Tax avoidance and audit quality: evidence from Spanish listed and non-listed companies

This study examines whether the association between audit quality and corporate tax avoidance (TA) is non-linear and heterogeneous across the TA distribution, applying a quantile regression approach to a sample of Spanish listed and non-listed firms over 2008–2019. Prior evidence on this relationship is limited and inconclusive, with sign reversals across institutional settings, firm types, and model specifications. The present study addresses this gap by jointly examining both types of firms in the same institutional setting and by documenting how the association varies across the entire TA distribution, rather than relying on a single conditional-mean estimate.

The literature on the determinants of corporate tax avoidance is extensive (Shackelford and Shevlin, 2001; Hanlon and Heitzman, 2010; Graham et al., 2012; Dyreng and Maydew, 2018; Wilde and Wilson, 2018), documenting a wide range of factors that influence corporate tax behaviour. However, the body of literature specifically examining the relationship between audit quality and TA remains comparatively limited, and the evidence is far from conclusive (Goh et al., 2013; Kovermann and Velte, 2019). For non-listed firms, Monterrey Mayoral and Sánchez Segura (2022) find a positive association between audit quality (proxied by BIG4) and tax aggressiveness in Spain, while Blaylock et al. (2025) report the opposite pattern in a cross-country sample of private firms, documenting that affiliates audited by non-BIG4 firms are more aggressive tax planners. For listed firms, Richardson et al. (2013) and Kanagaretnam et al. (2016) report a negative association, whereas other studies find no significant relation (McGuire et al., 2012; Gaaya et al., 2017). Earlier work by Maydew and Shackelford (2007) and Frank et al. (2009) already suggested that the link between audit-related factors and tax avoidance is likely to be more nuanced than a uniform positive or negative effect, given that the extent to which tax planning translates into lower reported tax expenses depends jointly on tax law, financial reporting rules, and the incentives of both the firm and its auditor.

These mixed results suggest that the association between audit quality and TA may not be homogeneous or linear and may differ across firm types. Rather than reflecting genuinely contradictory evidence, these mixed findings may partly result from the use of conditional-mean models to estimate a relationship whose sign and magnitude vary across the TA distribution. Furthermore, prior studies typically focus either on listed firms or on non-listed firms; however, evidence from Valle Ruiz et al. (2024) highlights significant behavioural differences between public and private companies, partly explained by the absence of market pressures in the latter. Our paper addresses this gap by jointly analysing listed and non-listed Spanish firms and by applying quantile regression to document how the association varies across the TA distribution. Spain offers a relevant setting for this analysis, as a developed European economy with regional tax specificities that we incorporate as controls. The sample starts in 2008 — the first year after the implementation of the Spanish accounting reform — and ends in 2019, thereby avoiding the COVID-19 distortions.

This study contributes to the limited literature on audit quality and tax avoidance in three specific ways. First, unlike prior studies that estimate only the average association, we examine whether the association between audit quality and TA varies across the TA distribution using quantile regression. This is not merely a methodological extension but follows directly from the theoretical framework developed in Section 2, which predicts that the auditor’s cost–benefit trade-off differs across levels of tax aggressiveness. Armstrong et al. (2015) demonstrated the value of this approach in tax research; our study applies it specifically to the audit quality–TA nexus, where such distributional heterogeneity is theoretically expected. Second, to the best of our knowledge, this is the first study that jointly examines listed and non-listed firms in the same institutional setting, and the results show that the association between audit quality and TA differs substantially between the two groups — a comparison that prior studies have not carried out simultaneously. Third, we incorporate auditor tenure (TENURE) as an additional empirical proxy of audit quality, alongside auditor size (BIG4). Audit quality is widely recognised as a multidimensional construct (DeFond and Zhang, 2014; Knechel et al., 2013), yet prior studies on the audit–TA relationship have focused almost exclusively on BIG4. The relationship between auditor tenure and TA has not been directly examined in the prior literature, as also noted by Kovermann and Velte (2019), partly because constructing a reliable firm-level tenure variable requires tracking the complete history of auditor–client engagements, which is considerably more demanding in terms of data collection. The joint use of BIG4 and TENURE allows us to capture two complementary dimensions of the audit quality construct: reputation and resources on the one hand, and firm-specific experience on the other (DeFond and Zhang, 2014; Tepalagul and Lin, 2015).

The paper is organized as follows. Section 2 develops the theoretical framework and derives the research hypotheses. Section 3 describes the methodology, the variables and data sources. Section 4 presents the main empirical results. Finally, Section 5 contains a summary of our main conclusions.

The relationship between audit quality and TA remains theoretically ambiguous. On the one hand, higher audit quality may constrain aggressive tax behaviour. Auditors perform an important monitoring function by enhancing the credibility of financial reporting and reducing information asymmetries between managers and external stakeholders (Richardson et al., 2013; Kanagaretnam et al., 2016). Auditors with stronger reputational concerns and greater exposure to litigation and regulatory risk are expected to be less tolerant of aggressive tax positions that could compromise reporting quality or expose the firm to future disputes with tax authorities. From this perspective, audit quality should be negatively associated with TA.

On the other hand, audit quality may also facilitate more sophisticated forms of tax planning. Auditors with greater technical expertise and deeper knowledge of the client’s operations may be better positioned to identify legitimate opportunities for tax minimisation and to assist firms in implementing complex but compliant tax strategies (McGuire et al., 2012; Chen et al., 2023). In this sense, higher audit quality does not necessarily imply stricter tax conservatism, but rather a greater ability to design and support efficient tax planning arrangements within legal boundaries. Under this alternative perspective, audit quality may therefore be positively associated with TA.

These competing arguments are consistent with the mixed empirical evidence reported in the literature (Lisowsky, 2010; Richardson et al., 2013; Kanagaretnam et al., 2016; Gaaya et al., 2017; Monterrey Mayoral and Sánchez Segura, 2022; Chen et al., 2023; Blaylock et al., 2025). Rather than reflecting contradictory findings, such evidence may indicate that the association between audit quality and TA is heterogeneous across firms and cannot be fully captured through conditional-mean estimators alone. Specifically, the association between audit quality and TA is likely to vary across the TA distribution because the cost–benefit trade-off faced by high-quality auditors is not symmetric across different levels of tax aggressiveness.

Firms with relatively low levels of TA may be better characterised as fiscally conservative or more risk-averse. Following Klassen et al. (2017), their tax outcomes may reflect a deliberate preference for tax compliance and a low tolerance for uncertain tax positions, rather than a failure to exploit available tax-planning opportunities. In this context, the reputational and litigation costs faced by high-quality auditors when supporting more assertive tax positions are relatively low, whereas the advisory benefits derived from their technical expertise could in principle be substantial. However, given the risk-averse profile of firms in this part of the distribution, it is not evident that better audit quality will systematically translate into higher levels of tax planning, as conservative firms may prefer to remain tax compliant regardless of the technical capabilities of their auditors. We therefore do not formulate a uniform directional prediction for this group of firms: the association between audit quality and TA may be weak, non-significant, or dependent on the specific TA measure considered.

By contrast, firms with relatively high levels of TA exhibit aggressive tax positions that increase regulatory scrutiny, litigation exposure, and the likelihood of future tax adjustments. In these cases, the reputational costs faced by high-quality auditors from being associated with additional tax aggressiveness are likely to outweigh the marginal benefits derived from supporting further tax planning. As a result, auditors may adopt a more restrictive stance and discourage excessively aggressive tax behaviour. Accordingly, audit quality may be negatively associated with TA among firms exhibiting relatively high levels of tax aggressiveness. This asymmetry in the auditor's cost–benefit trade-off across different levels of tax aggressiveness leads us to expect that the association between audit quality and tax avoidance will vary across the tax avoidance distribution.

This interpretation is also consistent with Armstrong et al. (2015), who document heterogeneous associations between corporate governance mechanisms and TA across the distribution of tax aggressiveness. Specifically, they report positive associations between governance characteristics and TA among firms with relatively low levels of tax avoidance, but negative associations among firms located in the upper tail of the distribution. Similarly, Boone et al. (2008) and Azizkhani et al. (2013) document non-linear relationships between auditor-related variables and other firm outcomes. The present study extends this distributional perspective to the relationship between audit quality and TA.

Additional support for this argument is provided by Frank et al. (2009), who document a strong positive association between tax aggressiveness and financial reporting aggressiveness. Their findings suggest that firms may simultaneously pursue upward earnings management and downward taxable income management when differences between accounting and tax rules permit such behaviour. In this setting, auditors face a dual monitoring role involving both financial reporting quality and tax-related risk. Aggressive financial reporting may therefore increase auditors’ scrutiny of tax positions, particularly in the presence of unusually large book–tax differences. This mechanism further reinforces the possibility that the association between audit quality and TA varies across firms with different levels of tax aggressiveness.

Related evidence provided by Erickson et al. (2004) and Lennox et al. (2013) also suggests that firms manipulating accounting earnings may avoid excessively reducing taxable income in order to minimise attention from tax authorities, auditors, and investors. More generally, these findings indicate that auditors are likely to respond differently when confronted with extreme forms of accounting or tax aggressiveness, as such behaviour may signal elevated reporting and litigation risk.

Finally, the relationship between audit quality and TA may differ between listed and non-listed firms. Listed firms operate within the capital market environment and are therefore subject to market scrutiny, demanding disclosure requirements, and close regulatory oversight, all of which may reinforce the disciplinary role of auditing. In these firms, the relevant information asymmetries primarily involve managers and dispersed capital market participants. Public visibility in capital markets also amplifies the reputational concerns faced by high-quality auditors, particularly when firms engage in highly aggressive tax strategies. More generally, the multiple monitoring mechanisms affecting listed firms may reduce variation in the association between audit quality and TA, resulting in a more homogeneous pattern across quantiles.

Non-listed firms, by contrast, are not exposed to the pressures or monitoring mechanisms specific to capital markets. In this context, the auditor may play a relatively more influential monitoring role, particularly when client-specific knowledge has accumulated through repeated engagements. At the same time, the absence of capital market visibility may reduce the reputational costs associated with moderately aggressive tax planning. As a consequence, the distributional heterogeneity of the association between audit quality and TA may be more pronounced among non-listed firms than among listed firms.

Taken together, the theoretical arguments developed above suggest that the association between audit quality and TA may vary systematically across the TA distribution. This distributional prediction provides the basis for the hypotheses developed in the next section and motivates the empirical approach adopted in Section 3.

The theoretical framework developed in the previous section suggests that the association between audit quality and tax avoidance is unlikely to be constant across firms. Rather, the association may depend on the firm's position within the tax avoidance distribution, reflecting the asymmetric cost–benefit trade-off faced by high-quality auditors across different levels of tax aggressiveness. Among firms with relatively high levels of TA, high-quality auditors are more likely to adopt a restrictive stance in order to limit exposure to regulatory and reputational risk. At the lower end, by contrast, the association is not expected to be uniformly directional, since the risk-averse profile of fiscally conservative firms may temper the potential advisory role of the auditor.

Accordingly, we formulate the following hypothesis:

In addition, the previous discussion suggests that the institutional environment in which firms operate may condition the role of audit quality in shaping tax behaviour. Listed firms operate within the capital market environment and are subject to market scrutiny and demanding disclosure and governance requirements, factors that may reinforce the monitoring role of auditors and attenuate distributional differences. Non-listed firms, by contrast, do not face capital market pressures, and the auditor–client relationship may therefore acquire greater relative importance in tax-related decisions. Consequently, the distributional pattern described above may differ between listed and non-listed firms.

This leads to the second hypothesis:

The hypotheses are formulated at the level of the audit quality construct. Section 3.1 discusses the empirical operationalisation of audit quality through the two proxies employed in this study — auditor size (BIG4) and auditor tenure (TENURE).

Consistent with the framework developed in Section 2, the empirical analysis is designed to test whether the association between audit quality and TA is heterogeneous across the TA distribution (H1) and whether this pattern differs between listed and non-listed firms (H2). The approach relies on quantile regression, following Koenker and Bassett (1978) and Koenker and Hallock (2001). As discussed in Section 2.1, we expect the association between audit quality and TA to differ across the distribution because the cost–benefit trade-off faced by high-quality auditors is asymmetric across different levels of tax aggressiveness. Quantile regression is the empirical implementation of this conceptual prediction: it estimates the association at different points of the conditional TA distribution and allows us to test whether the magnitude and statistical significance of the audit-quality coefficients vary across the distribution, as well as, in some cases, their signs. Yu et al. (2003) highlight the usefulness of quantile regression in economics and finance, and specifically in tax strategy research. However, to date, only a few studies on TA have adopted this methodology (Hsieh, 2012; Delgado et al., 2014, 2018; Armstrong et al., 2015), while other studies have considered non-linear effects through squared terms or sensitivity analyses (Delgado et al., 2023).

Since audit quality cannot be directly observed, we proxy it using auditor size (BIG4) and auditor tenure (TENURE), two complementary measures that capture the main dimensions of audit quality highlighted in the archival literature (DeFond and Zhang, 2014; Tepalagul and Lin, 2015). BIG4 reflects the auditor’s technical capacity, reputation and access to specialised tax expertise (Monterrey Mayoral and Sánchez Segura, 2022). TENURE reflects firm-specific experience derived from a sustained auditor–client relationship (McGuire et al., 2012; Goh et al., 2013). Both proxies are used jointly in all model specifications, and their coefficients are interpreted as capturing different dimensions of the same underlying construct. Although the literature has proposed additional proxies of audit quality based on financial reporting outcomes (e.g. Dechow and Dichev, 2002; Francis and Yu, 2009), we focus on BIG4 and TENURE because they are directly observable auditor characteristics that do not raise the endogeneity concerns potentially associated with reporting-based measures in a tax avoidance setting.

We estimate the relationship between audit quality and TA at nine quantiles (0.1, 0.2, …, 0.9), representing deciles of the conditional TA distribution. Quantile regression minimises an asymmetrically weighted absolute deviation rather than squared errors, making it robust to outliers and heteroscedasticity—properties particularly valuable given the well-documented skewness in tax avoidance distributions (Dyreng et al., 2008). Quantile regression is described as follows:

where Quant_θ (y_i|x_i) indicates the conditional quantile of y_i on the vector x_i. Therefore, the quantile θ (0<θ<1) solves the expression:

When quantile regression is used, linear benchmarks should be provided to assess the appropriateness of the methodology. The F Test and the Breusch–Pagan Lagrange Multiplier Test detect the presence of individual firm effects that critically influence TA, confirming that ordinary least squares (OLS) is not appropriate and that panel data methodology is required. In addition, the Wooldridge Test and the Wald Test verify the existence of autocorrelation and heteroscedasticity in our sample. Therefore, we also report PCSE estimates, following Beck and Katz (1995), which correct both problems simultaneously. This methodology has previously been used in TA research (Abubakar et al., 2021; Fernández-Rodríguez et al., 2021, 2023; Ismail and Megahed, 2022). The comparison between the two approaches allows us to show that the heterogeneity documented by the quantile regressions is systematically hidden under mean-based estimates. Following Armstrong et al. (2015), we interpret our estimates as associations rather than causal effects, given the presence of potential endogeneity concerns, which are discussed subsequently in the robustness analysis.

Model (1) presented below allows us to evaluate the two previously raised hypotheses:

where TA represents tax avoidance, for which no universally accepted definition exists in the prior literature (Hanlon and Heitzman, 2010). Therefore, it is common practice to use at least two different measures in order to enhance the validity and robustness of the results. As argued by Wang et al. (2020), the most widely used TA measures can be classified into two main categories: those based on the Effective Tax Rate (ETR) and those related to Book–Tax Differences (BTD) or the Book–Tax Gap (BTG). In our study, we use one ETR-based measure and one BTG-based measure as our primary proxies for TA. Additional measures are employed in the sensitivity analysis to further assess our findings.

Our first TA measure is based on the corporate tax payment (Cash Effective Tax Rate – CASHETR). Rather than using this variable directly, we employ its modified version (MCASHETR), which has been used in several prior studies (Amiram et al., 2013; Tang et al., 2017; Fernández-Rodríguez et al., 2021, 2023). The modified variable scales CASHETR by the Statutory Tax Rate (STR) in force in Spain, in order to account for the STR changes during the period studied. When the value of the modified variable is one, the legal rate and the effective rate coincide. If the value exceeds one, the effective rate is greater than the STR and, consequently, TA is lower. Conversely, the lower the modified variable, the more tax aggressive the firm is.

Our second measure of TA is the BTG, for which two main versions exist. The first, proposed by Manzon and Plesko (2002), calculates the difference between pre-tax book income and estimated taxable income, scaled by total assets. The second, developed by Hanlon et al. (2012), uses a logarithmic transformation of the mentioned difference. Following the approach of Monterrey Mayoral and Sánchez Segura (2022), we adopt the latter specification. Specifically, our BTG variable is computed as the logarithm of the absolute value of the difference between pre-tax book income and estimated taxable income. The estimated taxable income is obtained by dividing the total corporate income tax expense by the statutory tax rate applicable in each year of the sample (Monterrey Mayoral and Sánchez Segura, 2022). In line with the interpretive convention established in the prior literature (Comprix et al., 2011; Hanlon et al., 2012; Watrin et al., 2014; Monterrey Mayoral and Sánchez Segura, 2022), larger BTG values are generally associated with higher levels of tax planning or tax aggressiveness. However, because the measure is computed in absolute terms, it captures the magnitude, rather than the direction, of the difference between accounting and taxable income.

BIG4 and TENURE are the variables of interest. BIG4 represents auditor size and, following prior studies, is defined as a dummy variable that takes the value 1 if the company is audited by a BIG4 firm and 0 otherwise. TENURE represents auditor experience, measured as the logarithm of the number of years the auditor has remained as such in the same company (McGuire et al., 2012; Goh et al., 2013). It should be noted that audit tenure data have been obtained from SABI (Sistema de Análisis de Balances Ibéricos) including data from the earliest period available (1989).

Our model also includes the following control variables; their definitions are detailed in Appendix 1 (Supplementary Material): company size (SIZE), debt intensity or leverage (LEV), intangible fixed asset intensity (INTINT), property, plant and equipment intensity (PPEINT), inventory intensity (INVINT), return on assets (ROA), sales growth (SGROW), foreign activities (FOREIGN), domicile in a Foral territory (FORAL), domicile in the Canary Islands (CANARIAS), 2011 Consolidated Text of Audit Law (TRLAC11), 2015 Audit Law (LAC15), and GDP annual rate (GDP).

These control variables are commonly used in the prior TA literature and are mainly related to firm characteristics. In addition, we include some Spain-specific variables to control for institutional factors particular to the Spanish setting: FORAL and CANARIAS account for firms located in two Spanish territories that benefit from distinctive tax advantages (Monterrey Mayoral and Sánchez Segura, 2022), while TRLAC11 and LAC15 account for common changes in the audit environment linked to the two Spanish audit law reforms enacted during the sample period.

The sample is taken from the SABI database and, consistent with prior literature, focuses exclusively on non-financial firms. The sample comprises Spanish non-financial firms continuously audited during the 2008–2019 period. The financial information employed includes Corporate Income Tax (CIT), auditor size (BIG4 or non-BIG4) and audit tenure, all obtained from standalone financial statements. Following Monterrey Mayoral and Sánchez Segura (2022), we rely on individual accounts to avoid incorporating results from subsidiaries that are not taxable subjects of the CIT in Spain.

After applying these filters, a sample of 4,290 continuously audited companies is available, yielding 51,480 firm-year observations. We then conduct several data refining steps, eliminating observations with negative tax expense, tax payment, pre-tax income or equity, and excluding outliers in continuous variables below the 1st percentile or above the 99th percentile. The final dataset comprises 42,540 firm-year observations. Table 1 presents the sample construction procedure separately for listed and non-listed firms. The number of observations in the regression models reported in Tables 3 and 4 is smaller than the final dataset shown in Table 1 because each TA proxy requires specific variables whose availability varies across firm-year observations.

Table 2 reports mean comparison tests between listed and non-listed firms. We conducted mean difference tests using Student’s t-test with Welch’s correction for unequal variances and unbalanced sample sizes. In addition, Appendix 2 (Supplementary Material) contains the descriptive statistics for the full panel as well as separately for listed and non-listed companies, and the correlation matrix for all variables used in the model (1). Regarding the correlation matrix, it should be noted that it only reflects linear correlations and does not capture possible nonlinear associations.

The results presented in Appendix 2 (Supplementary Material) reveal substantial differences in the mean values of the two dependent variables between listed and non-listed firms. The mean comparison tests reported in Table 2 show statistically significant differences for all key variables. Specifically, non-listed firms exhibit average values of 0.887 for MCASHETR and 3.420 for BTG, compared to 0.563 and 8.599, respectively, for listed firms. Following the interpretive convention introduced in Section 3.1, these differences point to distinct tax outcomes between the two groups, with listed firms exhibiting lower effective tax burdens relative to the statutory tax rate and larger book–tax divergences than non-listed firms. Taken together, both patterns are broadly consistent with higher levels of tax aggressiveness among listed firms.

Regarding the variables of interest, the differences are also statistically significant for both BIG4 and TENURE. Listed firms are predominantly audited by BIG4 auditors (0.846), whereas the proportion is substantially lower among non-listed firms (0.261). In contrast, audit tenure is slightly higher among non-listed firms (1.822) than among listed firms (1.584), suggesting more stable auditor–client relationships in private companies.

Other notable differences between the two groups include firm size (with listed firms being larger), profitability (higher among listed companies), and location. None of the listed firms are domiciled in the Canary Islands, whereas the proportion of firms subject to the Foral tax regime is higher among listed firms.

These significant differences between listed and non-listed firms are consistent with the distinction underlying H2 and support the decision to estimate the model separately for each subsample rather than pooling both groups. Moreover, because non-listed firms substantially outnumber listed firms in the sample, pooled estimations would be dominated by the larger subsample, potentially obscuring the distinct behaviour of listed firms.

Tables 3 and 4 present our main quantile regression estimates, separately for non-listed and listed firms. In addition, Appendix 3 (Supplementary Material) provides the test of coefficient equality across quantiles using an F-Wald test based on the joint variance–covariance matrix obtained after performing the quantile regressions. This test allows us to assess whether the association between the explanatory variables and TA differs across the conditional TA distribution. Appendix 4 (Supplementary Material) presents Figures 1–4 with the coefficients by quantiles, along with those from the PCSE estimations.

For BIG4 in non-listed firms (Table 3), the MCASHETR results reveal substantial heterogeneity across the conditional TA distribution that is not captured by the PCSE estimation. Whereas the PCSE coefficient suggests a uniformly positive association between BIG4 and MCASHETR, the quantile regressions show that this association is statistically significant only in the upper quantiles (0.6–0.9). In the lower quantiles, where firms exhibit more aggressive tax behaviour, the coefficients are not statistically significant. The positive and increasing coefficients observed in the upper quantiles indicate that BIG4 is associated with higher effective tax burdens among firms that are already relatively conservative in their tax practices. F-Wald tests confirm that the coefficients differ significantly across quantiles.

These results are consistent with the idea that the association between BIG4 and tax behaviour varies depending on the firm’s position within the TA distribution. In particular, the evidence suggests a more compliance-oriented association among conservative firms, consistent with the reputational concerns discussed in Section 2.1, whereas no significant association is observed among the most aggressive firms.

Using BTG as the dependent variable, the results exhibit a different pattern. The coefficients are positive and statistically significant for quantiles 0.4–0.9, indicating that BIG4 is associated with larger book–tax gaps among firms located in the upper part of the BTG distribution. F-Wald tests again reject coefficient equality across quantiles, confirming the presence of distributional heterogeneity.

Considering MCASHETR and BTG jointly, the results suggest that the association between BIG4 and TA differs depending on the dimension of tax behaviour being captured. For firms located in the upper MCASHETR quantiles, BIG4 is associated with higher tax payments relative to the statutory tax rate. For firms located in the upper BTG quantiles, BIG4 is associated with larger book–tax gaps. Under the interpretive convention introduced in Section 3.1, this positive association would generally be interpreted as evidence of higher levels of tax planning activity among clients audited by BIG4 firms. This reading, however, must be qualified: as noted in Section 3.1, BTG captures the magnitude rather than the direction of the difference between accounting and taxable income, so the upper BTG quantiles should not be interpreted as unequivocally representing the most tax-aggressive firms. As discussed by Frank et al. (2009), larger book–tax gaps may reflect the interaction between tax planning and financial reporting incentives, such that auditors' technical expertise may be associated with wider gaps through the flexibility granted by legislation in certain areas, such as depreciation.

Overall, the evidence provides support for H1 in non-listed firms, although the pattern does not correspond to a simple sign reversal across quantiles. Rather, the results point to a more nuanced form of heterogeneity that depends both on the firm’s location within the TA distribution and on the specific TA measure employed. Importantly, this heterogeneity is largely obscured under mean-based estimators, as reflected by the comparatively uniform PCSE results.

For listed firms (Table 4), the MCASHETR results show a markedly different pattern. BIG4 shows a broadly homogeneous positive association with MCASHETR across most of the distribution, rather than the heterogeneous pattern documented for non-listed firms. Moreover, the F-Wald tests generally fail to reject coefficient equality across quantiles, indicating limited heterogeneity in the association between BIG4 and TA for listed firms.

This contrasts sharply with the evidence for non-listed firms and is consistent with the theoretical framework developed in Section 2.1. Specifically, the market scrutiny and regulatory visibility faced by listed firms may strengthen the reputational incentives of BIG4 auditors, resulting in a more homogeneous and conservative association with tax behaviour across the distribution.

Using BTG as the dependent variable, the evidence is substantially weaker. The PCSE coefficient is not statistically significant, and most quantile coefficients are likewise insignificant, with the exception of the upper quantiles (0.8–0.9). In addition, the F-Wald tests do not reject equality across quantiles.

This distinction is relevant because MCASHETR and BTG capture different dimensions of tax behaviour (Hanlon and Heitzman, 2010; Lisowsky et al., 2013). MCASHETR reflects taxes effectively paid—a highly visible metric subject to intense scrutiny by capital markets, analysts, and tax authorities (Dyreng et al., 2008). BTG, by contrast, incorporates deferred taxes and accrual-related components that may be less directly observable and less closely linked to immediate tax payments. The evidence therefore suggests that, in listed firms, the association between BIG4 and TA is concentrated in the most visible dimension of tax behaviour, consistent with BIG4 auditors prioritising reputational risk mitigation in the most observable component of tax outcomes (Francis and Wang, 2008). Accordingly, H1 is not supported for listed firms. BIG4 shows a broadly homogeneous positive association with MCASHETR across most of the distribution, rather than the heterogeneous pattern documented for non-listed firms.

Turning to the second proxy of audit quality, for TENURE in non-listed firms using MCASHETR (Table 3), the results show that audit tenure is not significantly associated with TA at any quantile. Therefore, TENURE does not appear to be related to the payment of corporate income tax in private firms.

However, for BTG the results reveal a heterogeneous distributional pattern. At the lowest quantile (0.1), TENURE is positive and significant, indicating that longer tenure is associated with larger book–tax gaps among firms located in the lower part of the BTG distribution. In the middle quantiles, the coefficients are not statistically significant, representing a transition zone. At the upper quantiles (0.8–0.9), TENURE is negative and significant, indicating that longer tenure is associated with smaller book–tax gaps among firms exhibiting the largest book–tax divergences. F-Wald tests reject coefficient equality across quantiles and reveal statistically significant differences between the extremes.

These results provide support for H1 and are consistent with Armstrong et al. (2015), who document similar non-linear patterns for board tenure and tax avoidance. Under the conventional interpretation adopted in the prior literature, these results would suggest that longer auditor tenure is associated with greater tax-planning activity among firms in the lower part of the distribution and with lower tax-planning activity among those in the upper part. However, given the absolute-value construction of BTG, the two ends of the distribution cannot be unequivocally identified with conservative and more tax-aggressive firms, respectively. The sign reversal should therefore be interpreted primarily as evidence of the distributional heterogeneity anticipated by the framework developed in Section 2.1 and obscured by mean-based estimators.

For listed firms, TENURE shows no significant association with TA in most quantile regressions, with either TA measure. The only exception is MCASHETR at quantile 0.9, where the coefficient is marginally significant at the 10% level. F-Wald tests generally do not reject equality of coefficients, except for a marginal difference between quantiles 0.90 and 0.50. These results suggest that auditor tenure has a limited association with TA for listed firms. This finding is consistent with the theoretical argument in Section 2.1 that listed firms, which operate within the capital market environment, may be less influenced by the auditor–client relationship, as their tax strategies are shaped by multiple stakeholders rather than primarily by auditor-specific knowledge.

Taken together, the evidence provides support for H1 among non-listed firms. For this subsample, both proxies of audit quality exhibit distributional heterogeneity in their association with TA. BIG4 shows a pattern that varies across the two TA measures, in line with the interpretive considerations discussed above, while TENURE–BTG displays the clearest sign reversal across the distribution. In both cases, the PCSE estimates fail to capture this heterogeneity, yielding non-significant or uniformly positive coefficients that conceal the distributional variation documented by the quantile regressions. For listed firms, by contrast, neither BIG4 nor TENURE exhibits significant heterogeneity across the TA distribution, and H1 is not supported. The sharp contrast between the two subsamples provides support for H2: the association between audit quality and TA across the distribution differs substantially between listed and non-listed firms, with the non-linear patterns being more pronounced among private firms—consistent with the absence of capital market pressures and greater reliance on the auditor–client relationship anticipated in Section 2.1.

Lastly, several robustness tests are presented in Appendix 5 (Supplementary Material). The models have been re-estimated using two additional measures of TA—BTD (Manzon and Plesko, 2002; Monterrey Mayoral and Sánchez Segura, 2010) and DTAX (Frank et al., 2009)—and, for listed firms, using three quantiles (0.25, 0.50, and 0.75) given the smaller sample size. The results are broadly consistent with the main findings. Finally, similar to Armstrong et al. (2015), we acknowledge that our estimates are descriptive and do not necessarily imply causality. Although we discuss the main endogeneity concerns in Appendix 5 (Supplementary Material), reverse causality and correlated omitted variables cannot be fully ruled out. Nevertheless, quantile regression allows us to uncover patterns that would remain hidden under mean-based approaches, providing valuable insights into firm behaviour at different levels of tax aggressiveness.

This paper examines whether the association between audit quality and corporate tax avoidance (TA) is heterogeneous across the TA distribution (H1) and whether this association differs between listed and non-listed firms (H2). Using quantile regression on a sample of Spanish firms over 2008–2019, we operationalise audit quality through two complementary proxies—auditor size (BIG4) and auditor tenure (TENURE)—and compare the distributional patterns observed across both types of firms. Our results indicate that the association between audit quality and TA cannot be adequately characterised through mean-based estimators. The quantile regressions reveal substantial heterogeneity across the conditional TA distribution, particularly among non-listed firms, whereas the corresponding PCSE estimations often conceal these patterns.

For non-listed firms, the BIG4 results suggest that the association between auditor size and TA differs depending on both the firm’s position within the distribution and the dimension of tax behaviour being captured. BIG4 is associated with higher effective tax burdens among conservative firms (upper MCASHETR quantiles), whereas no significant association is observed among the most aggressive firms. For BTG, BIG4 is associated with larger book–tax gaps in the upper quantiles. Under the conventional interpretation of BTG, this positive association would be consistent with higher levels of tax planning or tax aggressiveness. However, because BTG captures the magnitude rather than the direction of the difference between accounting and taxable income, these quantiles should not be interpreted as unequivocally representing the most tax-aggressive firms. The observed pattern may reflect more complex book–tax positions arising from the interaction between accounting and tax reporting incentives. These findings support H1 for non-listed firms, although the pattern does not correspond to a simple sign reversal but rather to a more nuanced form of heterogeneity. For listed firms, by contrast, BIG4 is associated with higher tax burdens across most of the distribution, with limited heterogeneity across quantiles. H1 is therefore not supported for listed firms.

Regarding TENURE, the results further differentiate the two subsamples. For non-listed firms, TENURE–BTG displays the clearest sign reversal across the distribution: longer tenure is associated with larger book–tax gaps at the lower end of the distribution and with smaller gaps at the upper end. Under the conventional interpretation of BTG, this pattern could suggest that the auditor’s client-specific experience facilitates tax planning among firms with relatively low levels of TA, while constraining it among firms with relatively high levels of TA. However, this interpretation is subject to the limitation concerning the absolute-value construction of BTG discussed above. For listed firms, TENURE shows no significant association with TA, suggesting that the multiple monitoring mechanisms associated with stock market participation may reduce the relative influence of auditor–client tenure. The contrast between the two subsamples provides support for H2.

Our findings contribute to the literature in several ways. First, we show that the association between audit quality and TA is heterogeneous across the TA distribution among non-listed firms (H1) and therefore cannot be fully characterised using conditional-mean approaches alone. Rather than focusing on whether audit quality increases or decreases TA on average, we examine how this relationship varies across firms with different levels of tax aggressiveness. In doing so, we provide evidence that part of the mixed findings reported in the prior literature may stem from the use of mean-based estimators to analyse a relationship that is inherently heterogeneous across firms, with such heterogeneity being particularly pronounced among non-listed firms and largely absent among listed firms. Second, whereas prior studies typically focus either on listed or on non-listed firms, we jointly examine both groups within the same institutional setting and document substantial differences between them (H2). Third, we complement the traditional BIG4 proxy by incorporating auditor tenure as an additional dimension of audit quality related to the auditor’s client-specific experience accumulated through the auditor–client relationship, a variable that has received limited attention in the prior literature partly due to the complexity of its construction.

Our findings have several implications. For regulators and policymakers, the results indicate that the association between audit quality and tax behaviour differs between listed and non-listed firms. Regulatory assessments should therefore avoid assuming that audit-related mechanisms operate uniformly across the two groups. For audit firms and practitioners, the evidence that the association between audit quality and TA varies across the distribution highlights the importance of adapting audit procedures and risk assessments to the client's tax profile, particularly in non-listed firms where the auditor–client relationship appears to play a more influential role. For researchers, our study confirms that distributional methods can reveal patterns that mean-based approaches systematically obscure and demonstrates the applicability of quantile regression to the audit quality–TA nexus.

This study is subject to several limitations. First, as in Armstrong et al. (2015), our estimates should be interpreted as descriptive associations rather than as causal effects, since reverse causality and omitted-variable concerns cannot be fully ruled out. Second, audit quality is a multidimensional construct (Knechel et al., 2013; DeFond and Zhang, 2014), and our two proxies capture only specific dimensions; the incorporation of additional proxies based on financial reporting outcomes—such as abnormal accruals (Dechow and Dichev, 2002) or audit report characteristics (Francis and Krishnan, 1999)—could provide a more comprehensive picture, provided that the potential endogeneity challenges are adequately addressed. Third, the relatively small size of the listed-firm sample may reduce statistical power and should be taken into account when interpreting the comparison between listed and non-listed firms. The additional three-quantile robustness analysis helps to address this limitation. Finally, our evidence is limited to the Spanish institutional context, and cross-country analyses would allow testing whether the distributional patterns documented here generalise to other jurisdictions with different audit and tax regulatory environments.

Overall, the results highlight the usefulness of quantile regression for uncovering forms of heterogeneity that remain hidden under conventional mean-based approaches. More broadly, the evidence suggests that the association between audit quality and tax avoidance is more nuanced and context-dependent than previously assumed in the literature.

The supplementary material for this article can be found online.