Evidence on the Relation between Audit and Earnings Quality. Do Clients of Higher Quality Auditors Provide Better Financial Reporting?

Content

Abbreviations

Register of Tables and Figures

1. Introduction

2. Prior Research and Hypothesis Development
2.1. Audit quality
2.2. Earnings Quality

3. Research Question and Design
3.1. Audit Quality
3.2. Discretionary Accruals
3.3. Income Smoothing

4. Sample and Data
4.1. Sample Composition
4.2. Descriptive Statistics
4.3. Correlations

5. Distributional Properties
5.1. Linearity and Normality
5.2. Heteroskedasticity

6. Empirical Results
6.1. Discretionary Accruals
6.2. Income Smoothing

7. Discussion

8. Conclusion

References

A. Appendix

B. Description of the formulas
I. Modified Jones Model
II. Income smoothing model

Abstract

This paper studies the relation between audit and earnings quality. It examines whether firms audited by a Big 4 member engage in higher earnings management activities as proxied by the magnitude of discretionary and absolute accruals, as well as an income smoothing measure. The author predicts that large auditors have higher competencies and incentives to deliver a higher quality audit. Therefore, their clients are expected to reveal less sophisticated earnings management and thus higher earnings quality. The results do not support this relation.

JEL classifications: M41, M42

Keywords: earnings management, audit quality.

Abbreviations

ABSDACC absolute discretionary accruals

Abbildung in dieser Leseprobe nicht enthalten

Register of Tables and Figures

Table 1:Measuring earnings quality

Table 2: Overview of accrual-based EQ approaches

Table 3: Expected signs for the beta coefficients

Table 4: Sample selection process

Table 5: Descriptive statistics

Table 6: Correlation matrix

Table 7: Tests of the residuals for statistical assumptions for the full sample

Table 8: OLS regression with DACC and ABSDACC as the dependent variable

Table 9: Two sample t-test with DACC and ABSDACC

Table 10: OLS regression with IS as the dependent variable

Table 11: Two sample t-test with IS

Table 12: Different model specifications for H2

Table 13: Correlation matrix for controls

Figure 1: Research Model

Figure 2: Plots of the regression residuals that were estimated during the main tests to detect heteroscedasticity and non-normality

Figure 3: Plots of the regression residuals that were estimated during the main tests to detect heteroscedasticity and non-normality

Figure 4: Histogram for DACC and ABSDACC

Figure 5: Histogram for IS

1. ntroduction

Since a long time^[1] that standardsetters have been concerned about managers’ use of discretion to manage earnings in their financial reports, an increasing amount of empirical research was conducted to address this issue, additionally to regulation. While independent auditors (aim to) assure that these statements are in accordance with legal compliance, the actual audit quality can be grasped as the contingency that the auditor exposes and discloses an anomaly in their clients’ financial reports.^[2] Whereas numerous audit scandals^[3] threat the trustworthiness of well-known large auditors, there is various research revealing that Big N audited firms are supposed to disclose financial reports of higher quality.

Supplementing misguiding accrual accounting practices in this regard, this study also addresses another proxy for earnings management: income smoothing. Burgstahler and Dichev (1997) explain corporate income smoothing with the fact that managers avoid revealing earning decreases and losses to diminish costs arising from transactions with stakeholders.^[4] Similarly, Degeorge, Patel and Zeckhauser (1999) show that managers smooth earnings to meet analysts’ forecasts.

On the other hand there are various contrary studies.^[5] DeFond and Jimbalvo (1993) found that auditor-client disagreements resulting from earnings management, are more present in Big 4 audited firms. They explain this with the properties of the “common” Big 4 clients. For the reason of the ambiguous results, it is interesting to study the effects and compare them with prior evidence to answer the question whether Big 4 auditors deliver “higher” quality in terms of a “better” financial reporting. The terms are operationalized using a discretionary accruals and income smoothing measure and analyzed for (non-)Big 4 audited UK-firms in the period 2005-2011.

The paper can be classified as an empirical quantitative research using archival data and it consists of eight sections. Section 2 addresses the literature on audit and earnings quality, along with the specification of the hypotheses. Section 3 deals with methodological aspects. The sections 4 to 7 present and discuss empirical results and limitations, and finally section 8 concludes with the principal contributions of this study.

2. Prior Research and Hypothesis Development

2.1. Audit quality

Analyzing the effects of external accounting would be incomplete without related issues of auditing. Managers prepare the financial statement but opportunistic behavior, i.e. maximizing their personal utility functions, may lead to a misleading picture of the actual financial performance and further prevent the information function of accounting. Therefore, it is necessary that an independent third party makes sure the statements are in accordance with legal compliance. Then stakeholders can make proper decisions which improves resource allocation and contracting efficiency.^[6] According to Jensen and Meckling (1976) audit services are mandatory arising from a market-induced mechanism to reduce agency costs emerging of conflicts of interest between owners and managers. Hence that it is obvious auditing is required, it is further critical to know how good the quality is. The basic relation is that higher AQ leads to higher quality financial reporting. In their literature survey DeFond and Zhang (2014) evaluate studies of these dimensions and conclude that no proxy covers a complete picture of AQ.

2.2. Earnings Quality

Researchers deal with earnings quality (EQ) by investigating observed financial reporting data. The focus of the literature is on the accounting’s information function. Decision makers can make superior judgements in case of higher quality earnings as they provide more sound information about the firm’s financial performance.^[7] The literature examines certain characteristics of accounting which postulate that they are important for a high quality financial reporting that is providing “more information”. Table 1 sums up the empirical measurement approaches by differentiating between their theoretical underlying.

A commonly used proxy for EQ is the magnitude of accruals: Extreme accruals, resulting either from the application of accounting rules and principles or from earnings management, indicate low EQ because they represent a less persistence and sustainability of earnings which points to prevent evaluating future cash flows. There is evidence that large auditors contain higher competencies and reputation capital^[8] resulting in higher quality financial reporting compared to their smaller competitors. E.g. Becker et al. (1998) and Francis, Maydew and Sparks (1999) found that non-Big N audited firms enclose higher discretionary accruals misleading stakeholders from their actual financial performance. To maintain and protect their reputation, Big N auditors are supposed to have a lower acceptance regarding earnings management and therefore would deliver higher EQ. Summing up, the first hypothesis to be tested is:

H1: Big 4 audited firms deliver a better financial reporting quality resulting in a lower magnitude of discretionary accruals relative to firms not audited by Big 4 members.

“Smoothness” means that earnings consist of a low volatility relative to the cash flow from operating activities (OCF). In theory, accounting systems have an inherent smoothing function through accruals, which distribute payments over time, e.g. via depreciation. Smoother earnings indicate that managers used accounting policies (“window dressing”) that mislead stakeholders from their actual financial performance, e.g. to meet or beat certain target values or to influence contractual outcomes.^[9] As studies like Subramanyam (1996a) conclude that managers make use of discretionary accruals (DACC) to smooth their earnings figures, this issue will be analyzed. According to Dechow, Ge and Schrand (2010) there is evidence that income smoothing is a common corporate practice in firms. Therefore, the following hypothesis is tested:

H2: Big 4 audited firms deliver a better financial reporting quality resulting in a positive value of the income smoothing measure as compared to firms not audited by Big 4 members with a negative value.

3. Research Question and Design

3.1. Audit Quality

In this paper an input-based measure, namely auditor size, is used to proxy for AQ. The dummy variable equals 1 if the firm is audited by a Big 4 and 0 if not. Lennox (1999) and DeFond and Zhang (2014) state that Big N membership is the most common used proxy in the literature.

3.2. Discretionary Accruals

An overview of accrual-based EQ approaches can be found in Table 2. It should be taken into account that, according to Dechow, Richardson and Tuna (2003), there is no significant difference between these models. Moreover, no empirical model can fully capture EQ.

The first hypothesis is tested following the approach of Kothari, Leone and Wasley (2005).^[10]

Considering innate factors helps isolating the sole effects induced by external accounting. A firm’s size, growth and profitability could inherently influence earnings quality, also because large and well-being firms tend to have more balanced operations.^[11] Therefore, the following controls are added to obtain more robust results: firm size, sales growth and profitability. As there is evidence that firms with a higher debt ratio tend to claim smoother earnings for meeting creditor’s expectations, leverage gets also controlled for.^[12] The multiple regression model used to test hypothesis H1 is:

As DACC is a signed value variable proxying for the degree of conservatism^[13], the absolute value (measuring the overall propensity to manage earnings regardless of income increasing or decreasing incentives) gets added in a second OLS regression for obtaining more solid results (ABSDACC). As the Big 4 dummy in the regression model compares the average of the two groups, a negative and significant β1 coefficient is expected for both regressions^[14]. The performance controls are expected to be negative in both regressions, while leverage is predicted to be positive.

3.3. Income Smoothing

Managers may make use of accrual accounting to conceal weak economic performance or underreport strong current performance to create reserves for the future. In this regard, income smoothing is measured following Leuz, Nanda and Wysocki (2003).^[15]

Hypothesis 2 is tested with the following regression model, while the above mentioned innate characteristics are again controlled to obtain greater robustness.

A negative correlation of the computed IS value refers to accrual accounting, indicating that managers made use of accrual accounting to buffer operating cash flow shocks. Combining this with the postulated “higher quality” auditing competencies of large auditors, the measure is expected to be positive and of higher values for Big 4 and negative for the other group.^[16] The performance controls are expected to be positive and leverage is predicted to be negative, to support the hypothesis H2.

Table 3 sums up the predicted coefficient signs for all models. The hypotheses are presented in Figure 1.

4. Sample and Data

4.1. Sample Composition

The initial sample comprises 3,027 UK-firms based on 15,368 firm-year observations in the period of 2005 until 2011 obtained from Worldscope. Table 4 describes the sample selection process which is consistent to prior research.^[17] The final sample consists 6,081 observations.

4.2. Descriptive Statistics

While the following provides a short overview of some interesting findings, Table 5 reveals the full descriptive statistics for both models’ applied variables.^[18] The mean number of firms within year-industry is 91.26, which is slightly below the value of 100 that other studies postulate to have at least.^[19] Audit quality, proxied by the Big 4 dummy, is almost equally distributed (0.52) in the sample, which makes the analysis more robust. While the mean and median value of DACC is nearly zero (-0.01), the range is relatively large (from -1.09 to +1.16). The IS measure is negative on average (-0.61) and also its median value (-0.80), which is a first hint, that income smoothing is used by the average of the sample.

4.3. Correlations

The bivariate correlation matrix is presented in Table 6. While there is no remark between DACC and the Big 4 dummy, DACC reveals certain significant correlations with IS (-0.0396, p<1%), profitability (0.0302, p<5%) and leverage (-0.0559, p<0.1%). On the contrary, the absolute measure of DACC (ABSDACC) highly significantly negatively correlates with the dummy variable on the 0.1% level indicating the postulated hypothesis H1 because the proxy reveals the magnitude of earnings management and therefore a lower EQ. This measure also significantly negatively correlates with all of performance control variables and positively correlates with leverage, which also meets the above mentioned expected relation, that large and well-run firms tend to have smoother operations and firms with a higher leverage tend to use more earnings management to meet their creditors’ expectations and covenants.

Regarding the income smoothing measure, there is a significant positive correlation to leverage (0.0520, p<0.1%) and significant negative correlations to: Big 4 (-0.0672, p<0.1%), size (-0.131, p<0.1%), sales (-0.189, p<1%).

The key findings of the correlation matrix are as follows: (i) the control variables reveal a relation to the dependent variables; (ii) the controls’ correlation signs meet the authors’ expectations for ABSDACC, but surprisingly not for DACC, which might be a hint that ABSDACC fits better than the signed value DACC; (iii) the signs for the correlations with IS are exactly the opposite of the ones of DACC, supporting the above described theoretical part; (iv) the negative correlation between the dummy and IS (-0.0672) contradicts the expected relation because a higher (signed) value means less earnings management and therefore higher EQ; the same contradiction is given for Big 4 and DACC indicating income increasing activities; and finally (v) the expected signs apply only for ABSDACC in the data and are not in accordance with the predictions for DACC^[20] and IS.

5. Distributional Properties

5.1. Linearity and Normality

Analyzing the scatterplot of residuals and the fit of the models’ variables provide insight regarding linearity and normality. Albeit, the Shapiro-Wilk rejects the normality of the standardized residuals, the graphs clearly show that these are approximately normal (see Table 7, Figure 2 and Figure 3).

5.2. Heteroskedasticity

The scatterplots in Figure 2 and Figure 3 compare the standardized residuals with the predicted residuals, which indicates whether the variance of the residuals is constant (homogeneous). The residuals display a clear dispersing shape (values do not keep going further from the zero line) and residuals are spread randomly enough. This shows that the model is approximately homoscedastic. Nevertheless, to correct for remaining heteroskedasticity observed in the figures, the regression parameters get estimated with robust standard errors in the forthcoming analysis part.

6. Empirical Results

6.1. Discretionary Accrual s

Using the above specified OLS regression model, Table 8 concludes that Big 4 members do not have a differential effect on the earnings quality (beta=0.0045, p>10%). The explanatory power (R²) is 0.003, which means that the model variables explain 0.3% of the variation in DACC. The F-statistic is insignificant (F=1.239, p>10%). The significant effect of leverage (-0.02130, p<10%) contradicts the expected sign, as well as the sales growth and profitability control variables and as well for the dummy.

The regression using ABSDACC, as shown in the second column, is highly significant (R²=0.094, F=64.99, p<0.01%) and nearly all controls are in accordance with the expected coefficient signs and moreover are significant on the 0.1% level. Nevertheless, the significance is not induced by the Big 4 variable. Hence, this result cannot be used to draw a conclusion regarding the research question. Furthermore, the β1 (0.00822) is not negative.

Figure 4 plots the histogram for DACC split by the two sub-samples. It shows that the Big 4 group has a higher density around 0 of DACC indicating actually greater EQ. The same result can be seen for the absolute values of the proxy, as depicted in the second histogram of Figure 4. Additionally, a two-sample t test is used for comparing the mean values of the two groups. Table 9 tests for both signed values (DACC) and absolute values (ABSDACC). Testing the alternative hypothesis that the difference is not equal to zero for DACC and greater than zero for ABSDACC, the first measure shows an insignificant difference of -0.0006 (t= -0.12). The absolute measured difference, as given in the second row, is 0.0401 (t=0.38), which is highly significant on the 0.1% level under the alternative hypothesis that the difference is greater than 0. In the absence of a significant impact by the Big 4 dummy on accruals using OLS regression, the mean-comparison-test would accept the hypothesis, but this evidence is not robust.

6.2. Income Smoothing

Table 10 reveals the regression results of the second model. Albeit the beta of the dummy is as predicted positive, it is not significant (beta=0.0037, p>10%). However, the results get interesting when looking at the F-statistics, which indicates the fit of the model with the underlying variables. The F-value is 25.95 and highly significant (p<0.1%). But this fact is probably due to the control variables: size (beta= -0.01565, p<1%), sales (beta= -0.0000, p<5%), profitability (beta= -0.12184, p<1%) and leverage (beta= -0.04782, p<5%), where only leverage meets the sign-expectation.

The histogram for both groups (Figure 5) displays the tendency, that there is a higher density in negative values for the IS measure. Even firms audited by a Big 4 overall tend to engage in income smoothing activities. Again, an additional t-testis conducted. Table 11 reveals that the mean values of both groups are negative. This indicates a discretionary use of accrual accounting. It is remarkable, and against the authors’ expectation, that the Big 4 audited firms have a higher value of the income smoothing measure (-0.6433). The p-value for the probability that the difference of 0.06911 is lower than 0 is not significant (t=4.47, p=1). Overall, the hypothesis H2 is rejected.

7. Discussion

This paper addressed the relation between audit and earnings quality to achieve a better understanding whether the Big 4 auditors deliver higher quality in terms of a “better” financial reporting.

Referring to the insignificant dummy coefficient value (beta=0.0045, p>10%) from the first OLS regression using DACC as the dependent variable, the only conclusions are: (i) managers of Big 4 audited firms tend to engage more in income increasing activities through discretionary accrual accounting relative to the other group, and surprisingly (ii) its magnitude is on average higher compared to the firms audited by a smaller auditor. The first fact counteracts the theory of auditor conservatism. If large auditors are considered to deliver higher quality, then they should be more sensitive to income-increasing practices. Hirst (1994) suggests that litigation costs are more likely to be charged for income overstatement than the opposite. Finally, the F-value indicates that this model does not fit and only one variable (size) meets the expected sign.

Though the model fit of the second regression using ABSDACC was highly significant, this is not affected by the dummy variable because it is insignificant. Again, its positive coefficient signals that the average of the Big 4 audited firms have higher values of absolute discretionary accruals which points to earnings management and therefore a lower EQ. Only the two-sample t-test revealed a high significant relation, but the author does not accept this result because of statistical weaknesses, as explained in the note of Table 9.

The analysis part for the second hypothesis H2 did also not show a significant relation regarding the research question. Though the fit of the postulated regression model was highly significant, this was affected by the control variables. Therefore, Table 12 presents several model specifications to provide a deeper understanding about which model, containing the controls, fits best.^[21] As it was not possible to find a relation to the Big 4 audited firms, more research is necessary in regard of the control variables in addition to the parent auditor. The specifications also contribute to future research.

A possible explanation of the stated regression models’ outcome is: Though the postulated higher quality of larger auditors might lead to a prevention of a greater fraction of unwarranted accruals, their clients might also have relatively higher levels of earnings management activities before the audit process, which makes the audit more challengeable relatively to the (smaller) clients of smaller auditors with lower pre-audit earnings management. Also, the accounting issues of a large auditor’s client might be more complex and therefore leaving higher space for both earnings management and disagreements.

Overall, it needs to be noted that the “findings” of this study are subject to certain limitations. Maijoor and Vanstraelen (2006) studied the effects of member state audit environment, audit firm quality and international capital markets in three distinct legal traditions, where one of those is the UK^[22]. Their results showed that firms in countries with flexible audit quality regimes report significantly higher ABSDACC compared to companies in countries with strict audit quality regimes. In this regard, they concluded that national differences in earnings management are the dominant factors and further independent of the presence of a Big 4 auditor. Regarding this fact, the insignificant results as shown in this study, are surprisingly, because Maijoor and Vanstraelen (2006) classify the UK’s audit quality regime as the second most strict.^[23] Nethertheless, due to institutional differences, the outcome of this work, cannot be generalized.

8. Conclusion

Theory suggests that “high quality” auditors are more likely to detect and correct financial reports with misleading accounting policies. Consistent with previous studies, this work used a Big 4 membership dummy to proxy for audit quality. Earnings management was evaluated with signed and absolute values of discretionary accruals using the modified Jones model and secondly with an income smoothing measure. The author hypothesized that firms audited by a Big 4 reveal a higher earnings quality based on relatively less earnings management practices as compared to firms audited by smaller competitors.

Although or even because this paper did not approve the suggested and postulated theory, it makes considerable contributions for external stakeholders and smaller auditors. Healy and Wahlen (1999) explain earnings management with the principal agent theory. They suggest that managers have incentives to abuse their control power to maximize their own utility function. As this research did not find the supposed positive relation between AQ and EQ, stakeholders do not have to be “afraid” if a financial report is not audited by a Big 4 member. Additionally, implying that smaller auditors do at least not deliver lower quality services in this regard, they can use this works’ outcome for convincing their potential clients.

To conclude, the academic field shout not exclusively focus on research about whether mangers use earnings management or not and if so, whether large auditors are more sensitive with this issue. The author postulates that it would be more effective to analyze which balance sheet items of accruals or other positions are most subject to discretionary activities.^[24] Standardsetters would benefit from these findings and could adjust the respective law sections and therefore diminish the margins of discretions and valuation leeways.^[25]

[...]

---

^[1] See Levitt (1998).

^[2] See DeAngelo (1981).

^[3] A small excerpt: Enron (2001); AOL (2002); AIG (2004); Lehman Brothers (2010); Toshiba (2015).

^[4] If stakeholders base their decisions on heuristic cutoffs, transactions costs are supposed to be reduced by disclosing smooth earnings.

^[5] See e.g.: Subramanyam (1996b); Becker et al. (1998); Tucker and Zarowin (2006).

^[6] See DeFond and Zhang (2014), p. 275.

^[7] See Dechow, Ge and Schrand (2010), p. 344.

^[8] Such as: more sophisticated recruitment approaches, high learning curves, technology, networks.

^[9] Further motives for smoothing are decreasing the perceived economic riskiness by increasing the firm value because of reported smoother earnings or the managers’ will to secure their job position. See Trueman and Titman (1988); Fudenberg and Tirole (1995); Healy and Wahlen (1999)

^[10] The approach is described in Bartov Eli, Gul and Tsui (2000) and Dechow, Sloan and Sweeney (1995). For readers, who want to gain a deeper understanding of the Modified Jones Model, a description of its application is provided in chapter B.I.

^[11] See e.g. Dou, Hope and Thomas (2013). Dechow, Ge and Schrand (2010) provide an overview of used controls in EQ literature.

^[12] See e.g. Gassen and Fülbier (2015) with additional references; DeFond and Zhang (2014), p. 291.

^[13] Positive values indicate income increasing earnings management and vice versa.

^[14] Regarding DACC, it needs to be taken into account that both signs indicate earnings management in terms of income increasing (decreasing) activities for positive (negative) signs.

^[15] The formula for the income smoothing measure is depicted in chapter B.II.

^[16] See e.g. Dechow (1994), Leuz, Nanda and Wysocki (2003). However, a certain degree of income smoothing is “accepted” considering the accrual reversals.

^[17] See for example Kothari, Leone and Wasley (2005) or Dou, Hope and Thomas (2013).

^[18] For reasons of brevity, the author renounces presenting two separate tables for the two models. For the income smoothing model, unavailable variables necessary to calculate the respective variables were dropped additionally, which led to 1,680 deletions. However, not to overwhelm the reader, this is not depicted in the descriptive statistics (which does not change the overall interferences). But the correlation matrix and all regression models contain the dropped data.

^[19] See Dou, Hope and Thomas (2013).

^[20] Except of size_m, which shows the expected negative relation, though not significant.

^[21] As it seems that size (specification 1) and profitability (specification 3) have the highest explanatory power (R²=1.7% and 3.6%), specification 5, containing the two of them, would be the best model explaining 4.2% of the variation in income smoothing.

^[22] LaPorta et al. (1998) describe that the UK, as the originating country of the English common law differs from other jurisdiction in crucial accounting environment factors.

^[23] This is due to several factors, e.g.: audit rotation; the disclosure of audit fees; auditors are subject to reviews by regulators; high litigation risk; stronger investment protection environment.

^[24] To say it with a positive theory approach: Why do managers use earnings management and how exactly are they engaging in those practices (meaning which exact positions are most used for it) and how do stakeholders react?

^[25] E.g. depreciation method choices, the capitalization versus expense decision, inventory valuation and others.