (1), which eliminates ${\eta }_i$:

$$\begin{array}{c}\Delta y_{it}=\rho \Delta y_{i,t-1}+{\mathsf{\beta }}^{\prime } \Delta {\mathrm{x}}_{it}+\Delta {\lambda }_t+\Delta {\epsilon }_{it}\end{array}$$

(3)

Differencing introduces a new problem: $\Delta y_{i,t-1}$ correlates with $\Delta {\epsilon }_{it}$ through the shared ${\epsilon }_{i,t-1}$ term. The estimator solves this with internal instruments. Provided ${\epsilon }_{it}$ is serially uncorrelated, lags of the dependent variable dated $t-2$ and earlier are uncorrelated with $\Delta {\epsilon }_{it}$, yielding the moment conditions:

$$\begin{array}{c}E\left[y_{i,t-s} \Delta {\epsilon }_{it}\right]=0 \mathrm{f}\mathrm{o}\mathrm{r}s\ge 2,\hspace{0.33em}t=3,\dots ,T\end{array}$$

(4)

5.5. System GMM

Difference GMM performs poorly when the series is persistent, because lagged levels are then weak instruments for first differences (Blundell & Bond, 1998). Several of our performance series are persistent, EPS with an autoregressive coefficient above 0.8, which makes this weakness operative rather than hypothetical and is the principal reason we do not rely on difference GMM as the main estimator. Blundell and Bond (1998) augment the differenced equation with the original levels equation, instrumenting the levels with lagged first differences, adding the moment conditions:

$$\begin{array}{c}E\left[\Delta y_{i,t-1} \left({\eta }_i+{\epsilon }_{it}\right)\right]=0\end{array}$$

(5)

The validity of Equation (5) requires a stationarity condition on initial observations, namely that deviations of the initial values from long-run firm means are uncorrelated with the firm effects:

$$\begin{array}{c}E\left[\left(y_{i1}-{\displaystyle \frac{\alpha +{\eta }_i}{1-\rho }}\right){\eta }_i\right]=0\end{array}$$

(6)

This condition is plausible in our setting because the sample opens at the start of the SA REIT regime, a common institutional starting point, rather than at heterogeneous points in individual firm life cycles. We also test the level-equation instruments directly with the difference-in-Hansen statistic, reported with the results. The combined system GMM estimator stacks the differenced and level equations and applies the moment conditions in Equations (4) and (5) jointly. In the main specification the board variables enter the instrument matrix as standard (IV-style) instruments and the year dummies as exogenous instruments; the endogenous-regressor variant is described in the Endogenous Board Variables subsection.

5.6. Two-Step Estimation and the Windmeijer Correction

Stacking the moment conditions as $E\left[{{\mathrm{Z}}_i}^{\prime }{\mathrm{u}}_i\right]=0$, where ${\mathrm{Z}}_i$ is the instrument matrix and ${\mathrm{u}}_i$ the stacked errors for firm $i$, the GMM estimator minimises:

$$\begin{array}{c}\widehat{\mathsf{\theta }}=\arg \underset{\mathsf{\theta }}{\mathrm{m}\mathrm{i}\mathrm{n}}\left({\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^N}{\mathrm{u}}_i(\mathsf{\theta }{)}^{\prime }{\mathrm{Z}}_i\right)W\left({\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^N}{{\mathrm{Z}}_i}^{\prime }{\mathrm{u}}_i\left(\mathsf{\theta }\right)\right)\end{array}$$

(7)

The one-step estimator uses a weight matrix assuming homoskedastic errors; the two-step estimator replaces it with the efficient weight matrix built from one-step residuals:

$$\begin{array}{c}{\mathrm{W}}_2={\left({\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^N}{{\mathrm{Z}}_i}^{\prime }{\widehat{\mathrm{u}}}_i{{\widehat{\mathrm{u}}}_i}^{\prime }{\mathrm{Z}}_i\right)}^{-1}\end{array}$$

(8)

Two-step standard errors are severely downward biased in finite samples because ${\mathrm{W}}_2$ is itself estimated. All reported standard errors therefore apply the Windmeijer (2005) finite-sample correction, which adjusts the variance estimate for the sampling variability of the weight matrix. Without it, two-step inference in a panel of 30 firms would be unusable.

5.7. Instrument Set and the Proliferation Problem

The GMM-style instruments use lags 2 through 4 of the dependent variables. Restricting lag depth is the first defense against instrument proliferation: the number of moment conditions in an uncollapsed instrument matrix grows quadratically in $T$, and too many instruments overfit the endogenous variables, bias coefficients toward their non-instrumented values, and weaken the Hansen test to the point of meaninglessness (Roodman, 2009), the symptom being Hansen p-values near one.

Estimations on this panel confirmed the risk concretely. An earlier difference GMM specification produced 59 instruments against 30 firms with Hansen p-values between 0.9975 and 0.9995, and an uncollapsed system GMM specification with year dummies produced 52 instruments with Hansen p-values between 0.898 and 1.000. Following Roodman (2009), the main specification therefore collapses the GMM instrument matrix, replacing the full set of lag-and-period-specific instruments with one instrument per lag distance by summing the moment conditions over $t$:

$$\begin{array}{c}E\left[{\displaystyle \sum _t}{y}_{i,t-s} \Delta {\epsilon }_{it}\right]=0\hspace{1em}\mathrm{f}\mathrm{o}\mathrm{r}\mathrm{e}\mathrm{a}\mathrm{c}\mathrm{h}s\ge 2\end{array}$$

(9)

This reduces the instrument count to 19, comfortably below the number of groups, and restores the discriminating power of the overidentification test: the Hansen p-values move from the uninformative near-unity range into an informative middle range, direct evidence that the test regained power rather than merely passing. The earlier specifications are retained and reported in Appendix A as sensitivity checks, and the original difference GMM estimates are reported there for transparency about how the results change under the corrected design.

5.8. Specification with Endogenous Board Variables

To relax the assumption that board structure is exogenous, we re-estimate every model treating all four board variables as endogenous, instrumenting each GMM-style with its own lags 2 to 3, collapsed, alongside the collapsed lag-2-to-3 instruments for the dependent variable and the firm-level controls as standard instruments, and the year dummies as exogenous instruments. The tighter lag range keeps the instrument count at 29, below the number of groups. Under this specification the identifying assumption for the board coefficients is no longer strict exogeneity but the absence of second-order serial correlation in the idiosyncratic error, the same assumption already required for the lagged dependent variable and testable via the AR(2) statistic. This directly addresses the reverse-causality channel in which past performance shapes current board composition.

5.9. Specification Tests

Three diagnostics accompany every model. The Arellano–Bond serial correlation tests examine the differenced residuals: first-order correlation, AR(1), should be present and negative by construction, while second-order correlation, AR(2), must be absent, because AR(2) in differences implies first-order correlation in levels, which would invalidate the $t-2$ lags as instruments. The statistic for order $m$ is asymptotically standard normal:

$$\begin{array}{c}AR\left(m\right)={\displaystyle \frac{\sum _i\Delta {{\widehat{\epsilon }}_{i,t}}^{\prime } \Delta {\widehat{\epsilon }}_{i,t-m}}{\sqrt{\widehat{Var}\left(\sum _i\Delta {{\widehat{\epsilon }}_{i,t}}^{\prime } \Delta {\widehat{\epsilon }}_{i,t-m}\right)}}}\sim N\left(0,1\right)\end{array}$$

(10)

The Hansen (1982) J test of overidentifying restrictions evaluates joint instrument validity. Under the null that all moment conditions hold:

$$\begin{array}{c}J={\left({\displaystyle \frac{1}{N}}{\displaystyle \sum _i}{{\mathrm{Z}}_i}^{\prime }{\widehat{\mathrm{u}}}_i\right)}^{\prime {\mathrm{W}}_2\left({\displaystyle \frac{1}{N}}\sum _i{{\mathrm{Z}}_i}^{\prime }{\widehat{\mathrm{u}}}_i\right)}\cdot N\hspace{0.33em}\sim \hspace{0.33em}{\chi }_{\left(L-K\right)}^2\end{array}$$

(11)

where $L$ is the instrument count and $K$ the number of estimated parameters. The Hansen statistic is preferred over Sargan because it is robust to heteroskedasticity; rejection signals invalid instruments, while p-values approaching one signal a weakened test. We additionally report the difference-in-Hansen statistics for the system GMM level-equation instruments, computed as the difference between the system and difference GMM Hansen statistics with degrees of freedom equal to the difference in overidentifying restrictions.

The Wald test evaluates the joint significance of the five substantive regressors, the lagged dependent variable and the four board variables, using the Windmeijer-corrected covariance matrix:

$$\begin{array}{c}W={\widehat{\mathrm{b}}}^{\prime {[\widehat{Var}(\widehat{\mathrm{b}})]}^{-1}\widehat{\mathrm{b}}}\hspace{0.33em}\sim \hspace{0.33em}{\chi }_{\left(5\right)}^2\end{array}$$

(12)

5.10. Robustness Estimator: Bias-Corrected LSDV

System GMM was designed for panels with large $N$ and small $T$; with $N=30$ its asymptotic properties are strained, and Monte Carlo evidence shows the bias-corrected least squares dummy variable estimator (LSDVC) can outperform GMM in small-$N$ dynamic panels (Kiviet, 1995; Judson & Owen, 1999; Bruno, 2005). We therefore re-estimate the ROA and ROE models, the two carrying the central findings, with a bootstrap-based LSDVC in the spirit of Everaert and Pozzi (2007).

The procedure is as follows. First, estimate Equation (1) by least squares with full firm and year dummies, obtaining ${\widehat{\mathsf{\theta }}}_{LSDV}$ and residual variance ${\widehat{\sigma }}^2$. Second, treating ${\widehat{\mathsf{\theta }}}_{LSDV}$ as the data-generating process and holding the regressors, firm effects, year effects, and each firm’s actual initial condition fixed, simulate $B=500$ artificial panels by drawing errors ${\epsilon }_{it}^{\left(b\right)}\sim N\left(0,{\widehat{\sigma }}^2\right)$ and generating $y_{it}^{\left(b\right)}$ recursively. Third, re-estimate the LSDV model on each artificial panel. The bootstrap estimate of the Nickell bias and the corrected estimator are:

$$\begin{array}{c}\widehat{bias}={\displaystyle \frac{1}{B}}{\displaystyle \sum _{b=1}^B}{\hat{\mathsf{\theta }}}^{\left(b\right)}-{\widehat{\mathsf{\theta }}}_{LSDV}\end{array}$$

(13)

$$\begin{array}{c}{\widehat{\mathsf{\theta }}}_{LSDVC}={\widehat{\mathsf{\theta }}}_{LSDV}-\widehat{bias}\end{array}$$

(14)

Standard errors are the standard deviations of the bootstrap replications. The estimator corrects the dynamic panel bias exactly where GMM’s small-sample weakness lies but treats the board variables as exogenous; the two estimators thus have complementary strengths, and agreement between them on sign and magnitude constitutes evidence robust to the identification strategy.

5.11. Variable Definition and Measurement

Table 1 below provides a summary of variable definitions and measurements:

Table 1. Variable Definition and Measurement.

Variable(s)	Acronym(s)	Description
Funds from operations per share	FFO_PS	FFO divided by average number of outstanding shares
Dividend yield	DIV_YIELD	Dividend per share divided by price per share
Return on asset	ROA	Net profit divided by total assets
Return on equity	ROE	Net profit divided by average shareholders’ equity
Return on invested capital	ROIC	EBIT divided by the sum of total equity and total debt
Earnings per share	EPS	Net profit less preferred dividends, divided by average shares outstanding
Board Activity	B_ACTIV	Reflects the level of board engagement and oversight through meeting frequency.
Board Size	B_SIZE	Measures the total number of directors serving on the board.
Board Independence	BIND	Measures the extent of independent oversight within the board structure.
Board Tenure	BOARD_TEN	Measures the average years of service of board members.
Firm Size	SIZE	Natural logarithm of total assets
Leverage	LEV	Total debt divided by total assets
Growth	GROWTH	Annual change in the natural logarithm of total assets

5.12. Estimation Summary

All GMM models are estimated by two-step system GMM with Windmeijer-corrected standard errors using the pydynpd implementation of the Arellano–Bond and Blundell–Bond estimators in Python 3.12. The main specification uses collapsed instruments (lags 2 to 4 of the dependent variable), year dummies, firm-level controls, and IV-style board variables (22 instruments). Robustness specifications comprise: the endogenous-board-variable system GMM with controls (29 instruments); the collapsed specification without firm-level controls, reported for comparison with the smaller estimation sample the controls impose (Appendix A Table A1); uncollapsed system GMM with and without year dummies (52 and 42 instruments, respectively, Appendix A); the original two-step difference GMM (59 instruments, Appendix A, for comparison); the LSDVC estimator with controls for ROA and ROE; and outlier checks re-estimating the ROA, ROE and ROIC models with the dependent variable winsorized at the 1st and 99th percentiles, motivated by the non-normality reported in the descriptive statistics. Statistical significance is assessed at the 1, 5, and 10 percent levels throughout.

6. Findings

We applied the above equations to the selected dataset to analyse the descriptive statistics of the dependent (performance measures) and independent variable (activity, size, independence, and tenure), with their results shown in Table 2, Table 3 and Table 4 below.

6.1. Descriptive Statistics, Correlations and VIF

Table 2 presents the descriptive statistics for the six performance measures and four board structure variables under investigation. Funds from operations per share yields a mean value of 1.8712 with a standard deviation of 2.9228, spanning from a minimum of −9.90 to a maximum of 15.20. Dividend yield averages 8.29 percent, whereas return on assets, return on equity, and return on invested capital exhibit mean values of 4.19 percent, 6.98 percent, and 4.93 percent, respectively. Earnings per share displays a mean of 1.9844 and exhibits the highest degree of dispersion within the sample, with a standard deviation of 6.8340, which reflects the underlying earnings heterogeneity among the observed real estate investment trusts. The reported skewness and kurtosis statistics demonstrate that most of the continuous variables deviate from a normal distribution. Specifically, earnings per share, funds from operations per share, dividend yield, and board activity are positively skewed, whereas return on assets, return on equity, return on invested capital, board independence, and average board tenure exhibit negative skewness. Furthermore, return on invested capital displays extreme kurtosis at 297.6, which is driven by a limited number of pronounced negative values, a distributional property subsequently addressed through a winsorization robustness test. Regarding board structure attributes, board size averages 9.36 directors, and board independence records a mean proportion of 58.26 percent. Board activity averages 6.35 meetings per year, while the mean tenure across boards is 6.50 years. The firm-level control variables demonstrate a mean log of total assets of 16.39 with a standard deviation of 1.23, a mean leverage ratio of 36.3 percent of total assets with a standard deviation of 11.6 percentage points, and a mean annual asset growth rate of 8.4 percent. These magnitudes align with the established gearing standards and corporate expansion activities characteristic of the JSE listed property sector during the period under study. Finally, while firm size and leverage maintain approximately symmetric distributions, asset growth exhibits pronounced right-skewness, with a skewness of 5.85 and kurtosis of 61.5, attributable to a small number of substantial acquisition events, most notably the 2015 asset-base expansion of Industrials REIT.

Table 2. Summary of descriptive statistics.

Variable(s)	Mean	Median	Maximum	Minimum	Std. Dev.	Skewness	Kurtosis	Obs.
FFO_PS	1.8712	1.0900	15.2000	−9.9000	2.9228	1.3301	9.1664	323
DIV_YIELD	0.0829	0.0744	0.9250	0.0000	0.0770	5.1551	49.5380	320
ROA	0.0419	0.0508	0.3796	−0.2624	0.0681	−0.9596	7.8361	326
ROE	0.0698	0.0872	0.5840	−0.7493	0.1291	−1.6951	11.5481	323
ROIC	0.0493	0.0574	0.3702	−2.1707	0.1260	−16.8046	297.6221	326
EPS	1.9844	1.1300	44.6300	−37.7700	6.8340	1.1545	17.6534	326
B_ACTIV	6.3528	6.0000	16.0000	1.0000	2.2576	0.7257	4.3576	326
B_SIZE	9.3620	9.0000	16.0000	5.0000	2.2445	0.1125	2.7355	326
BIND	0.5826	0.6000	0.8571	0.0000	0.1443	−0.9613	4.9261	326
BOARD_TEN	6.4991	6.5000	8.2000	4.6000	0.7524	−0.1252	2.3750	326
SIZE	16.3918	16.4156	18.9629	13.1543	1.2317	−0.1025	2.7302	309
LEV	0.3631	0.3693	0.7572	0.0007	0.1159	−0.2339	5.3536	300
GROWTH	0.0840	0.0457	3.0421	−0.4583	0.2621	5.8523	61.4552	278

Notes: FFO_PS = funds from operations per share; DIV_YIELD = dividend yield; ROA = return on assets; ROE = return on equity; ROIC = return on invested capital; EPS = earnings per share; B_ACTIV = number of board meetings; B_SIZE = number of directors on the board; BIND = proportion of independent directors; BOARD_TEN = average board tenure (years); SIZE = natural logarithm of total assets; LEV = total debt divided by total assets; GROWTH = annual change in the natural logarithm of total assets. Sample: 30 JSE-listed REITs, 2013 to 2025.

Table 2. Summary of descriptive statistics.

Variable(s)	Mean	Median	Maximum	Minimum	Std. Dev.	Skewness	Kurtosis	Obs.
FFO_PS	1.8712	1.0900	15.2000	−9.9000	2.9228	1.3301	9.1664	323
DIV_YIELD	0.0829	0.0744	0.9250	0.0000	0.0770	5.1551	49.5380	320
ROA	0.0419	0.0508	0.3796	−0.2624	0.0681	−0.9596	7.8361	326
ROE	0.0698	0.0872	0.5840	−0.7493	0.1291	−1.6951	11.5481	323
ROIC	0.0493	0.0574	0.3702	−2.1707	0.1260	−16.8046	297.6221	326
EPS	1.9844	1.1300	44.6300	−37.7700	6.8340	1.1545	17.6534	326
B_ACTIV	6.3528	6.0000	16.0000	1.0000	2.2576	0.7257	4.3576	326
B_SIZE	9.3620	9.0000	16.0000	5.0000	2.2445	0.1125	2.7355	326
BIND	0.5826	0.6000	0.8571	0.0000	0.1443	−0.9613	4.9261	326
BOARD_TEN	6.4991	6.5000	8.2000	4.6000	0.7524	−0.1252	2.3750	326
SIZE	16.3918	16.4156	18.9629	13.1543	1.2317	−0.1025	2.7302	309
LEV	0.3631	0.3693	0.7572	0.0007	0.1159	−0.2339	5.3536	300
GROWTH	0.0840	0.0457	3.0421	−0.4583	0.2621	5.8523	61.4552	278

Notes: FFO_PS = funds from operations per share; DIV_YIELD = dividend yield; ROA = return on assets; ROE = return on equity; ROIC = return on invested capital; EPS = earnings per share; B_ACTIV = number of board meetings; B_SIZE = number of directors on the board; BIND = proportion of independent directors; BOARD_TEN = average board tenure (years); SIZE = natural logarithm of total assets; LEV = total debt divided by total assets; GROWTH = annual change in the natural logarithm of total assets. Sample: 30 JSE-listed REITs, 2013 to 2025.

Table 3 reports the parametric Pearson and non-parametric Spearman pairwise correlation matrices, both of which yield consistent empirical patterns. The accounting-based performance measures, namely ROA, ROE, ROIC, and EPS, demonstrate strong positive intercorrelations, with ROA and ROE recording a correlation coefficient of 0.94 across both estimations. FFO_PS correlates positively with ROA (0.28), ROE (0.28), and EPS (0.45), while exhibiting a negative correlation with DIV_YIELD (−0.27), all statistically significant at the one percent level. Among the board governance attributes, B_ACTIV displays a significant negative bivariate association with ROA (−0.25) and ROE (−0.22); however, subsequent multivariate regression analysis indicates that this relationship attenuates upon controlling for period fixed effects, supporting the notion that B_ACTIV responds to macro-level sector stress rather than directly influencing firm-level accounting performance. BIND demonstrates positive correlations with B_ACTIV (0.27) and BOARD_TEN (0.22). The control variables demonstrate associations with performance metrics in directions consistent with theoretical expectations. LEV is inversely related to both ROA (−0.36) and ROE (−0.26), whereas GROWTH is positively associated with ROA (0.41) and ROE (0.35), with all coefficients significant at the one percent level. Crucial to the empirical identification strategy is the positive correlation between B_SIZE and SIZE (0.57), which confirms that larger real estate investment trusts maintain larger boards in levels and verifies the presence of the scale confounding effect that the econometric specification is designed to mitigate. Subsequent panel regressions demonstrate that B_SIZE and SIZE remain separately identified via within-firm variation. Beyond this relationship, pairwise correlations among the independent variables remain strictly below 0.58, well under the conventional 0.80 threshold indicative of severe multicollinearity concerns, a conclusion further corroborated by variance inflation factor diagnostics.

Table 3. Pearson and Spearman correlation matrices of all variables for all company years.

Variables(s)	1	2	3	4	5	6	7	8	9	10	11	12	13
(1) FFO_PS	1.000	−0.279 ***	0.406 ***	0.420 ***	0.048	0.689 ***	0.063	0.118 **	0.104 *	−0.061	0.352 ***	−0.092	0.245 ***
(2) DIV_YIELD	−0.273 ***	1.000	−0.139 **	−0.133 **	0.243 ***	−0.262 ***	0.031	0.128 **	0.029	0.057	−0.014	0.018	−0.031
(3) ROA	0.280 ***	−0.224 ***	1.000	0.941 ***	0.397 ***	0.755 ***	−0.274 ***	−0.030	−0.145 ***	−0.220 ***	−0.141 **	−0.349 ***	0.660 ***
(4) ROE	0.278 ***	−0.266 ***	0.943 ***	1.000	0.386 ***	0.780 ***	−0.228 ***	−0.037	−0.123 **	−0.229 ***	−0.087	−0.213 ***	0.655 ***
(5) ROIC	0.006	−0.033	0.091	0.083	1.000	0.140 **	−0.147 ***	−0.062	−0.144 ***	−0.049	−0.301 ***	−0.177 ***	0.319 ***
(6) EPS	0.450 ***	−0.233 ***	0.523 ***	0.588 ***	0.027	1.000	−0.100 *	0.004	−0.043	−0.142 **	0.142 **	−0.193 ***	0.499 ***
(7) B_ACTIV	0.026	0.051	−0.247 ***	−0.219 ***	0.016	−0.041	1.000	0.036	0.414 ***	0.090	0.344 ***	0.231 ***	−0.320 ***
(8) B_SIZE	0.087	0.090	−0.089	−0.102 *	−0.036	0.010	0.051	1.000	0.027	0.117 **	0.583 ***	0.058	−0.039
(9) BIND	0.103 *	0.071	−0.078	−0.072	0.013	−0.031	0.274 ***	−0.000	1.000	0.212 ***	0.341 ***	0.207 ***	−0.185 ***
(10) BOARD_TEN	0.014	0.019	−0.179 ***	−0.177 ***	−0.033	−0.055	0.067	0.145 ***	0.221 ***	1.000	0.146 **	0.076	−0.237 ***
(11) SIZE	0.161 ***	0.004	−0.204 ***	−0.140 **	−0.081	0.047	0.349 ***	0.566 ***	0.211 ***	0.120 **	1.000	0.165 ***	−0.079
(12) LEV	0.031	0.111 *	−0.356 ***	−0.258 ***	−0.075	−0.128 **	0.212 ***	0.073	0.172 ***	0.021	0.216 ***	1.000	−0.308 ***
(13) GROWTH	0.091	−0.106 *	0.410 ***	0.389 ***	0.012	0.188 ***	−0.215 ***	−0.007	−0.121 **	−0.228 ***	−0.084	−0.177 ***	1.000

Notes: The bottom left half of the table contains Pearson’s parametric correlation coefficients, while the upper right half shows Spearman’s non-parametric correlation coefficients. ***, **, * represent correlation significance at 1%, 5% and 10%, respectively.

Table 3. Pearson and Spearman correlation matrices of all variables for all company years.

Variables(s)	1	2	3	4	5	6	7	8	9	10	11	12	13
(1) FFO_PS	1.000	−0.279 ***	0.406 ***	0.420 ***	0.048	0.689 ***	0.063	0.118 **	0.104 *	−0.061	0.352 ***	−0.092	0.245 ***
(2) DIV_YIELD	−0.273 ***	1.000	−0.139 **	−0.133 **	0.243 ***	−0.262 ***	0.031	0.128 **	0.029	0.057	−0.014	0.018	−0.031
(3) ROA	0.280 ***	−0.224 ***	1.000	0.941 ***	0.397 ***	0.755 ***	−0.274 ***	−0.030	−0.145 ***	−0.220 ***	−0.141 **	−0.349 ***	0.660 ***
(4) ROE	0.278 ***	−0.266 ***	0.943 ***	1.000	0.386 ***	0.780 ***	−0.228 ***	−0.037	−0.123 **	−0.229 ***	−0.087	−0.213 ***	0.655 ***
(5) ROIC	0.006	−0.033	0.091	0.083	1.000	0.140 **	−0.147 ***	−0.062	−0.144 ***	−0.049	−0.301 ***	−0.177 ***	0.319 ***
(6) EPS	0.450 ***	−0.233 ***	0.523 ***	0.588 ***	0.027	1.000	−0.100 *	0.004	−0.043	−0.142 **	0.142 **	−0.193 ***	0.499 ***
(7) B_ACTIV	0.026	0.051	−0.247 ***	−0.219 ***	0.016	−0.041	1.000	0.036	0.414 ***	0.090	0.344 ***	0.231 ***	−0.320 ***
(8) B_SIZE	0.087	0.090	−0.089	−0.102 *	−0.036	0.010	0.051	1.000	0.027	0.117 **	0.583 ***	0.058	−0.039
(9) BIND	0.103 *	0.071	−0.078	−0.072	0.013	−0.031	0.274 ***	−0.000	1.000	0.212 ***	0.341 ***	0.207 ***	−0.185 ***
(10) BOARD_TEN	0.014	0.019	−0.179 ***	−0.177 ***	−0.033	−0.055	0.067	0.145 ***	0.221 ***	1.000	0.146 **	0.076	−0.237 ***
(11) SIZE	0.161 ***	0.004	−0.204 ***	−0.140 **	−0.081	0.047	0.349 ***	0.566 ***	0.211 ***	0.120 **	1.000	0.165 ***	−0.079
(12) LEV	0.031	0.111 *	−0.356 ***	−0.258 ***	−0.075	−0.128 **	0.212 ***	0.073	0.172 ***	0.021	0.216 ***	1.000	−0.308 ***
(13) GROWTH	0.091	−0.106 *	0.410 ***	0.389 ***	0.012	0.188 ***	−0.215 ***	−0.007	−0.121 **	−0.228 ***	−0.084	−0.177 ***	1.000

Notes: The bottom left half of the table contains Pearson’s parametric correlation coefficients, while the upper right half shows Spearman’s non-parametric correlation coefficients. ***, **, * represent correlation significance at 1%, 5% and 10%, respectively.

Following the analysis of the Pearson and Spearman correlation matrices highlighting relationships and potential multicollinearity among variables, it is important to assess the extent of multicollinearity using the Variance Inflation Factor (VIF). VIF measures multicollinearity in a regression model (Shrestha, 2020). Multicollinearity occurs when independent variables in the model are highly correlated, leading to unreliable estimates of regression coefficients (Shrestha, 2020). A VIF value greater than 10 indicates significant multicollinearity, while values below 10 suggest negligible collinearity (Putri et al., 2023).

Table 4 presents the VIF and its reciprocal for each independent variable, including the firm-level controls. VIF measures multicollinearity in a regression model, with values greater than 10 indicating significant multicollinearity (Shrestha, 2020; Putri et al., 2023). The reported VIF values range from 1.12 to 1.71 with a mean of 1.29, far below the conventional cut-off. The two highest values belong to SIZE (1.71) and B_SIZE (1.57), reflecting their 0.57 correlation, and even these remain modest, confirming that the regressors, board variables and controls alike, are not collinear to any degree that threatens estimation.

Table 4. VIF Tests for independent variables.

Independent Variable(s)	VIF	1/VIF
B_ACTIV	1.25	0.80
B_SIZE	1.57	0.64
BIND	1.16	0.86
BOARD_TEN	1.12	0.89
SIZE	1.71	0.59
LEV	1.13	0.89
GROWTH	1.12	0.89
Mean VIF	1.29

Table 4. VIF Tests for independent variables.

Independent Variable(s)	VIF	1/VIF
B_ACTIV	1.25	0.80
B_SIZE	1.57	0.64
BIND	1.16	0.86
BOARD_TEN	1.12	0.89
SIZE	1.71	0.59
LEV	1.13	0.89
GROWTH	1.12	0.89
Mean VIF	1.29

6.2. Main Specification

Table 5 reports the main results: two-step system GMM with collapsed instruments and year dummies, one column per performance measure. Instruments number 19 in every model against 30 groups, satisfying the rule that instruments should not exceed groups (Roodman, 2009).

Table 5. Two-Step System GMM Estimates of Board Structure and REIT Performance (Main Specification: Collapsed Instruments, Year Dummies, Firm-Level Controls).

Variable(s)	FFO_P/S	DIV_YIELD	ROA	ROE	ROIC	EPS
DV(t−1)	−0.0368 (0.2728)	0.2716 *** (0.0627)	0.1331 * (0.0685)	0.1426 ** (0.0596)	0.0228 ** (0.0099)	0.7381 *** (0.0591)
B_ACTIV	−0.0913 (0.1234)	0.0062 ** (0.0030)	−0.0040 ** (0.0019)	−0.0070 ** (0.0033)	0.0001 (0.0024)	−0.0976 (0.1718)
B_SIZE	−0.0615 (0.0733)	0.0026 (0.0029)	−0.0032 * (0.0017)	−0.0100 *** (0.0034)	−0.0003 (0.0018)	−0.1179 (0.1067)
BIND	3.0883 (2.9369)	0.0475 (0.0451)	0.0618 (0.0380)	0.1080 * (0.0653)	0.0128 (0.0261)	2.9976 (3.3314)
BOARD_TEN	−0.0164 (0.3937)	0.0010 (0.0064)	0.0085 (0.0063)	0.0150 (0.0110)	−0.0012 (0.0045)	0.2671 (0.2945)
SIZE	0.4351 (0.3249)	−0.0114 ** (0.0047)	0.0002 (0.0048)	0.0077 (0.0109)	−0.0033 (0.0044)	0.2594 (0.2593)
LEV	−1.4026 (3.6688)	0.0856 ** (0.0413)	−0.1627 *** (0.0553)	−0.2834 ** (0.1170)	−0.0290 (0.0464)	−3.1733 (5.1569)
GROWTH	0.2242 (0.7277)	−0.0101 (0.0371)	0.1037 *** (0.0168)	0.1753 *** (0.0351)	0.0162 (0.0151)	2.4496 (1.7914)
Year dummies	Yes	Yes	Yes	Yes	Yes	Yes
N	235	233	240	238	240	240
Groups	30	30	30	30	30	30
Instruments	22	22	22	22	22	22
AR(1) p-value	0.265	0.132	0.055	0.046	0.304	0.080
AR(2) p-value	0.531	0.885	0.430	0.526	0.438	0.947
Hansen J p-value	0.529	0.120	0.549	0.569	0.724	0.348

Notes: N = 233 to 240 firm-year observations across 30 JSE-listed REITs, 2013 to 2025. Each column reports a separate two-step system GMM (Blundell–Bond) estimation with Windmeijer-corrected standard errors in parentheses. The instrument set comprises collapsed GMM-style instruments using lags 2 to 4 of the dependent variable, standard instruments for the board variables and firm-level controls, and year dummies. Year dummy coefficients are omitted for brevity. DV_(t−1) = lagged dependent variable; SIZE = ln(total assets); LEV = total debt/total assets; GROWTH = annual change in ln(total assets). * p < 0.10. ** p < 0.05. *** p < 0.01.

The lagged dependent variable demonstrates a positive and statistically significant relationship at the ten percent level or better across five of the six estimated models, specifically for DIV_YIELD (0.272, $p<0.01$), EPS (0.738, $p<0.01$), ROE (0.143, $p<0.05$), ROIC (0.023, $p<0.05$), and ROA (0.133, $p<0.10$). Only FFO_PS exhibits no statistically significant persistence. The estimated coefficient for EPS highlights strong persistence in per-share earnings, reflecting the forward-rolling nature of contracted rental escalations, a distributional property that justifies the selection of system GMM over difference GMM for this panel. Furthermore, these persistence estimates are lower than those observed in the baseline specification lacking control variables, as LEV and GROWTH absorb a portion of the temporal dynamics previously captured by the lag structure alone.

The firm-level control variables align closely with theoretical expectations within the real estate investment trust literature, validating their inclusion and contextualizing the interpretation of the governance coefficients. LEV exerts a negative and statistically significant influence in both accounting return models (−0.163, $p<0.01$ for ROA; −0.283, $p<0.05$ for ROE), indicating that higher gearing compressed net returns over an observation period characterized by a complete interest-rate cycle. Conversely, GROWTH exhibits a positive and highly significant relationship within the same specifications (0.104 and 0.175, both $p<0.01$). SIZE yields statistically insignificant coefficients across all return models and exhibits a negative association with DIV_YIELD (−0.011, $p<0.05$), consistent with larger entities trading at lower yields.

Holding the lagged dependent variable, remaining governance metrics, firm-level controls, and both firm and year fixed effects constant, B_SIZE maintains a negative and statistically significant association with both accounting return measures (−0.0032 for ROA, $p=0.070$; −0.0100 for ROE, $p<0.01$). Specifically, each additional director is associated with a 1.0 percentage point reduction in ROE and a 0.32 percentage point reduction in ROA. The ROE point estimate increases in magnitude compared to the specification without controls (−0.0073), contradicting the pattern expected had B_SIZE merely served as a proxy for omitted firm scale. This identification is further reinforced by the finding that SIZE remains statistically insignificant while B_SIZE retains its significance, confirming that board scale and balance-sheet scale are separately identified. Economically, this effect is substantial relative to baseline profitability. Given a sample mean ROE of 6.98 percent and a standard deviation of 12.9 percentage points, a one standard deviation increase in B_SIZE (2.2 directors) corresponds to a 2.2 percentage point decline in ROE, representing approximately one-third of mean sample profitability. Because direct board maintenance costs are insufficient to explain differentials of this scale, the observed effect likely operates through managerial decision-making processes, including capital allocation, acquisition discipline, capital structure choices, and operational responsiveness during market stress.

B_ACTIV carries a negative coefficient that achieves statistical significance at the five percent level in both return models (−0.0040 for ROA; −0.0070 for ROE). This result requires cautious interpretation, as the coefficient attenuates once treated under endogenous regressor frameworks, a pattern consistent with reactive governance mechanisms where board meetings increase in response to firm-specific distress. BIND demonstrates a positive association with return measures, reaching the ten percent significance level for ROE (0.108), whereas its association with DIV_YIELD loses statistical significance upon the inclusion of control variables (0.048, $p=0.29$). Finally, BOARD_TEN remains statistically insignificant across all specifications.

6.3. Diagnostic and Specification Checks

The primary econometric specifications satisfy all necessary diagnostic criteria across the estimated models. The Arellano–Bond test for second-order serial correlation, AR(2), fails to reject the null hypothesis of no second-order autocorrelation across all models, with $p$-values ranging from 0.348 to 0.947, thereby confirming the validity of the second lagged ($t-2$) instruments. Similarly, the Hansen $J$-test of overidentifying restrictions fails to reject the null hypothesis across all specifications, yielding $p$-values between 0.120 and 0.724. These statistics fall well within the optimal diagnostic range, remaining clear of both the rejection threshold and the near-unity values typically associated with instrument proliferation.

The difference-in-Hansen test evaluating the additional moment conditions of the system GMM level equations fails to reject the null hypothesis for ROA ($p=0.242$) and ROE ($p=0.207$), supporting the validity of the initial-conditions assumption for the models containing the primary empirical findings. Conversely, the test rejects the null hypothesis at the five percent significance level for DIV_YIELD ($p=0.016$), which warrants caution regarding the robust interpretation of the board independence and payout association.

The Arellano–Bond test for first-order serial correlation, AR(1), rejects the null hypothesis at or near the five percent level for ROA ($p=0.055$), ROE ($p=0.046$), and EPS ($p=0.080$), while failing to reject for DIV_YIELD, FFO_PS, and ROIC. Finally, the FFO_PS model exhibits the weakest diagnostic performance within the analysis, displaying neither statistically significant persistence nor significant explanatory regressors. Consequently, while its results are reported for completeness, they carry no interpretive weight in the overall findings.

6.4. Results with Endogenous Board Variables

Table 6 relaxes the exogeneity assumption on board structure, instrumenting all four board variables GMM-style by their own collapsed lags 2 to 3 while retaining the firm-level controls (29 instruments). This specification directly addresses reverse causality from performance to board composition.

Table 6. Two-Step System GMM Estimates Treating Board Variables as Endogenous (With Firm-Level Controls).

Variable(s)	FFO_P/S	DIV_YIELD	ROA	ROE	ROIC	EPS
DV(t−1)	−0.2140 (0.1426)	0.2912 *** (0.0522)	0.0883 (0.0956)	0.1091 (0.1367)	0.0072 (0.0148)	0.6477 *** (0.0629)
B_ACTIV	−0.1330 (0.2864)	0.0091 (0.0062)	0.0034 (0.0046)	0.0002 (0.0112)	0.0010 (0.0063)	−0.0200 (0.3585)
B_SIZE	−0.3216 * (0.1840)	0.0008 (0.0067)	−0.0115 * (0.0067)	−0.0226 * (0.0120)	−0.0008 (0.0067)	−0.6643 (0.4430)
BIND	1.5045 (3.4275)	0.0606 (0.1212)	0.0510 (0.1237)	−0.0025 (0.2076)	0.0224 (0.0694)	−1.7283 (6.2165)
BOARD_TEN	0.7706 (0.4809)	−0.0095 (0.0135)	0.0157 (0.0123)	0.0186 (0.0163)	0.0009 (0.0090)	0.8830 (0.6432)
SIZE	1.1138 ** (0.5192)	−0.0115 (0.0119)	0.0063 (0.0094)	0.0241 (0.0173)	−0.0043 (0.0068)	1.1693 (0.7401)
LEV	−4.8265 (2.9783)	0.0342 (0.0588)	−0.1957 ** (0.0988)	−0.2892 * (0.1752)	−0.0108 (0.0653)	−4.8300 (6.3100)
GROWTH	0.1301 (0.7913)	−0.0154 (0.0129)	0.1050 *** (0.0214)	0.1712 *** (0.0415)	0.0142 (0.0205)	2.6504 (1.7704)
Year dummies	Yes	Yes	Yes	Yes	Yes	Yes
N	235	233	240	238	240	240
Groups	30	30	30	30	30	30
Instruments	29	29	29	29	29	29
AR(1) p-value	0.382	0.135	0.053	0.063	0.304	0.097
AR(2) p-value	0.857	0.733	0.435	0.553	0.643	0.849
Hansen J p-value	0.561	0.440	0.302	0.594	0.938	0.718

Notes: Two-step system GMM with year dummies and collapsed instruments, treating all four board structure variables as endogenous, instrumented GMM-style by their own lags 2 to 3, with the dependent variable instrumented by its collapsed lags 2 to 3 and the firm-level controls entering as standard instruments. Windmeijer-corrected standard errors in parentheses. * p < 0.10. ** p < 0.05. *** p < 0.01.

Several prominent empirical insights emerge from these endogeneity-corrected specifications. Notably, the negative relationship between B_SIZE and accounting profitability remains statistically meaningful across both return models, yielding point estimates of −0.0226 for ROE ($p=0.059$) and −0.0115 for ROA ($p=0.084$). These point estimates are approximately twice the magnitude of their counterparts in the baseline specifications, an expansion in scale consistent with the mitigation of attenuation bias arising from reverse causality, whereby underperforming real estate investment trusts expand board capacity in response to operational stress. Furthermore, the estimated coefficients for B_ACTIV, which previously demonstrated statistical significance, attenuate to near-zero values under instrumentation (+0.0034 for ROA and +0.0002 for ROE). This attenuation substantiates the hypothesis that board meeting frequency operates primarily as a reactive mechanism to firm-specific difficulties rather than a structural determinant of corporate performance. Additionally, LEV and GROWTH retain their expected theoretical signs and statistical significance, while the Hansen $J$-test $p$-values remain within the non-rejection range of 0.302 to 0.594 for the return models, confirming instrument validity. Ultimately, the inverse relationship between B_SIZE and accounting profitability constitutes the sole governance metric that demonstrates consistent empirical robustness across every identification framework evaluated in this investigation.

6.5. LSDVC Robustness

Table 7 reports the bootstrap bias-corrected LSDV estimates for ROA and ROE with the firm-level controls included. The correction behaves as theory predicts on the persistence parameter, raising the downward-biased LSDV estimates toward the GMM values.

Table 7. Bootstrap Bias-Corrected LSDV (LSDVC) Estimates for ROA and ROE.

Variable(s)	ROA	ROE
DV(t−1)	0.1140 * (0.0643)	0.1989 *** (0.0629)
B_ACTIV	−0.0034 * (0.0020)	−0.0064 (0.0048)
B_SIZE	−0.0042 * (0.0024)	−0.0091 * (0.0053)
BIND	0.0723 * (0.0376)	0.1437 * (0.0775)
BOARD_TEN	0.0117 (0.0077)	0.0155 (0.0165)
SIZE	0.0102 (0.0162)	0.0324 (0.0342)
LEV	−0.2328 *** (0.0692)	−0.4409 *** (0.1418)
GROWTH	0.0746 *** (0.0228)	0.1578 *** (0.0474)
Firm and year FE	Yes	Yes
N	241	239
Groups	30	30
Bootstrap replications	500	500

Notes: Dynamic fixed effects estimates with firm and year fixed effects and firm-level controls. The Nickell bias is removed by parametric bootstrap correction (500 replications, firm-specific initial conditions held at actual values), following the approach of Everaert and Pozzi (2007). Bootstrap standard errors in parentheses. Regressors are treated as exogenous; the estimator corrects dynamic panel bias but not simultaneity. * p < 0.10. *** p < 0.01.

The empirical results across the alternative specifications demonstrate convergence between the estimation techniques. Under the LSDVC estimator, board size maintains a negative and statistically significant association with accounting returns at the ten percent level (−0.0042 for ROA, $p=0.085$; −0.0091 for ROE, $p=0.085$), displaying magnitudes remarkably close to the primary system GMM estimates. Additionally, the leverage and asset growth controls preserve both their directional alignment and strong statistical significance observed in the GMM models. In baseline specifications lacking firm-level controls, the LSDVC board size coefficients exhibited identical signs and comparable point estimates but failed to achieve statistical significance ($p=0.31$ for ROA; $p=0.26$ for ROE). This prior divergence between GMM and LSDVC inferences prompted two competing explanations: attenuation bias in LSDVC due to its strict exogeneity assumptions, or the understatement of standard errors within system GMM given a cross-sectional dimension of 30 firms. Both considerations retain analytical validity. Indeed, with a 30-group panel, system GMM asymptotic properties face inherent constraints, and while the Windmeijer correction mitigates finite-sample downward bias in two-step standard errors, it may not eliminate it entirely. Nevertheless, the controlled empirical results establish a robust finding: the inverse relationship between board size and accounting returns is consistently identified across two distinct estimators characterized by different bias structures and identifying assumptions. The effect remains significant between the one percent and ten percent levels under GMM, achieves ten percent significance under LSDVC, and increases in magnitude when board variables are instrumented. Furthermore, the LSDVC estimates indicate that board independence exhibits a positive and marginally significant relationship with profitability (0.072 for ROA, $p=0.054$; 0.144 for ROE, $p=0.064$), a pattern that warrants further consideration within the broader discussion.

6.6. Outlier Robustness and Sensitivity Across Specifications

The descriptive statistics reveal pronounced non-normality across the performance metrics, evidenced by excess kurtosis values of 7.8 for ROA, 11.5 for ROE, and 297.6 for ROIC. Given that the central empirical conclusions of this study depend on the accounting return specifications, the ROA and ROE models are re-estimated using dependent variables winsorized at the 1st and 99th percentiles. These robust estimations are executed across both the baseline specification and the endogenous-regressor framework, with the corresponding results reported in Table 8.

Table 8. Winsorized ROA and ROE: Two-Step System GMM With Firm-Level Controls.

Variable(s)	ROA (main)	ROE (main)	ROA (endog.)	ROE (endog.)
DV(t−1)	0.1211 (0.0767)	0.0945 (0.0633)	0.0814 (0.0971)	0.1015 (0.1216)
B_ACTIV	−0.0038 ** (0.0019)	−0.0070 * (0.0036)	0.0023 (0.0039)	0.0011 (0.0104)
B_SIZE	−0.0030 * (0.0016)	−0.0098 *** (0.0032)	−0.0097 (0.0062)	−0.0210 * (0.0115)
BIND	0.0580 * (0.0329)	0.1035 * (0.0617)	0.0409 (0.1147)	−0.0154 (0.2115)
BOARD_TEN	0.0077 (0.0056)	0.0172 * (0.0100)	0.0171 (0.0115)	0.0192 (0.0158)
SIZE	−0.0004 (0.0038)	0.0025 (0.0080)	0.0028 (0.0086)	0.0206 (0.0149)
LEV	−0.1411 *** (0.0441)	−0.1873 ** (0.0753)	−0.1473 * (0.0764)	−0.2278 * (0.1214)
GROWTH	0.0958 *** (0.0145)	0.1482 *** (0.0271)	0.0955 *** (0.0168)	0.1622 *** (0.0337)
Year dummies	Yes	Yes	Yes	Yes
N	240	238	240	238
Instruments	22	22	29	29
AR(2) p-value	0.453	0.616	0.441	0.600
Hansen J p-value	0.795	0.546	0.331	0.627

Notes: Dependent variables winsorized at the 1st and 99th percentiles. Main columns replicate the Table 5 specification; endogenous columns replicate the Table 6 specification. Windmeijer-corrected standard errors in parentheses. * p < 0.10. ** p < 0.05. *** p < 0.01.

The estimated B_SIZE coefficients remain virtually unchanged following winsorization, recording values of −0.0098 ($p<0.01$) for ROE and −0.0030 ($p=0.065$) for ROA in the baseline specification, and −0.0210 ($p=0.068$) and −0.0097 ($p=0.117$) under the endogenous regressor framework. These results confirm that the primary empirical finding is not driven by extreme tail observations. Re-estimating the specifications without firm-level controls yields a coefficient of −0.0077 ($p=0.0001$) for ROE, aligning closely with the unadjusted estimates. Furthermore, the winsorized ROIC model exhibits a persistence coefficient of 0.513 ($p<0.01$) while all board governance variables remain statistically insignificant at the five percent level. This outcome verifies that raw ROIC dynamics were heavily influenced by a limited set of extreme values and demonstrates that the negative BOARD_TEN relationships observed under the initial difference GMM specification does not persist following model correction and outlier adjustment. Across all econometric specifications evaluated in this study, the empirical findings demonstrate a clear hierarchy of robustness. The inverse relationship between B_SIZE and ROE achieves statistical significance at the one percent level in the main controlled specification and remains robust under the endogenous regressor treatment ($p=0.059$), dependent variable winsorization ($p<0.01$), winsorization combined with endogeneity controls ($p=0.068$), and the LSDVC estimator ($p=0.085$), while consistently maintaining its sign and magnitude across all uncontrolled specifications. The inverse association between B_SIZE and ROA exhibits an identical pattern, albeit at broader significance levels ranging between one percent and twelve percent. BIND maintains a positive direction across all return specifications, reaching marginal significance under the LSDVC estimator and within the controlled ROE model. However, its association with DIV_YIELD attenuates upon the inclusion of control variables, and the specification fails the difference-in-Hansen diagnostic, preventing these BIND patterns from being interpreted as conclusive. The negative B_ACTIV coefficients observed in controlled instrumental variable specifications attenuate completely under the endogenous regressor treatment, aligning with the reactive meeting hypothesis. BOARD_TEN displays no statistical significance across any specification. Importantly, none of the empirical associations detected under the initial difference GMM design (specifically, BIND on FFO_PS, B_SIZE on DIV_YIELD, and BOARD_TEN on ROIC) survive the updated model specification. This divergence demonstrates within a single dataset how instrument proliferation and omitted period fixed effects can generate spurious governance relationships within small-sample panel structures

7. Discussion

The empirical findings indicate that board structure relates to real estate investment trust performance through specific operational channels rather than uniformly across metrics. These channels align directly with institutional expectations, concentrating in accounting efficiency and payout measures, while remaining absent from per-share earnings constructs.

The central finding of this investigation is the negative association between B_SIZE and accounting profitability. This relationship remains stable in sign and magnitude across every estimator, robust to outlier adjustment through winsorization, and identified by both system GMM and the bias-corrected LSDVC estimator once firm-level controls enter the specification. Two features of these controlled results bear directly on its economic interpretation. Firm size (SIZE), LEV, and asset growth (GROWTH) behave as predicted in the corporate finance literature. Notably, SIZE is statistically insignificant in the return models, whereas B_SIZE maintains strong statistical significance, confirming that the observed effect attaches to governance structure rather than balance-sheet scale. Furthermore, the magnitude of the B_SIZE coefficient expands rather than shrinks upon inserting firm-level controls, ruling out an omitted-scale artifact. The negative effect strengthens further when board composition is instrumented under the endogenous regressor framework, a pattern consistent with reverse causality attenuating rather than manufacturing the estimate.

These empirical patterns align with the coordination cost and communication friction arguments established in classic governance literature (Lipton & Lorsch, 1992; Jensen, 1993; Yermack, 1996; Eisenberg et al., 1998; Guest, 2009). The real estate investment trust setting sharpens this theoretical interpretation. The two primary mechanisms through which larger boards generate value, namely advising on complex and opaque operations and securing external resources through network ties, are less pronounced in real estate investment trusts relative to general industrial firms. The underlying business model is transparent, revenues derive from contractual leases on physical assets, and mandatory payout regulations constrain managerial discretion over free cash flow. Where managerial discretion is structurally circumscribed by regulation, the marginal monitoring benefit of additional directors diminishes, whereas coordination costs persist. Consequently, additional directors add procedural complexity without providing commensurate oversight value.

This result contrasts with positive board-size findings reported in emerging-market studies of opaque corporations (Hasan et al., 2024; Zureigat et al., 2024) and with the positive association between board size and DIV_YIELD observed in uncorrected baseline specifications. This contrast defines the operational boundaries of resource dependence theory at the precise institutional boundaries that real estate investment trusts inhabit. Conversely, stewardship theory, which posits performance gains from board scale and continuity, receives no empirical support within this sector.

Evidence regarding BIND is directionally positive across all specifications but nowhere decisive, and the inclusion of firm-level controls clarifies its empirical standing. The positive association with DIV_YIELD, which represented the strongest result for this variable in unadjusted models and is theoretically grounded in the outcome model of agency dividends (La Porta et al., 2000; Viviers et al., 2023), falls short of statistical significance once controls are introduced ($p=0.29$). Moreover, the associated specification fails the difference-in-Hansen test, precluding it from being treated as a robust finding.

Nevertheless, the BIND coefficient remains strictly positive across every return model in all estimated specifications. It achieves marginal statistical significance for ROE in the main controlled model and for both ROA and ROE under the LSDVC estimator. Unlike the directional instability observed in the tenure metric, this persistent positive alignment suggests a weak underlying relationship that the current sample size lacks the statistical power to identify definitively. While Kitulazzi et al. (2025) report that independent directors correlate negatively with short-term accounting returns in listed real estate firms, the dynamic panel estimates presented here point in the opposite direction, albeit at significance levels requiring cautious interpretation.

The findings regarding B_ACTIV provide methodological insights into governance dynamics. In specifications omitting year fixed effects, B_ACTIV demonstrates a negative relationship with accounting returns, whereas introducing year dummies eliminates this bivariate association. Re-introducing firm controls yields a negative coefficient significant at the five percent level in standard instrumental variable specifications, but this association attenuates to zero once B_ACTIV is instrumented by its own collapsed lags.

This progression illustrates the reactive meetings mechanism operating within corporate boards (Vafeas, 1999). Board meeting frequency increases in response to both sector-wide macroeconomic stress, which is absorbed by year fixed effects, and firm-specific operational difficulties, which are exposed when firm controls sharpen the specification. The raw pairwise correlations reported between B_ACTIV and accounting performance reflect this bivariate phenomenon. Consequently, studies asserting negative causal performance relationships from meeting frequency without accounting for time fixed effects or endogeneity should be interpreted with caution. Board meeting frequency functions primarily as an endogenous reaction to operational stress rather than a structural determinant of firm performance.

The empirical null regarding BOARD_TEN represents a genuine structural finding rather than a specification artifact. BOARD_TEN remains statistically insignificant across every estimated model and specification, exhibiting unstable directional signs. The single significant tenure coefficient generated in baseline difference GMM estimations for ROIC vanishes following model correction and outlier adjustment. This null is plausible given the sample characteristics because board tenure is concentrated within a narrow range averaging 4.6 to 8.2 years, excluding both highly inexperienced and deeply entrenched boards where theoretical effects typically manifest. Within this intermediate range, competing theoretical mechanisms regarding accumulated managerial expertise versus board entrenchment appear to offset each other.

The divergence across performance metric families carries substantive analytical value. Board governance attributes demonstrate statistical associations with ROA, ROE, and DIV_YIELD, but exhibit no relationship with EPS or FFO_PS. The per-share metrics incorporate capital structure decisions, equity issuance histories, and share counts, all of which represent variations not directly governed by board oversight. In contrast, accounting return measures isolate underlying operating and allocative efficiency, which is the specific domain where board oversight binds. Additionally, FFO_PS exhibits no temporal persistence and fails joint significance tests, indicating measurement noise within that series for this panel.

These governance findings should be interpreted relative to the broader macroeconomic determinants of real estate investment trust performance. Macroeconomic uncertainty, market sentiment, and systemic shocks represent first-order drivers of real estate returns (Lin et al., 2009; Bossman et al., 2022; Mensi et al., 2023; Hadad et al., 2024). The year fixed effects utilized in this study absorb the common component of these macro-level forces, as evidenced by the attenuation of several baseline governance signals once time effects are specified. At the firm level, LEV and GROWTH exert substantially larger and more precisely estimated impacts on performance than any governance variable. This hierarchy reflects the operational reality of a transparent, capital-intensive, and highly geared sector. The governance coefficients documented here represent conservative within-firm effects that survive after controlling for dominant systemic drivers and key balance-sheet determinants. That most board variables remain statistically insignificant under this benchmark confirms that within a regulated and distribution-constrained sector, board scale constitutes the primary internal governance attribute with measurable performance implications.

8. Conclusions

We examined whether board structure is associated with the performance of South African REITs, focusing on board activity (B_ACTIV), board size (B_SIZE), board independence (BIND), and board tenure (BOARD_TEN) over the period 2013 to 2025. Employing a two-step system GMM framework with collapsed instruments, year fixed effects, Windmeijer-corrected standard errors, and firm-level controls for firm size (SIZE), leverage (LEV), and asset growth (GROWTH), the empirical strategy was further supported by an endogenous-regressor specification, a bias-corrected LSDVC estimator, and outlier-adjusted robustness checks. Together, these estimators provide REIT-specific evidence on how board architecture relates to cash-flow-based and accounting-based performance metrics.

The empirical evidence supports a channel-specific reading of the overarching hypothesis. B_SIZE is inversely associated with accounting returns: each additional director corresponds to a 1.0 percentage point reduction in ROE and a 0.32 percentage point reduction in ROA in the primary specification. Furthermore, the ROE effect, expanding to −2.3 percentage points per director when board composition is instrumented, holds across every identification strategy, dependent variable winsorization, and control specification. A one-standard-deviation increase in B_SIZE corresponds to an ROE reduction of approximately 2.2 percentage points, representing roughly one-third of mean sector profitability. Critically, company SIZE itself remains statistically insignificant in these return models, whereas LEV and GROWTH carry their expected theoretical signs and significance. Consequently, the B_SIZE effect is identified separately from balance-sheet scale and cannot be attributed to omitted firm-level attributes. In contrast, BIND exhibits directionally positive but statistically fragile associations with returns and payout ratios, while B_ACTIV and BOARD_TEN display no robust association with any performance measure. The B_ACTIV effects that appear in exogeneity-assuming specifications attenuate entirely under instrumentation, consistent with meeting frequency reacting to firm-specific and sector-wide stress rather than driving financial returns.

For corporate boards and nomination committees, these findings counsel restraint regarding board expansion. No offsetting benefits of scale manifest anywhere in the corrected empirical results, and the estimated cost to accounting returns is material. For investors, B_SIZE offers a low-cost, observable governance screen, while BIND provides a tentative directional payout signal. For regulators and governance bodies, the results support independence-focused guidance while cautioning against prescriptions that mechanically enlarge boards, such as mandated additional committee appointments, or relying on meeting counts and tenure thresholds as primary governance metrics. Numerical optima for board size, independence proportions, or tenure limits are deliberately avoided because the empirical evidence identifies directions and magnitudes of association rather than structural optima.

Beyond the specific institutional setting, this study offers a methodological contribution to the broader emerging-market governance literature. Within a single dataset, three statistically significant governance relationships generated under a baseline difference GMM specification with proliferated instruments and omitted time effects were eliminated upon collapsing the instrument matrix, adding year fixed effects, and transitioning to system GMM. Concurrently, a distinct and more coherent set of empirical findings emerged. Given the prevalence of small governance panel datasets in this literature, the specification sensitivities documented here are likely to generalize across settings.

Several limitations qualify these conclusions. Firstly, the cross-sectional dimension is small. While thirty firms represent substantially the entire population of the JSE REIT sector and enhance external validity for this market, the small $N$ strains large-sample system GMM asymptotics. Secondly, while the firm-level controls encompass the principal balance-sheet drivers, they do not include property subsector composition, development pipelines, or market-based valuation ratios. Furthermore, listwise deletion across these controls reduces the estimation samples to between 233 and 240 observations, representing a trade-off between specification completeness and sample size that was resolved in favour of the former. Thirdly, Arellano–Bond AR(1) diagnostics do not reject the null of no first-order serial correlation for DIV_YIELD, FFO_PS, and ROIC, meaning the dynamic panel specification for these metrics relies primarily on endogeneity arguments rather than residual correlation structures. Lastly, the governance variables remain coarse proxies. For instance, independence proportions do not capture director quality, committee dynamics, or informal board influence, and meeting counts do not reflect meeting substance. Future research could extend this analysis to other emerging-market REIT regimes; incorporate additional governance dimensions such as board gender diversity, audit committee effectiveness, and CEO duality; examine interaction terms across crisis periods, including interactions of firm-level exposures with economic policy uncertainty and geopolitical risk indices; and combine quantitative methodologies with qualitative insights from board evaluations.

Finally, board structure represents a non-neutral feature of REIT governance. In the South African listed property market over the 2013 to 2025 period, board scale exerted a measurable cost on accounting profitability that firm scale, leverage, and growth cannot explain. Conversely, board independence provided directional but fragile positive signals, and easily countable metrics such as meeting counts and tenure served carried no persistent performance information. These findings advance the governance debate beyond generic prescriptions, highlighting the specific structural levers that matter within a transparent, capital-intensive, and regulated sector.