Our panel covers U.S. non-financial, non-utility firms from 1996–2023 matched across Compustat, Center for Research in Security Prices (CRSP), and RiskMetrics. We exclude American Depositary Receipts (ADRs) and non-ordinary shares and winsorize continuous variables at the 1st/99th percentiles. Following De Franco et al. (2011), we require at least 12 non-missing quarters among the prior 16 to estimate firm-specific accounting mappings and at least five same-industry (2-digit SIC) peers to compute pairwise comparability. The final sample includes 7020 firms and 74,860 firm-year observations.

We evaluate the connectedness of a firm’s board by examining inter-organizational links formed through shared directorships. Two directors are considered connected when they served on the same corporate board. Consistent with prior work on board networks (Larcker et al., 2013; Chiu et al., 2013; Bakke et al., 2024), we capture each director’s position within the overall network using four widely adopted metrics: degree, closeness, betweenness, and eigenvector centrality.

Financial statement comparability is constructed following the framework of De Franco et al. (2011), who conceptualize an accounting system as the function that maps underlying economic events into reported financial outcomes. Two firms are therefore considered more comparable when their accounting systems produce similar earnings responses to identical economic conditions. In practice, stock returns serve as proxies for economic events, while earnings represent the resulting accounting outputs.

To mitigate concerns that cross-firm differences in operating scope may confound the relation between board-network centrality and comparability, we augment our baseline models with several Compustat segment-based measures: SEGCOUNT (number of operating segments), GEOCOUNT (number of geographic segments), FOREIGN_SALES (foreign-to-total sales ratio), and SEG_HHI (the Herfindahl index of segment sales concentration, where larger values indicate more specialization). We also construct DIVERSITY_PC1, which is the first principal component extracted from standardized SEGCOUNT, GEOCOUNT, and FOREIGN_SALES. All additional variables are lagged one period and winsorized at the 1st and 99th percentiles.

Using rolling information from the prior sixteen quarters, we estimate for each firm-year the following firm-specific accounting-return relation:

$$Earning{s}_{it}={\alpha }_i+{\beta }_i*Retur{n}_{it}+{ϵ}_{it}$$

Here, $Retur{n}_{it}$ denotes the raw quarterly stock return, and $Earning{s}_{it}$ is quarterly net income before extraordinary items scaled by the market value of equity at the beginning of the quarter. The estimated coefficients ${\alpha }_i$ and ${\beta }_i$ characterize firm $i$’s accounting system—how its earnings respond to economic shocks. For comparison, we obtain analogous parameters ${\widehat{\alpha }}_j$ and ${\widehat{\beta }}_j$ for peer firm $j$ within the same two-digit SIC industry.

The similarity between firms $i$ and $j$’s accounting functions is evaluated by comparing predicted earnings for a common economic shock. Using firm $i$’s stock returns, we compute the predicted earnings from each accounting system:

$$E{\left(Earnings\right)}_{iit}={\widehat{\alpha }}_i\mathrm{}+{\widehat{\beta }}_i*Retur{n}_{it}$$

$$E{\left(Earnings\right)}_{ijt}={\widehat{\alpha }}_j\mathrm{}+{\widehat{\beta }}_j*Retur{n}_{it}$$

The accounting comparability between firms $i$ and $j$($COM{P}_{ijt}$) is then defined as minus one times the mean absolute difference between these predicted earnings over a rolling window of sixteen quarters (De Franco et al., 2011; Gao et al., 2024):

$$CompAcc{t}_{ijt}=-\frac{1}{16}{\sum }_{t-15}^t\left|E{\left(Earnings\right)}_{iit}-E{\left(Earnings\right)}_{ijt}\right|$$

A value closer to zero implies nearly identical predicted earnings (high comparability), whereas more negative values indicate divergence in their accounting responses. We also construct two additional comparability outcomes: (i) the average of the four largest comparability scores for firm $i$ in period $t$, and (ii) the median comparability for firm $i$ relative to all firms in its industry in period $t$. Similar measures are formed for cash flow comparability—based on quarterly cash flow from operations—and for accruals comparability—using quarterly accruals. For each case, we compute both the average of each firm’s top four scores and the industry-level median.

Table 1A and Table 1B summarizes the distribution of observations over time, showing a reasonably balanced sample across the years. Descriptive statistics appear in Table 2. By construction, comparability measures display negative means with upper bounds at zero. Their relatively large standard deviations suggest meaningful variation across firms, including instances of very low comparability. The centrality variables exhibit means near zero, consistent with many firms lacking board interlocks. Sample firms have an average book-to-market ratio of 1.68, long-term leverage of roughly 15.91% of assets, and mean profitability of –0.11% of assets. Roughly 31.31% of sample firms report losses, while average annual sales growth is 18.31%.

Table 3 presents correlations among the primary variables. The various comparability measures are positively associated with one another, as expected. Centrality metrics correlate with several control variables in predictable ways. Importantly, unconditional correlations between centrality and comparability are not indicative of the true multivariate relationship because the regression models absorb firm-level heterogeneity and time-specific shocks. Once these factors are accounted for, the baseline regressions reveal a negative and statistically meaningful association between board-network centrality and financial statement comparability, consistent with the central hypothesis.

Table 4A and Table 4B presents our baseline regression results examining the impact of board network centrality on FSC. We find that greater board connectedness is significantly associated with lower financial statement comparability. This negative relationship holds across all four-centrality metrics, implying that firms with more influential or networked directors tend to have accounting systems that diverge more from their peers.

Our alternative comparability measures, Comp4 and CompIND, corroborate this pattern. These results suggest that well-connected boards may facilitate idiosyncratic or opaque reporting choices, possibly due to greater managerial discretion or the dissemination of non-standard practices via shared board members.

We re-estimate the baseline adding SEGCOUNT, GEOCOUNT, FOREIGN_SALES, and SEG_HHI; results are unchanged in sign and remain statistically significant with modest attenuation. A parsimonious specification with DIVERSITY_PC1 delivers similar inferences. Interactions of centrality with diversity proxies are negative and significant, indicating that broader operational scope strengthens the negative centrality–FSC relation. See Appendix Table A.

Beyond statistical significance, the magnitudes are economically meaningful. Using the CompIND specification (Table 4A), a one–standard-deviation (1$\sigma$) increase in Eigenvector centrality (SD = 0.0140) is associated with a 0.00225 decrease in comparability (–0.1605 $\times$ 0.0140), which is about 0.23$\sigma$ of CompIND (SD = 0.0096) and $\approx$10% of its sample mean magnitude ($|$mean$|$ = 0.0218). Likewise, a 1$\sigma$ increase in Betweenness (SD = 0.0024) and Degree (SD = 0.0020) maps to –0.00149 and –0.00129, i.e., $\approx$0.15$\sigma$ and $\approx$0.13$\sigma$ of CompIND, respectively. To give a more intuitive yardstick, moving from the 25th to the 75th percentile of board connectedness implies sizable drops in comparability. For example, Eigenvector rises from 0.0000 (p25) to 0.0147 (p75), which corresponds to –0.00246 in CompIND ($\approx$0.26$\sigma$ of CompIND; $\approx$11% of the mean magnitude). The analogous p25$\to$p75 moves for Betweenness (0.0000$\to$0.0030) and Degree (0.0000$\to$0.0027) imply –0.00192 ($\approx$0.20$\sigma$) and –0.00181 ($\approx$0.19$\sigma$), respectively. Results are similar—if anything slightly larger in standard-deviation units—when using Comp4 (Table 4B). A 1$\sigma$ increase in Eigenvector centrality implies a –0.00101 change in Comp4 (–0.0721 $\times$ 0.0140), which is $\approx$0.37$\sigma$ of Comp4 (SD = 0.0027) and about 48% of its mean magnitude ($|$mean$|$ = 0.0021). The corresponding 1$\sigma$ effects for Degree and Betweenness are –0.00057 each ($\approx$0.21–0.21$\sigma$ of Comp4; $\approx$27% of $|$mean$|$). Note that by construction higher (less negative) values mean greater comparability, so the negative coefficients translate into economically meaningful declines in comparability as connectedness increases.

Tables 5A,5B and Tables 6A,6B extend the analysis to cash flow and accrual comparability, respectively. We observe that board centrality is likewise negatively associated with both types of comparability, reinforcing our interpretation. These results indicate that the influence of board networks extends beyond accrual-based measures of earnings quality to encompass broader aspects of financial reporting. Using cash-flow comparability (Table 5A), a 1$\sigma$ increase in Eigenvector centrality implies –0.00290 in CompIND_Cash (–0.2073 $\times$ 0.0140), which is $\approx$0.26$\sigma$ of CompIND_Cash (SD = 0.0112) and $\approx$12.5% of its mean magnitude ($|$mean$|$ = 0.0232). Degree and Betweenness deliver –0.00203 and –0.00253 ($\approx$0.18–0.23$\sigma$). For accrual comparability (Table 6A), the 1$\sigma$ effect for Eigenvector is –0.00348 (–0.2483 $\times$ 0.0140), $\approx$0.27$\sigma$ of CompIND_Acc (SD = 0.0128) and $\approx$10.7% of its mean magnitude ($|$mean$|$ = 0.0325).

The negative association between board connectedness and FSC supports the argument that interlocking directors can serve as channels for the transmission of aggressive or inconsistent accounting practices, consistent with research on negative contagion in board networks (e.g., Fich and Shivdasani, 2006). These directors may promote strategies they encountered elsewhere, including earnings management or strategic disclosure tailored to specific firm interests, thereby reducing consistency across firms.

Taken together, our findings provide robust empirical support for the hypothesis that well-connected boards reduce the comparability of financial statements, raising concerns about the informational quality and transparency of such firms’ financial disclosures.

This subsection examines why board-network centrality is associated with lower financial statement comparability. We evaluate four candidate channels using interaction and exposure tests and report the key patterns that align with the baseline results.

FSC_i,t = $\alpha$ + $\beta$1 CENT_i,t-1 + $\beta$2 Z_i,t-1 + $\beta$3 (CENT $\times$ Z)_i,t-1 + $\gamma$X_i,t-1 + $\mu$_j + $\tau$_t + $\epsilon$_i,t

Monitoring frictions (primary). The centrality slope is more negative when boards are busier or less independent and is weaker when audit environments are stronger. These patterns indicate that weaker oversight expands discretion in recognition and classification, reducing comparability (see Table 7).

Proprietary-cost and competition (secondary). The centrality effect is more negative in settings where rents are salient, such as high competition and high R&D intensity. These findings support a strategic differentiation motive, although the magnitudes are smaller than for monitoring or diffusion (see Table 8).

FSC_i,t = $\alpha$ + $\beta$1 CENT_i,t-1 + $\beta$2 ExpoFSC_i,t-1 + $\beta$3 ExpoEM_i,t-1 + $\gamma$X_i,t-1 + $\mu$_j + $\tau$_t + $\epsilon$_i,t

Network diffusion (primary). Exposure to lower comparability or to earnings management among director-linked firms predicts lower subsequent comparability for the focal firm. Including exposure measures reduces the absolute size of the centrality coefficient, consistent with diffusion of reporting practices through interlocks (See Table 9).

Organizational complexity (supporting). Interactions with segment breadth and foreign activities are negative and significant, indicating that greater operational scope amplifies discretion. Effect sizes are smaller than for monitoring or diffusion (see Table 10).

In sum, the evidence points to monitoring frictions and network diffusion as the main mechanisms, with proprietary-cost considerations and organizational complexity contributing to a lesser extent. These results are consistent across alternative comparability measures and remain robust to fixed effects and control sets. Additional robustness results are reported in Appendix Tables B.1,B.2,B.3.

Around years when a new interlock is created (e.g., a top-quartile eigencentral director joins), event-time regressions show flat pre-trends and a decline in FSC beginning in event year 0 that persists for two years (Appendix Table B.3).

We estimate $\mathrm{\Delta }$FSC on lagged $\mathrm{\Delta }$CENT with year fixed effects and $\mathrm{\Delta }$controls. In line with the baseline, increases in board connectedness predict subsequent declines in comparability (Appendix Table B.2).

A potential endogeneity concern is reverse causality: firms with inherently lower comparability might proactively appoint well-connected directors, rather than network connectedness reducing comparability. To address this, we first re-estimate the baseline model using lagged centrality. Specifically, we measure board-network centrality at year and regress comparability at year on this lagged measure, while retaining all other covariates at and including firm and year fixed effects. As reported in Table 11A and Table 11B, the coefficients on board connectedness remain negative and statistically significant across all network measures, indicating that greater connectedness precedes and is associated with lower comparability in the subsequent year.

Second, we augment the specification with lead terms for centrality (i.e., centrality at t+1) to probe reverse timing. If firms with declining comparability strategically appoint connected directors, future centrality should load negatively in year. We find that placebo leads are near zero and insignificant, whereas contemporaneous/lagged terms are negative and significant (Appendix Table A), which is inconsistent with reverse timing.

Third, we estimate event-time regressions around the first appointment of a highly connected director (top-quartile by eigencentrality among new appointees). Normalizing to zero, we observe flat pre-trends for and a decline in comparability beginning in the appointment year and persisting thereafter. Overall, these results jointly indicate that our findings are unlikely to be driven by firms with low comparability selectively appointing well-connected directors.

To further assess whether the economic structure of firms conditions the effect of board interlock networks on financial statement comparability, we formally examine the moderating role of operational diversity. Table 12 presents the interaction between board centrality and firm-level operational diversity (DIVERSITY_PC1). Across specifications, the interaction term (e.g., Eigenvector $\times$ DIVERSITY_PC1) is statistically significant and negative, indicating that higher operational diversity reduces the magnitude of the relation between board centrality and comparability.

The economic significance of these moderating effects is meaningful. For example, in the first specification, the coefficient estimate for the interaction term (0.0431) is approximately 31 percent of the magnitude of the main effect of Eigenvector (0.1369). Across other specifications, this ratio ranges from 5 to 32 percent, suggesting that while board networks exert a robust influence on comparability, a nontrivial portion of this effect is conditioned by firms’ operational structures. These results reveal that operational diversity does not eliminate the effect of board interlocks on comparability but meaningfully attenuates its economic impact.

To address the concern that our findings may reflect differences in firms’ underlying economic structures rather than governance mechanisms, we include SEGCOUNT (Compustat segment count) as an additional control variable in the baseline specifications. After controlling for operational diversity, the main coefficients of interest retain their sign and statistical significance (untabulated results available upon request). These supplementary tests alleviate the concern that firms with interlocking boards simply exhibit lower comparability due to having more diversified operations, rather than due to governance-related mechanisms.