ACCOUNTING QUALITY VERSUS AUDITOR CHOICE UNDER STRONG Tax-GAAP CONFORMITY : THE CASE OF BRAZIL

Brazilian companies with a Big 4 auditor have better accounting quality than those with a local auditor, when accounting quality is measured either by compliance with GAAP or by conservatism. However, the cross-sectional pattern of discretionary accruals—often used to measure accounting quality in other countries—is unrelated to quality in Brazil. In fact, companies with Big 4 auditors tend to recognize income more aggressively than companies with a local auditor. This is the opposite of what happens in other countries and it is consistent with local auditors interpreting the tax code (rather than GAAP) more aggressively than Big 4 auditors, due to the strong linkage that exists between the Brazilian tax code and fi nancial reporting standards.


INTRODUCTION
Is it true in Brazil-as it appears to be in many other countries-that companies that hire a Big 4 1 auditor produce better accounting disclosures than companies that hire a local auditor?This question involves more than a replication of similar studies conducted for other countries because of two special features of the Brazilian environment.The fi rst is leniency in the enforcement of generally accepted accounting principles by Brazilian regulatory agencies (La PORTA, LOPEZ-DE-SILANES and SHLEIFER, 1998;PATEL, BALIC and BWAKIRA, 2002.)The second feature is the existence of a strong linkage between the Brazilian tax code and fi nancial reporting principles (HUNG, 2001;Appendix A.) Leniency does not imply that quality is irrelevant for companies and auditors because alternative mechanisms arise to impose accounting discipline.For example, a higher quality of earnings is associated with lower costs of capital (BOTOSAN, 1997;RICHARDSON and WELKER, 2001;HRIBAR and JENKINS, 2004;and FRANCIS, LaFOND and OLSSON, 2004.)In addition, where enforcement is weak but the GAAP framework is strong, fi rms may attempt to differentiate themselves in terms of accounting quality by choosing a reputable auditor (HOLTHAUSEN, 2003.)Still, it is an empirical question whether Brazilian fi rms rely on auditor choice as a signal of quality.
A strong linkage between tax and fi nancial reporting transforms into a serious handicap a common feature of many studies of quality and auditor choice (e.g: JONES, 1991; DECHOW, SLOAN and SWEENEY, 1995): the assumption that Big 4 auditors, because of their large reputation capital and "deep pockets," will try to impose conservative accounting methods (i.e., income reducing accruals), more often than their smaller competitors will.The problem with this assumption, under strong tax-GAAP conformity, is that minimization of income in the fi nancial statements leads to minimization of income in the tax return, and the latter may cause problems with the tax authority.Thus, it is no longer clear whether the Big 4 would prefer income increasing or income decreasing discretionary accruals.
Using a sample of 97 publicly traded fi rms that adopted the REFIS tax amnesty of 2000, it was found that there is no difference in quality between Big 4 and non-Big 4 auditors if quality is measured by the behavior of discretionary accruals, but signifi cant differences were found in quality using compliance with GAAP and conservatism as indicators of quality.These results support the hypothesis that Brazilian fi rms use Big 4 auditors as a signal of quality, and show that discretionary accrual behavior is an unreliable indicator of quality under strong tax-GAAP conformity.
1 Throughout this paper it was used the current "Big 4" terminology, even though in 2000 there were still fi ve large multinational accounting fi rms: Andersen, Deloitte, Ernst & Young, KPMG, and PricewaterhouseCoopers. 2 cash fl ow from operations as funds from operations have been computed (as published under Brazilian GAAP) minus working capital accruals, interest expense and infl ationary adjustments.

Discretionary Accruals
Accruals can be broken down into discretionary and non-discretionary components.Discretionary accruals, measured by the original or by the modifi ed Jones model (JONES, 1991;DECHOW, SLOAN and SWEENEY, 1995), are often used in studies of quality because they are vulnerable to manipulation.Non-discretionary accruals, being driven by the fi rm's level of operations, are assumed to be unrelated to quality.
Two measures of accruals were used: total accruals (net income minus operating cash fl ow 2 ), and working capital accruals (increase in net working capital excluding short term debt, cash and cash equivalents.)Discretionary total accruals based on the original and modifi ed versions of the Jones model, for company i at year t, are denoted respectively as DTA 1 it and DTA 2 it , and are estimated as the error terms in: and in: The following notation is used: (i is the fi rm, t is time) A it-1 total assets at the end of t -

Compliance with GAAP
The second indicator of accounting quality is compliance with GAAP.In particular, it was observed the degree to which listed companies that accepted the REFIS tax amnesty of 2000 complied with reporting requirements of the Brazilian Securities Commission (CVM).The CVM's rules for the amnesty-Instruction No. 346 (of 09/2000)-are clear and the amnesty's impact on most acceptors was signifi cant, which means that involuntary errors and omissions should be rare and lack-of-materiality should not be a common reason for avoiding disclosure.
A key requirement of I#346 was for companies to disclose the net benefi t from the amnesty as an extraordinary item and to provide a detailed breakdown of the benefi t in footnotes.The gain attributable to the amnesty can be traced to penalties waived, previously unrecognized tax assets or liabilities and discounts on net operating loss carry forwards acquired.(Table 2, Panel B, contains a summary of I#346.) The information content of the disclosures required by I#346 is not uniform.For example, item A (the impact of the amnesty on income) is a key input for estimating the value of securities issued by each company.However, the information provided in items E though H can be obtained easily from other sources or estimated by investors.Since the quality implications of not complying with any given item in I#346 depends on that item's relevance to decision making by investors, two subsets of disclosures were considered: items needed and items not needed for decision making by investors.
To assess compliance with I#346 management's discussion and analysis, the auditor's letter, fi nancial statements, and footnotes included in year 2000 fi nancial reports for all publicly traded companies that accepted the amnesty were consultedand it was checked whether these companies complied in full, partially, or not at all with the mandatory requirements of I#346.It was assigned to each company 100 points in the fi rst case, 50 in the second, and zero in the third.Finally, a company's decision-making and overall scores as the ratio of points assigned divided by the maximum points available in each category was computed, expressed as a percentage.

Conservatism
The third indicator of accounting quality is conservatism, defi ned as timely recognition of contingent losses.Accrual of contingencies can be used as a measure of quality in Brazil because they do not generate a deductible expense for tax purposes.Conservatism was measured using evidence regarding liabilities that were revealed for the fi rst time in December of 2000-when companies joined the amnestythat probably should have been disclosed before.
Companies that accepted the amnesty were committed to paying a reduced liability over an extended period of time at below market interest rates.Prior to accepting the amnesty, these companies were disputing the claim that taxes were overdue and they could have continued to do so had they rejected the amnesty.It was assumed that companies that accepted the amnesty weighed the expected liability value of rejecting the amnesty against the expected value of accepting the amnesty, and found the former to be greater.
According to Brazilian GAAP, the need and the format for disclosing contingent liabilities depends on the anticipated probability of loss. 3There are three cases: remote contingencies need not be disclosed; reasonably possible contingencies must be disclosed but not accrued; and probable contingencies must be described in footnotes and accrued if their values can be estimated.Brazilian standards are not precise about numerical probability thresholds for these three categories.Therefore, past due taxes included in the amnesty program could be in one of four possible conditions before December of 2000: (1) disclosed on the balance sheet as part of an installment agreement; (2) disclosed on the balance sheet as a probable contingent liability; (3) disclosed in the footnotes as a reasonably possible contingent liability; or (4) they could be undisclosed.It was assume that all category (4) liabilities included in the amnesty were in violation of GAAP and should have been at least mentioned in footnotes by December of 1999.If perceived probability thresholds in Brazil are similar to those in the U.S., this is equivalent to assuming that all previously undisclosed liabilities included in the amnesty had loss probabilities equal to or greater than 25%, the highest estimate of the remote/reasonably possible threshold in the U.S. (RAGHUNANDAN et al., 1991;REIMERS, 1992;AHARONY and DOTAN, 2004.) 3 Accounting for loss contingencies under Brazilian GAAP is described in IBRACON Statement No. XIII.

DATA AND DESCRIPTIVE STATISTICS
In 2000 there were 769 listed companies in Brazil according to the CVM's website (http://www.cvm.gov.br/).Of these, 224 are eliminated for being government-controlled or classifi ed as utilities or fi nancial institutions.Another 148 are eliminated for the following reasons: 77 reported zero revenues in 2000; 38 had trading in their securities suspended; 3 were not operating or reported zero net income in 1999-2000; 11 were bankrupt; and 19 did not fi le fi nancial reports for 2000.For each of the remaining 398 companies it was obtained fi nancial statements for 1999 and 2000, and also management's discussion and analysis, footnotes, and the auditor's letter for 2000.Of those 398 companies, 97 joined the tax amnesty of 2000.
In order to estimate discretionary accruals information on funds generated by operations was needed.In 1999 and 2000, 108 and 86 respectively of the remaining 398 companies in the sample did not fi le, or fi led incomplete funds fl ow statements.This reduces the effective sample size for estimation of discretionary accruals to 290 in 1999 and to 312 in 2000.In the sample used to estimate discretionary accruals in 2000, most fi rms are engaged in manufacturing activities, the proportions being 64% in the full sample and 72% in the amnesty sub-sample.About 57% of fi rms in the full sample are audited by one of the Big 4, but only 38% in the amnesty sub-sample.In terms of exchange listing, a higher percentage of companies that choose a Big 4 auditor are listed on Bovespa and/or a U.S. exchange (72%) than companies that choose otherwise (65%).
Table 1 contrasts median values of total and working capital discretionary accruals, size, asset turnover, leverage and profi tability according to auditor choice in 1999-2000.There are signifi cant differences in discretionary accruals measured by the modifi ed Jones model (DTA 2 and DWKA 2 ) in 1999-when fi rms with a Big 4 auditor tend to post less income-decreasing accruals.There are no signifi cant differences in discretionary accruals in 2000.Firms with a Big 4 auditor are consistently larger based on mean and median total assets, and more profi table based on the median ratio of operating profi ts to sales.There is also indication in 2000 that fi rms with a Big 4 auditor carry less debt, based on the median ratio of liabilities to total assets.Table 2, Panel A contains compliance statistics for each item in I#346.The fi rst column describes the disclosure, the second column has the number of fi rms in the sample that should have provided that disclosure, and the last three columns have the percentages of fi rms that complied in full, partially, or not at all.The percentage of fi rms that completely ignored each disclosure requirement varies from 3.3% to 90.6%, with 44.3% of the 97 acceptors providing no information at all about the impact of the amnesty on income (item A).Another indicator of poor compliance is that, although all 74 fi rms that chose to pay the amnesty over a variable term were required to disclose the present value of the liability (item H), only 8.1% complied in full, 18.9% provided some information, and 73% provided no information regarding the present value of the liability.Of the 97 amnesty companies, only 40 provided suffi cient information to allow estimation of the values of accepting and rejecting the amnesty.These are the companies for which the third measure of accounting quality (conservatism) can be evaluated: 40 of these companies achieved an average decision-making score of 69.8 versus 28.1 for the remaining 57 companies.
In order to verify the degree to which the CVM enforced I#346, "Market Alert/Rectify and Republish Actions" (since 06/2001) and "Decisions of the CVM's Collegiate in Appeals Processes" (since 08/2000) on the CVM's website were consulted. 4Most, if not all disciplinary actions taken by the CVM regarding annual reports for fi scal 2000 should be included in one of these two archives.Of the 97 fi rms in the sample of amnesty acceptors, only three were ordered by the CVM to republish year 2000 fi nancial statements due to violations of I#346.5This is consistent with the generally low rankings achieved by Brazil in international comparisons of GAAP enforcement (La PORTA, LOPEZ-DE-SILANES and SHLEIFER, 1998;PATEL, BALIC and BWAKIRA, 2002.)Table 3 contains a breakdown of amnesty liabilities for this group of 40 acceptors.The breakdown is according to whether, prior to the amnesty, liabilities had been recognized as payable on the balance sheet (P-type), contingent on the balance sheet or in the notes (C-type), or unrecognized (U-type).The fractions of the aggregate amnesty liability that had been recognized as P, C, and U were 40%, 10% and 50% respectively.The fraction of unrecognized liabilities is higher for the 18 companies with a local auditor (68%) than for the 22 companies with a Big 4 auditor (40%).

Discretionary Accruals
If discretionary accruals were driven mostly by earnings manipulation motives it would be expected that companies with a Big 4 auditor display more income-reducing discretionary accruals than companies with a local auditor.The study showed that the opposite would be expected, however, if accruals are heavily infl uenced by tax minimization motives.Although it is a priori unknown which of these two motives dominates in Brazil (where tax-to-GAAP conformity is strong), it is worthwhile to observe the pattern of discretionary accruals as a function of auditor choice.If we can assume that Big 4 audits produce higher quality fi nancial disclosures, and companies audited by the Big 4 have less income-decreasing discretionary accruals, then we can conclude that the Big 4 are less tolerant of aggressive tax minimization strategies.
It was measured the association between auditor choice on total discretionary accruals by means of the following linear regression model, estimated for 1999 and 2000: where i is a fi rm index, and: DTA K i discretionary total accruals obtained with the original (k = 1) and modifi ed (k = 2) versions of the Jones model; AUD i auditor choice, 1 if auditor is Big 4 (AA, DT, EY, KPMG, PWC in 2000), 0 otherwise; LST i listing choice, 1 if fi rm is listed on Bovespa or any U.S. exchange, 0 otherwise; HILEV i indicator of very high leverage, 1 if the company is in the highest decile according to the ratio of debt to total assets at year-end, 0 otherwise; NEGSE i indicator of negative shareholders' equity, 1 if shareholders' equity is negative at year-end, 0 otherwise; SIZ i fi rm size, equal to the natural logarithm of total assets at year-end; LABTA i magnitude of accruals, equal to the natural logarithm of absolute total accruals; Model [3] is similar to the model used in (BECKER et al., 1998), except that two variables were dropped-change in shares outstanding and change in auditor-due to data collection costs, and it was added the choice of exchange listing and negative shareholders' equity variables.High leverage can be associated with income-increasing accruals, in the case of fi rms that are close to violating debt covenants, or to income-decreasing accruals, in the case of distressed companies (DeFOND and JIAMBALVO, 1994).Negative shareholders' equity is introduced as an additional control for highly distressed companies.Since companies audited by the Big 4 are on average larger than companies with local auditors by value of total assets (Table 1), and size can proxy for omitted variables, the natural log of total assets is included.The last variable is magnitude of accruals, because of the possibility that large dis-cretionary accruals are associated with large total accruals.(There is some indication in Table 1 that, in 2000, working capital accruals are greater for the Big 4.) To examine the association between auditor choice and working capital discretionary accruals, DWKA k i was substituted on the left-hand side, and LABTA i was replaced by the log of absolute working capital accruals on the righthand side.
Results for 1999 and 2000 are given in Table 4.In general, if discretionary accruals are determined by the original Jones model, it cannot be rejected the null hypothesis that auditor choice is unrelated to discretionary accruals in 1999 or in 2000.Switching to the modifi ed Jones model, it still cannot reject the null in 2000, but it does reject it in 1999.In 1999 the coeffi cient of AUD is positive in the models of total accruals (DTA 2 ) and working capital accruals (DWKA 2 ), implying that the expected value of discretionary accruals conditional on choosing a Big 4 auditor is greater (i.e., more income-increasing) than if a local auditor is chosen.
These results imply that tests based on discretionary accruals fail to discriminate accounting quality in Brazil in 1999-2000.If audit quality and accounting quality are related-in the sense that Big 4 auditors would engage in less income-increasing accruals than local auditors if there were no tax constraints on fi nancial reporting-and if the true coeffi cient of AUD is on average positive over time for the modifi ed Jones model, the implication is that Big 4 auditors tolerate a less aggressive interpretation of the tax code than local auditors.

Compliance with GAAP
To test whether compliance with I#346 is affected by auditor and exchange listing choices, it was estimated the following cross-sectional regression model: where i is a fi rm index, and: CSCR i compliance score with I#346, measured by the ratio of points obtained to maximum possible decision making points (or overall points), expressed as a percentage, CON i concentration of control, measured by percentage of common and preferred shares owned by controlling shareholders, and the other variables are as defi ned before.
Compliance to be positively related with the choices of a Big 4 auditor and with the decision to list on Bovespa or on a U.S. exchange is expected.It is also expected that this association be stronger when compliance refers to the decision-making items (more relevant disclosures to investors), than when compliance refers to absolutely all items of the CVM's instruction.The regression model includes four other variables.Indicators for high leverage and for negative shareholders' equity, as well as a measure of size, are included for consistency with model [3].Risk and size proxy for reputation at stake and other potentially omitted variables (BECKER et al., 1998).The fourth variable is concentration of control, which allows for the possibility that fi rms with concentrated ownership produce better disclosures (DeFOND and JIAMBALVO, 1991). 6 Table 5, Panel A, shows the results of estimating model [4] with OLS for the decision-making and for the overall compliance scores.With the decision-making score as the 6 It was not used the concentration of control variable in model [3] due to data collection constraints.
dependent variable the null hypothesis (H 0 ) that accounting quality is either unrelated to auditor choice or worse for Big 4 auditors is rejected, in favor of the alternative that quality is better for companies with a Big 4 auditor (p = .025).With the overall compliance score as the dependent variable it can only be rejected H 0 at or above the 5.4 % level of signifi cance.The stronger result for auditor choice when the decision-making score is used as a measure of accounting quality is consistent with a heavier emphasis by auditors on disclosures that are relevant for capital allocation decisions.The p-values for exchange listing choice are about.003for both measures of compliance.

Timeliness of Loss Recognition
The model for timeliness versus auditor and exchange listing choices is: where i is a fi rm index, and: TIM i timeliness (compliance with GAAP for contingencies) measured by the negative of the ratio of previously undisclosed tax liabilities to all tax liabilities, and the other variables are as defi ned before.
It was expected timeliness to be positively related with the choices of a Big 4 auditor and with the decision to list on Bovespa or on a U.S. exchange.
The other independent variables are the same as in model [4], and are included for similar reasons.
Table 5, Panel B, has the results of estimating model [5] with OLS, which led to rejecting the hypothesis that timeliness is unrelated to auditor choice, or worse for Big 4 auditors, in favor of the alternative that timeliness is better for companies with a Big 4 auditor (p -.041).It cannot be rejected, however, the hypothesis that timeliness is unrelated to choice of exchange listing.

CONCLUSION
In this paper it was examined how publicly traded Brazilian companies that accepted a national tax amnesty in 2000 complied with the CVM's disclosure requirements for the amnesty and for contingent liabilities.Based on these observations it was concluded that Brazilian companies with a Big 4 auditor have better accounting quality than those with a local auditor when accounting quality is measured by compliance with GAAP or by conservatism.This result supports Holthausen's (2003) conjecture that companies attempt to signal accounting quality by means of auditor choice in countries with good accounting principles but lenient enforcement of those principles.
However, the cross-sectional pattern of discretionary accruals-often used to measure quality in other countries-is unrelated to quality in Brazil.In fact, companies with Big 4 auditors in Brazil tend to recognize income more aggressively than companies with a local auditor.This is the opposite of what happens in other countries and it is consistent with local auditors interpreting the tax code (rather than GAAP) more aggressively than Big 4 auditors due to the strong linkage that exists between the Brazilian tax code and fi nancial reporting standards.

Tax-to-GAAP Conformity in Brazil in 2000
To determine degree of fi nancial-tax conformity in Brazil in 2000, Hung's (2001) method was used, according to which tax-to-GAAP conformity is high if the combined score from the six items below is positive; otherwise, conformity is low.
1.  2002) reports on the ability of tax authorities in Latin America to challenge transactions based on economic substance.E&Y reports that Brazil has historically applied a form over substance approach, but a new law passed in January 2001 will allow authorities to disregard transactions created with the sole purpose of minimizing taxation.Since these developments occurred after the time period studied, the answer is yes.4. Is additional accelerated depreciation allowed?Yes-1; Limited-1 ½; No-0 Additional accelerated depreciation means methods other than declining balance or sum-of-years digits.There are two forms of accelerated depreciation in Brazil.According to articles, 313-323 of the tax code (RIR) fi rms can depreciate at straight-line rates above the usual rates for a variety of special cases in which the government wants to encourage investment.Although the increase in depre-ciation expense caused by this method is tax deductible, it cannot be recognized in the fi nancial statements, and therefore leads to deferred tax liabilities (i.e., no tax-to-GAAP linkage).According to article 312 of RIR, fi rms that work for a single daily 8-hour shift must apply standard straightline rates; fi rms that work two shifts can apply 150% of the standard rates; and fi rms that work three shifts can apply 200% of the standard rates.These increases in depreciation expense must be recognized in the books, are not equivalent to any of the traditional depreciation methods, and are deductible for tax purposes.In this case, there is a linkage between tax and fi nancial reporting.Therefore, yes. 5. Do amortization periods depend on tax laws?Yes-1; Limited-1 ½; No-0 According to article 327 of RIR the minimum is 5 years, which is refl ected in IBRACON (Instituto Brasileiro de Contadores) Statement VIII.6.Does lease capitalization depend on tax laws?Yes-1; Limited-1 ½; No-0 Article 415 of RIR establishes criteria for capital leases.Brazilian GAAP in 2000 do not differentiate between capital and operating leases, requiring only that the asset and liability effects of capitalizing the lease be shown as a footnote.Therefore, item 6 is not applicable.The combined score is greater than zero, which implies a high degree of tax-GAAP conformity.Additional supporting evidence includes: depreciation rates used in fi nancial reports are stipulated in the tax code; LIFO is never used, because it is not acceptable for tax purposes; and the amortization period for R&D assets is based on the tax code (KPMG, 2001).
WKA it working capital accruals, ∆CA it -∆CL it + ∆STD it -∆CASH it , divided by A it-1 ; ∆REV it revenues during t minus revenues during t -1, divided by A it-1 ; ∆REC it receivables at the end of t minus receivables at the end of t -1, divided by A it-1 ; PPE it net property plant and equipment divided by A it-1 .WKA it replacing TA it on the left-hand side and without the PPE term on the right-hand side.
1;TA it total accruals divided by A it-1 ;

variable Constant AUD > 0 LST > 0 HILEV NEGSE SIZ CON > 0 N R 2 Adj-R 2 F statistic Panel A: Accounting Quality as Compliance with GAAP (CVM Instruction No. 346)
The sample consists of 290 fi rm observations in 1999 and 312 fi rm observations in 2000.The data is obtained from the CVM's website.Government-controlled fi rms, utilities and fi nancial institutions, as well as fi rms that were suspended from trading, bankrupt, did not fi le a funds fl ow statement, or were not operating at the end of 2000 are excluded.Means and medians are given for discretionary accruals, the magnitude of discretionary accruals, total assets, asset turnover,leverage, and profi tability.The hypothesis that the medians (means) of fi rms with Big 4 or local auditors are equal is verifi ed with the Kruskall-Wallis test (2-tailed t test)."***" indicates that both null hypothesis (of equal means and medians are rejected)."*" indicates that one of the hypothesis is rejected, but not the other.Models of Accounting Quality as Compliance with Brazilian GAAP and as Conservatism Breakdown of the total amnesty liability by auditor type and according to whether, prior to the amnesty(i,e., by year-end 1999), the liabilities that these companies included in the amnesty were: [P-type] recognized as payable on the balance sheet; [C-type] recognized as contingent either on the balance sheet or in the notes to the fi nancial statements; or [U-type] absolutely undisclosed.The 40 companies represented in this table are those that provided suffi cient information in the 2000 annual report for the estimation of the values of accepting and rejecting the amnesty.

Accounting Quality as Conservatism (Timeliness of Loss Recognition)
LST and CON are less or equal to zero, against the alternative that they are positive; and p-values for the hypothesis that each of the remaining coeffi cients is equal to zero, against the alternative that they are non-zero.CSCR score for compliance with CVM Instruction No. 346, measured by the ratio of points obtained to maximum possible decision-making points (or overall points), expressed as a percentage;TIM score given for timeliness, measured by the negative of the ratio of previously undisclosed tax liabilities to all tax liabilities; AUD 1 if auditor is Big 4 (AA, DT, EY, KPMG, PWC), otherwise 0; LST 1 if fi rm is listed on Bovespa or any U.S. exchange, otherwise 0;HILEV 1 if company is in top decile of debt to total assets at year-end, otherwise 0; NEGSE 1 if shareholders' equity is negative at year-end, otherwise 0; SIZ fi rm size, measured by the natural log of total assets at year-end;CON concentration of control, given by percentage of common and preferred shares owned by controlling shareholders.