AUDIT COMMITTEE DISCLOSURE TONE AND EARNINGS MANAGEMENT

STRUCTURED ABSTRACT Purpose: This paper examines the effect of audit committee reporting, measured by the tone of audit committee disclosures, in improving financial reporting quality as proxied by earnings management. Design / Methods We focus on the textual properties of AC reports, particularly the tone of AC disclosure, and their impact on financial reporting quality proxied using real and accruals-based earnings management. For additional analysis, we use a financial reporting index and matched sample. Our analysis is based on a sample of UK FTSE 350 firms. Findings: Our analysis suggests that audit committee reports are not boilerplate but varied in language. We find audit committee reporting is negatively associated with both real and accruals-based earnings management. In our additional tests, we find a positive association between financial reporting quality index and reporting tone. Originality / value:


INTRODUCTION
Corporate governance can affect the quality of financial reporting.While there is evidence to suggest that governance structures reduce earnings management, there is limited evidence on how audit committee (AC) reporting affects financial reporting quality.Guidelines on AC role and responsibilities in the UK require ACs to report on significant issues considered in relation to financial reports and auditing (FRC, 2012).
This paper focuses on the textual properties of AC reports, particularly the tone of AC disclosures, and examines the effect of AC reporting in improving financial reporting quality as proxied by earnings management.
Recognizing that reporting by ACs in relation to their activities may impact on the quality of financial reporting, we draw on the literature on textual properties and focus on the tone of AC reports.Our paper extends the literature which has mainly investigated the relationship between ACs characteristics and financial reporting quality.For example, studies show that AC expertise enhances earnings quality (Beasley et al., 2009;Bedard and Gendron, 2010;Cohen et al., 2013).A number of studies finds a positive association between AC size and the quality of financial reporting (e.g., Mangena and Pike, 2005;Li et al., 2012;Mangena and Tauringana, 2008).Moreover, previous research has established the importance of active AC oversight of the financial reporting process for improving the quality reporting (Beasley et al., 2009;Zaman et al., 2011).Our paper also complements qualitative studies, mainly based on interviews with AC participants in small number of companies, on the AC process showing that AC actions play a key role in configuring meanings of effectiveness (Gendron and Bédard, 2006;Turley and Zaman, 2007;Beasley et al., 2009;Compernolle, 2018;Khemakhem and Fontaine, 2019).It also adds to the literature on AC disclosures (e.g., Guttierrez et al., 2018;Lennox et al., 2018 andCzerney et al., 2019;Sahyoun and Magnan, 2020) through a focus on the textual properties of AC reports, particularly the tone of AC disclosures, and examining the effects on financial reporting quality as proxied by earnings management.
Our analyses based on a sample of UK FTSE 350 firms support our hypothesis that the tone of AC report disclosures is negatively associated with earnings management and suggest the tone of AC disclosures, as expressed in the choice of words used in describing AC engagement, conveys information that affects the communication function of AC reporting thereby improving reporting quality.Our analysis indicates that AC reports are not boilerplate but varied in language choice which contributes to improving financial reporting quality.Moreover, our findings also hold when we control for endogeneity.When we use alternative proxies for financial reporting quality, we find that AC disclosure tone has a significant and negative association with accrual-based earnings and a positive and significant association with the financial reporting quality index.
In contrast to the predominant focus in existing studies on the US setting, our study is based on the UK context.The UK provides a meaningful and useful context for our study.First, as noted by Ghafran and O'Sullivan (2013), evidence on the financial reporting effects of ACs in the UK setting is quite limited.Key exceptions include: Peasnell et al. (2005), Mangena and Pike (2005), Mangena and Tauringana (2008) and Li et al. (2012).Second, compared to US regulations on corporate governance which are relatively rigid, the Financial Reporting Council (FRC) in the UK adopts a softer, comply or explain regulatory approach (FRC, 2012b).Third, the higher risk of litigation can inhibit the management of US firms from making detailed voluntary disclosures about governance practices (Francis et al., 1994;Baginski et al., 2002).Whereas, under the UK's flexible governance environment, risk of litigation from shareholders or other stakeholders is relatively low (Clatworthy and Jones, 2003;Black et al., 2005).
Though the importance of AC disclosures in improving reporting quality is well recognised in policy guidelines and governance recommendations, to the best of our knowledge, no study has employed computer-based textual analysis of AC reports and investigated the effect of AC disclosure tone and the role it can play in achieving higher reporting quality.Overall, our paper provides baseline evidence for future research and policy making and reveals that ACs reporting what they have done increases transparency and impacts on reporting quality.
The reminder of the paper is structured as follows.Section 2 summarizes prior literature.Section 3 develops the hypothesis.Section 4 describes the research method including model specification, variable measurement, and the sample.Section 5 presents the findings while in section 6 we provide a summary and conclusion.

LITERATURE REVIEW
The theoretical underpinning for ACs is based on agency theory which leads to an expectation that effective ACs will enhance governance and the quality of financial reporting and auditing (Turley and Zaman 2007;Ghafran and O'Sullivan, 2017).Beasley et al. (2009) argue that agency theory is relevant in enhancing our understanding of the role of governance in the audit process since the audit is focused on financial reporting and internal control.Agency theory suggests that factors that create the need for closer monitoring of management produce the need for effective ACs (Collier and Zaman, 2005;Turley and Zaman, 2007;Zhou et al., 2018).From an agency perspective firms with high information asymmetry and agency costs will be inclined to reduce such costs by providing substantive oversight and disclosure relating to governance (Piot and Janin, 2007).Earnings management intended to mislead stakeholders about the underlying economic performance of the firm can compromise reporting quality.A primary purpose of ACs is thus to constrain earnings management.Overall, prior research suggests that ACs are associated with improved financial reporting (Zaman et al., 2011;Ghafran and O'Sullivan, 2013;Badolato et al., 2014;Khemakhem and Fontaine, 2019).ACs reporting publicly on significant issues considered in relation to financial reporting and auditing, and how these issues have been addressed can increase the transparency of their monitoring.Moreover, it may also intensify AC substantive oversight and thereby improve quality of reporting.
A focus on AC reporting enables us to respond to the call for more research, inter-alia, by Beasley et al. (2009: p113): "of particular importance is whether variations in process are associated with variations in financial reporting and governance outcomes, above and beyond previously documented relations between AC characteristics and financial reporting outcomes".Our focus also resonates with reviews of AC literature by Bedard and Gendron (2010) as well as Broye and Johannes (2021) who suggest that despite the existing literature focusing on AC characteristics, our understanding of their effects is still limited and call for more research that helps to increase understanding of ACs.McNulty et al. (2013) also argue that examining AC attributes alone fails to capture what really goes on in ACs.Our focus on AC reporting thus complements research largely focused on AC attributes.

HYPOTHESIS DEVELOPMENT
Research on textual analysis of corporate disclosures focuses on textual analysis of the readability and tone dimensions of annual reports (see Li, 2008;Lehavy et al., 2011;Loughran and McDonald, 2014;Ajina et al., 2016).Several studies examine whether companies use tone in qualitative disclosures to convey fundamental information.
Bicudo de Castro et al. (2019) find that annual reports that convey positive tone are associated with lower audit fees.Davis et al. (2012) find that the increase in tone optimism of earnings press releases leads to an immediate stock price response to earnings announcement and Davis and Tama-Sweet (2012) argue that managers have incentives to use tone strategically in their qualitative disclosures and find that tone pessimism predicts poor future performance.Huang et al. (2013) find that abnormal positive tone misleads investors and reveals managerial incentives to misinform investors about future performance.Bassyouni et al. (2020) report that firms with high level of board independence have less positive tone in their narrative reporting.Some recent auditing literature focuses on changes in auditors' reports and their informativeness.Textual analysis of audit reports is however limited and at an early stage.Gutierrez et al. (2018) investigate the UK requirement to expand auditors' reports and find that mere increase in disclosures is not associated with increase in audit fees or informativeness.Similarly, Lennox et al. (2018) examine the informativeness of new reporting standards in UK that require auditors to disclose the risks of material misstatements and find these disclosures lack incremental information content, suggesting that the expanded audit reports do not provide investors with additional value-relevant information.In contrast to Gutierrez et al. (2018) and Lennox et al. (2018), Reid et al. (2019) examine the impact of UK's auditor reporting changes on financial reporting quality and find that the new auditor reporting requirements are associated with a significant improvement in financial reporting quality.Czerney et al. (2019) examine the explanatory language added to unqualified audit reports and how investors respond to such language.They find that investors generally do not respond to explanatory language at the time of the audit report release.Sahyoun and Magnan (2020) analyse AC disclosures based on recommendations from multiple industry and governance organisations' reports and its association with banks' earnings management and find that the voluntary disclosures in AC reports are used as a tool for impression management.
There is scarcity of research focusing on the textual analysis of AC reports.We are able to identify one unpublished article that focuses on the textual analysis of AC reports.Draeger et al. (2019) undertake textual analysis of US firms' AC reports and find that AC reports have become shorter over time and include fewer voluntary disclosures.The study finds no evidence that AC characteristics are associated with the level of disclosure in AC reports.Draeger et al. (2019) argue that AC reports typically use boilerplate language and present limited information on AC oversight of the external auditor.Samaha et al. (2015) in their review note that audit committees play a pivotal role by improving the quality of information disclosed.To the best of our knowledge, no previous study has examined the tone of AC reports and their impact on financial reporting quality.The language used in AC reports can give an insight into the nature of communication to the user (Coram et al., 2011).For example, if there is additional disclosure relating to risks of material restatements or going concern risk, the tone of information disclosed in AC reports is likely to be negatively affected.The linguistic content of the corporate reports (including AC reports) is helpful in explaining firms' reporting quality (Loughran and McDonald, 2011).Therefore, we expect AC disclosure tone that conveys information which enhances users' understanding of the audit process will affect the communication function of AC reporting and increase the transparency of AC monitoring and thereby improve financial reporting quality as proxied by earnings management.
Overall, the tone of AC reporting is important because ACs help to create and maintain a reporting system and environment within firms that supports the integrity of the financial reporting process.AC disclosure tone can signal their attention and sharpen their focus on certain financial reporting items such as accounting adjustments and revenue recognition.AC disclosures with constrain tone to accounting figures (e.g., revenues recognition) can help increase reporting quality through lower earnings management.Moreover, AC disclosures with litigious tone relating to accounting entries (e.g., recording a provision) may help increase overall earnings quality.Our hypothesis is: Hypothesis.The tone of audit committee report disclosure is negatively associated with earnings management.

Sample and Data
The sample for our study consists of FTSE 350 companies listed on the London Stock Exchange (LSE) during 2013 and 2014.The FTSE 350 represents highest market capitalization and are the centre of attention for investors, regulators and professional bodies.The time period is appropriate for the study as the AC reporting patterns over the period of two years since the issuance of the FRC's revised guidelines for ACs published in 2010 means sufficient time has lapsed for companies to achieve compliance. 1 Financial data is collected from annual reports and DataStream.We remove financial firms and companies operating in the utility industry because these firms have unique regulatory environment.We also loose a few observations due to missing data for some of the financial variables.Table 1 shows how our sample which begins with 700 firm-years and ends with a final sample of 540 firm-years.

Earnings Management Measures
Following prior studies our main proxy of earnings management is the real earnings management metrics (EM_RM).To capture the total effects of real earnings management, we follow Cohen and Zarowin (2010) and use an aggregate measure of real earnings management calculated as the sum of abnormal discretionary expenses multiplied by negative one (so that the higher amount, the more likely it is that the firm is cutting discretionary expenses) and abnormal production costs (increasing production to spread the fixed costs of production over a large number of units). 2 (See untabulated Appendix 1 for detailed calculations of earnings management metrics).
Following Cheng et al. (2016), we also use the performance-matched real earnings management (EM_PRM) as a proxy for financial reporting quality, where real earnings management is adjusted for the performance of a matched firm.
1 The FRC revised guidelines recommend that the core functions of audit committees are expressed in terms of 'oversight', 'assessment' and 'review' of a particular function (FRC, 2010).
2 In untabulated analysis we include an aggregate measure of real earnings management that is equal to the sum of abnormal cash flows and abnormal discretionary expenses, both multiplied by negative one (so that the higher the values the more likely the firm is accelerating sales using aggressive price discounts and/or lenient credit terms and reducing the amount of discretionary expenses).Our inferences are qualitatively similar to those reported when we use this alternative measure, and therefore, for parsimony, we report the results using one aggregate measure of REM.
Additionally, we use the accrual-based earnings as an alternative measure of financial reporting quality.Prior studies often rely on the Jones Model (Jones, 1991) or the modified Jones Model (Dechow et al., 1995) to calculate accrual-based earnings.
We use the modified Jones Model to measure the accrual-based earnings (EM_MJM) (untabulated Appendix 1).3 Prior research on accruals-based earnings suggests that discretionary accrual models might be misspecified when applied to firms with extreme financial performance (e.g., Dechow et al., 1995;Kothari et al., 2005).The discretionary accrual measure is also adjusted for the accrual performance of a matched firm where matching is on the basis of return on assets and industry.
Following Kothari et al. (2005) we also estimate performance-matched earnings management proxy (EM_PEM).4Finally, we compute an index (FRQ_index) as an additional proxy by taking the standardised averages of EM proxies used in this study, i.e., EM_RM, EM_PRM, EM_MJM, and EM_PEM and multiply them by minus one so that it is increasing with reporting quality.5

AC Reporting Tone
In measuring the disclosure tone of AC reports we draw on prior literature that has focused on the textual analysis of financial reports. 6Various methods have been used to measure qualitative disclosures, such as the naïve Bayesian machine-learning algorithm (Li, 2010), General Inquirer and Diction (Tetlock, 2007;Tetlock et al., 2008;Kothari et al., 2009), and Linguistic Inquiry and Word Count (LIWC) (Li, 2008;Draeger et al., 2019;Smith, 2019).These methods are based on using word dictionaries that have been transferred from the field of social psychology into the finance context (Kang et al., 2018).Loughran and McDonald (2011) argue that these psychological dictionaries are not suitable for financial disclosures and suggest the use of financialcustomised word list to better reflect the finance context.The authors created financebased word lists to describe the negative, positive, litigation, constrain, strong and weak tones in financial disclosures.
Prior research on corporate disclosure has used various textual analysis measures to investigate the text-based information released by firms including readability (Li, 2008;Ajina et al., 2016;Kim et al., 2017;Lo et al., 2017;Luo et al., 2018), positive vs. negative tone (Davis et al., 2012), and aggregate tone (Huang et al., 2013;Kang et al., 2018).Kang et al. (2018) argue that when investors read financial reports and make investment decisions, they consider good and bad events alike.Therefore, separating the effects of positive and negative text-based information is more likely to lead to mixed findings.Purda and Skillicorn (2015) argue that linguistic analysis can be used effectively to detect unusual discrepancies in financial reporting.Their study uses language-based technique including the litigious and negative words lists form Loughran and McDonald (2011) as a tool for corporate fraud detection.Moreover, Bodnaruk et al. (2015) use textual analysis to measure the extent to which a firm is 2010, You and Zhang, 2009;Ertugrul et al., 2017;Kang et al., 2018), earnings press release (Davis et al., 2012;Davis and Tama-Sweet, 2012;Huang et al., 2013), media news (Tetlock, 2007;Tetlock et al., 2008), management discussion and analysis section of the annual report (Lo et al., 2017), and conference calls (Larcker and Zakolyukina, 2012).
financially constrained and create a constraining word list similar to the one used in Loughran and McDonald (2011).The study argues that companies facing financial challenges are more likely to use constraining tone in their 10-K fillings to communicate their concerns to shareholders, thereby reducing litigation exposure.
We proxy for AC tone using three alternative measures: ACT_Litigious, ACT_Constrain and ACT_LIWC.We use the litigation and constrain word dictionaries created by Loughran and McDonald (2011) in our main tests similar to Bodnaruk et al. (2015) and Purda and Skillicorn (2015).2018), we define the tone as the frequency difference between the numbers of positive and negative words divided by the total word count using LIWC custom dictionary of positive and negative word lists. 9

Control Variables
In testing our models, we control for AC, boards of directors and firm characteristics.
We use ACStrength, similar to Zaman et al. (2011), to measure the strength of an AC computed by totalling the proxies of five AC characteristics.We measure the components using dummy industry-adjusted variables to enable the calculation of the composite measure of AC strength.Accordingly, all variables are defined in Table 1.
Similar to our measurement of AC strength, we use a composite measure for the strength of the corporate board, i.e., BODStrength computed by totalling the proxies of six board characteristics.
Prior studies find firm-specific characteristics can affect earnings management we thus control for firm-specific variables.These are: firm size (SIZE) measured by the natural logarithm of total assets; firm age (FAGE) measured by the number of years since a firm has been listed on the London Stock Exchange; Business complexity (BSEG) calculated as the logarithm of the number of operating segments of the firm; leverage (LEV) measured by the ratio of total liabilities scaled by total assets; ownership concentration (OWN) measured by the sum of shares held by block-holders of greater than 5% divided by the total number of shares in issue; and firm profitability (ROA) measured by net income before extraordinary items divided by total assets; a proxy for liquidity (CFO) which represents cash flow from operating activities divided by total assets; the annual growth of sales (Sales_growth) calculated as the change in sales from previous year scaled by sales at the beginning of the year; and industry and year dummies.

Empirical Model
We use the following model to estimate the impact of AC report tone (ACTone) on earnings management (EM).Since earnings management must occur prior to the financial year-end, and AC report is written after the financial year-end (Zang, 2012), we expect that the tone of AC reports to be associated with earnings management in the following financial year.Therefore, we apply one-year lag between our dependent and independent variables.The measurements are described separately in sections 3.2 and 3.3, and the control variables are described in section 3.4.All variables are defined in Table 2.

Descriptive Statistics and Correlations
Table 3 presents the descriptive statistics for the variables in our model.The mean values of real earnings management variable (EM_RM) is -0.0328 which is higher than the mean value of -0.006 reported in Cheng et al. (2015), and the mean value of The mean value of the aggregate tone measure (ACT_LIWC) is 1.401 which is higher than the mean value of 0.901 reported in Smith (2019) based on the positive and negative words in the LIWC word lists.
Table 4 reports the Pearson correlation for the variables in our main analysis we find that ACT_Litigious, ACT_Constrain are negatively associated with real earnings management (EM_RM) and correlation coefficients are significant at the 5% level.ACT_LIWC is also negatively associated with EM_RM, the correlation coefficient is significant at 10% level.These coefficients suggest a negative correlation between AC disclosure tone and earnings management.ACStrength and BODStrength are also negatively correlated with earnings management.None of the correlations between control variables are high enough to raise a multicollinearity issue as the variance inflation factor (VIF) values range from 1.07 and 2.75 With a mean value of 2.17.
[Table 3 and 4 about here]

AC Report Disclosure Tone and Real Earnings Management
Table 5 presents our main findings on the impact of AC disclosure tone and financial reporting quality measured by real earnings management (EM_RM) and performance-matched real earnings management (EM_PRM).Models 5.1 and 5.2 use ACT_Litigious, Models 5.3 and 5.4 use ACT_Constrain, and Models 5.5 and 5.6 use ACT_LIWC as the AC disclosure tone measure.Our results show that ACT_Litigious has a negative and significant association with EM_RM at 5% level and at 5% level with EM_PRM.ACT_Constrain has a negative and significant association with EM_RM at 1% level and with EM_PRM at 5% level, and ACT_LIWC has a negative and significant association with EM_RM at 5% level and with EM_PRM at 10% level. 10he results are also economically significant, where economic significance is computed following Huang et al. (2018).For example, Model 5.1 shows that the coefficient of -0.0381 for ACT_Litigious indicates that moving from the first quartile (0.220) to the third quartile (0.500) of ACT_Litigious can decrease a firm's real earnings management by 1.067%.Similarly, in Model 5.3 the coefficient of -0.031 for ACT_Constrain indicates that moving from the first quartile (0.350) to the third quartile (0.650) of ACT_Constrain can decrease a firm's real earnings management by 0.933%.
In Model 5.5, the coefficient of -0.0114 for ACT_LIWC indicates that moving from the first quartile (1.224) to the third quartile (1.542) of ACT_LIWC can decrease real earnings management by 0.362%.Therefore, consistent with our hypothesis, we find that the tone of AC disclosures is associated with lower level of real earnings management activities.
Our results also show that ACStrength has a negative and significant association with real earnings management metrics at 1% level for Models 5.1-5.6.These findings are consistent with the view that ACs mitigate earnings management by supervising major accounting choices and supporting correction of discovered errors and generally improving quality of reporting and auditing (Klein, 2002;Xie et al., 2003;Piot and Janin, 2007;Zaman et al., 2011;Libby et al., 2015).Among other control variables, we find LEV has a significant and positive association, and Sales_growth has a significant and negative association with real earnings management proxies indicating that both capital structure and growth opportunities are likely to affect real earnings management (Cheng et al., 2016;Katmon and Al Farooque, 2017).
Prior literature argues that REM is less subject to auditor scrutiny (e.g., Kim and Park, 2014).The real economic actions that a company engage in to meet specific earnings targets is difficult to confront by auditors or regulators (Graham et al., 2005).
We investigate whether ACs are responding to aggressive REM by disclosing activities that capture the underlying risks of their firms, where ACs and auditors may perceive lower-magnitude REM as ordinary business operations.We divide the sample into firms that are the top quintile REM (i.e., higher-magnitude REM) and firms at the lower quintile REM (i.e., lower-magnitude REM) (see Greiner et al. 2017 for a similar approach).Our untabulated results show that for subsample with highermagnitude REM, ACT_Constrain has a negative and significant association with EM_RM at 10% level, and with EM_PRM at 1% level, and ACT_LIWC has a negative and significant association with EM_RM at 5% level, and with EM_PRM at 1% level.
For subsample with lower-magnitude REM, AC disclosure tone measures do not show an association with REM proxies suggesting that ACs may perceive lowermagnitude REM as ordinary business operations and their reporting is more likely to have an effect in constraining higher-magnitude REM rather than lower-magnitude REM.Interestingly, ACStrength has a negative and significant association with real earnings management metrics for lower-magnitude REM subsample, but not for higher levels REM, suggesting that aggressive REM requires additional substantive oversight and AC engagement through AC disclosure activities that capture the increased business risks, more than specific features of ACs.
Our paper contributes to the textual analysis literature.Prior studies explore how the readability of annual reports, used as an aspect of textual analysis, varies with earnings management (Ajina et al., 2016;Lo et al., 2017) and shows that firms most likely to engage in earnings manipulation have difficult to read annual reports.
Whereas we focus on the tone of a document, in particular of AC reports, and its association with earnings management.Overall, the evidence in Table 5 suggests that the tone of AC disclosures, as expressed in the choice of words used in describing AC engagement, conveys information that affects the communication function of AC reporting thereby improving reporting quality.ACs disclose activities that capture the underlying risks of their firms by utilizing litigious and constrain words in the AC reports.Such disclosure increases the transparency of AC monitoring and enhances the quality of reporting.Our analysis of AC report tone suggests that AC disclosures are not boilerplate but varied in language choice, and that this is associated with reduced earnings management and hence higher financial reporting quality.

Alternative measures of earnings management
We conduct additional analysis using alternative proxies for earnings management.
In untabulated test, we use accrual-based earnings management and financial reporting quality index (FRQ_Index) computed as the standardized average of the four proxies of earnings management multiplied by minus one so that it is increasing in reporting quality.In the additional tests we use ACT_Litigious, ACT_Constrain, and ACT_LIWC as the AC tone measure.Our results show that ACT_Litigious has a negative and significant association with EM_MJM at 1% level and ACT_Constrain has a negative and significant association with EM_MJM at 10% level.ACT_LIWC does not seem to have a significant association with accrual-based earnings management.
ACStrength has a negative and significant association with accrual-based earnings management metrics at 1% level.
We further find (results untabulated) that FRQ_Index has a positive and significant association with ACT_Litigious (at 10% level), ACT_Constrain (at 1% level) and ACT_LIWC (at 5% level) indicating that the disclosure tone used in AC reports helps in improving financial reporting quality.Also, ACStrength has a positive and significant association with FRQ_Index at 1% level.Our findings from these additional tests are consistent with our previous findings and support our hypothesis on the negative association between AC report disclosure tone and earnings management.
Overall, our findings show that tone of audit committee disclosures, reflecting the linguistic choice of audit committees in describing their activity, seems to be communicated in a manner that helps to improve financial reporting quality.

Controlling for endogeneity
Although our results indicate that AC disclosure tone is negatively associated with earnings management proxies and helps in improving financial reporting quality, and despite our attempt to include a comprehensive set of control variables, it is possible that there are other underlying factors that drive our results.Firms with high earnings management may have propensity to disclose less or strategically manipulate the tone of disclosures to hide bad news and make earnings management activities less visible.This suggests that earnings management metrics and AC disclosure tone can be endogenously determined and thus could bias the regression results.Moreover, studies examining voluntary disclosures face selection bias issues (Katmon and Al Farooque, 2017;Bratten et al., 2019).To address these issues, we conduct two supplementary tests that attempt to control for endogeneity resulting from sample selection bias and omitted variables issue.
To solve sample selection bias, we employ the Heckman (1979) two-step approach consistent with Katmon and Al Farooque (2017).In the first stage, we run a probit model where the dependent variable is an indicator variable (Constrain_indicator/ Litigious_indicator), which equals one for observations where ACT_Constrain/ACT_Litigious is greater than the median value and zero for observations where it is less than the median value, and regress it on AC, board and firm-specific variables.We then compute the Inverse Mills Ratio (IMR) and include it in the second-stage regressions reported in Table 6 in order to control for the sample selection bias in our dataset.11Our inferences remain unchanged when using the Heckman (1979) two-step approach.

[Table 6 about here]
To address the endogeneity that may result from model misspecification, we test the main findings on a matched sample using the propensity score matching technique (PSM).We first run a probit model that uses Constrain_indicator/ Litigious_indicator as the dependent variable and variables that determine AC disclosures such as firm-specific variables, AC and board variables as regressors.We then estimate the propensity score and match based on it for each year-industry group using a 1% radius matching approach (Shipman et al., 2017).The quality of the matching performed shows a Rubin's B value below 25% and a Rubin's R value between 0.5 and 2, with mostly insignificant differences in the variables between the treatment and control groups.The untabulated results confirm the main findings and the absence of endogeneity bias.

SUMMARY AND CONCLUSIONS
This paper examines the effect of AC reporting on financial reporting quality as proxied by earnings management.We focus on AC disclosure tone and use textual analysis to assess the tone of AC reports.We utilise the litigation and constrain word dictionaries created by Loughran and McDonald (2011), measured by the word count frequency in the AC reports.We also measure the tone of AC disclosures in terms of an aggregate number measured as the sum of the frequency of positive words and frequency of negative words divided by the total word count using LIWC custom dictionary of positive and negative word lists.We use various measures for earnings management, our proxy for financial reporting quality.Using a sample of FTSE 350 firms, we find AC disclosure tone has a significant and negative association with REM proxies.When we divide the sample into firms at the top quintile REM and firms at lower levels quintile REM, finding shows that AC disclosure tone is negatively associated with higher-magnitude REM indicating that AC reporting helps reduce aggressing earnings behaviour while lower-magnitude REM may be perceived as ordinary business operations.
Our additional tests show that AC disclosure tone has a significant and negative association with accrual-based earnings and a positive and significant association with financial reporting quality index.The results support our hypothesis that the tone of AC disclosures is associated with financial reporting quality as proxied by earnings management.The tone of AC disclosures, as expressed in the choice of words used in describing AC activity, seems to convey information that affects the communication function of AC reporting and thereby help improve reporting quality.
Our paper has a few limitations.Future research could improve on our measurements and examine alternative conceptualisation and measurement of AC engagement and reporting.Future research could also explore to what extent the tone of AC reports is separate from the tone of other elements of the report.While our paper is an early attempt at extending AC research by focusing on textual analysis of AC reports and moving beyond AC characteristics and in complementing qualitative studies, our paper is subject to both data and measurement limitations.Our focus in this paper has been on financial reporting quality as proxied by earnings management.
Examining alternative measures of reporting quality as well the impact of AC reporting on audit quality would make valuable extensions to our paper.Moreover, exploring AC reporting in different institutional settings and in particular time periods are also potential avenues for extending AC research.
Our paper provides a valuable baseline evidence for future academic research and policy making.It shows that AC disclosures are not boilerplate but are varied in use of language choice and this affects financial reporting quality.Overall, our research suggests that AC reporting can aid transparency and improve the quality of financial reporting.

7
Thus, we include ACT_Litigious measured by the word count frequency in the AC report based on the Loughran and McDonald Litigious Word List, ACT_Constrain is measured by the word count frequency in the AC report based on the Loughran and McDonald Constrain Word List. 8Also, similar to Kang et al. ( performance-matched real earnings management (EM_PRM) is -0.0185.The mean value of accrual-based earnings management (EM_MJM) is -0.0292, and the mean value of performance-matched accrual-based earnings is -0.0095.The mean value of our financial reporting quality index (FRQ_index) is -0.0138.The AC disclosure tone measures viz.ACT_Litigious and ACT_Constrain show that 0.375 of the words in the AC reports are included in Loughran and McDonald Litigious Word List and 0.508 of the words are included in Loughran and McDonald Constrain Word List respectively.