Do initial public offering firms understate the allowance for bad debts?

DO INITIAL PUBLIC OFFERINGFIRMS UNDERSTATE THEALLOWANCE FOR BAD DEBTS?

Scott B . Jackson, William E . Wilcox and

Joel M . Strong

ABSTRACT

In this study, we investigate whether initial public offering (IPO) firmsunderstate the allowance for bad debts in the two annual periodsadjacent to their IPOs. The evidence suggests that IPO firms understatethe allowance for bad debts in both periods, and that high qualityauditors have little effect on the extent to which the allowance for baddebts is understated. The evidence also indicates that the magnitude of theunderstatement is economically significant in relation to the recordedbalance in the allowance account. It is estimated that the mean (median)understatement of the allowance for bad debts by IPO firms is approxi-mately 40% (75%) and 35% (60%) of its recorded balance in the yearbefore and year after IPO, respectively .

1. INTRODUCTION AND MOTIVATION

Accounting information is provided to the investing public by firm managerswho have situation specific incentives to alter the profile of that information .Although generally accepted accounting principles (GAAP) and external

Advances in Accounting, Volume 19, pages 89-118 .Copyright © 2002 by Elsevier Science Ltd .All rights of reproduction in any form reserved.ISBN: 0-7623-0871-0

89

90

SCOTT B. JACKSON, WILLIAM E. WILCOX AND JOEL M. STRONG

auditors reduce the amount of discretion exercised by firm managers, the accrualaccounting system mandated by GAAP nonetheless affords them a substantialamount of accounting discretion . The purpose of this study is to explore whethermanagers of initial public offering (IPO) firms use the flexibility inherent inGAAP to systematically understate the allowance for bad debts in the twoannual periods adjacent to IPOs in order to bolster earnings and assets .'

This study is distinguished from prior research on earnings management by IPOfirms in the following ways. First, most prior studies focus on total accruals(Aharony et al ., 1993 ; Friedlan, 1994 ; Teoh et al., 1998b), while this study focuseson a single accrual account . As discussed below, there are several advantagesassociated with focusing on a single accrual account rather than total accruals .Second, while Teoh et al . (1998a) examine the allowance for bad debts of IPOfirms, they do so in the context of a broad analysis of earnings management byIPO firms and perform simple univariate tests on the allowance account. Incontrast, our study is dedicated exclusively to analyzing the allowance for baddebts and we perform a focused, in-depth analysis of discretionary behavior withrespect to this accrual account . For example, unlike Teoh et al . (1998a), wedevelop empirical models to estimate the discretionary component of theallowance for bad debts which remove the portion of the allowance that is dictatedby accounts receivable and sales and the year-to-year change in those accounts .The univariate tests conducted by Teoh et al . (1998a) do not take these variablesinto account . Third, unlike Teoh et al . (1998a), we provide estimates of the eco-nomic significance of the discretionary component of the allowance for bad debts,and examine whether high quality auditors mitigate any tendency of IPO firms tounderstate this accrual account . Finally, we motivate our analysis of the allowancefor bad debts by highlighting why this account is a likely target for manipulation .'

This study focuses on the allowance for bad debts of IPO firms for twoprimary reasons. First, the allowance for bad debts is likely to be a singularlymaterial accrual account . Initial public offering firms usually experienceunprecedented increases in both sales and accounts receivable in the periodssurrounding IPOs so discretion over the allowance for bad debts could have amaterial impact on earnings and assets . In addition, anecdotal evidence in thefinancial press indicates that part of the increase in accounts receivable in theperiods surrounding IPOs may be the result of managers accepting low qualitycredit sales. Hall and Renner (1988) suggest that firms may "cut a few corners"to register sales just before IPOs, and Khalaf (1992) suggests that IPO firmsmay try to "dress up their (accounting) numbers" before IPOs, particularly thoserelated to sales . More generally, these and other articles (Browning, 1998 ;Schroeder, 1994) allege that IPO firms use the leeway in GAAP to inflateearnings and assets just before stock offerings .

Do Initial Public Offering Firms Understate the Allowance for Bad Debts?

91

Second, the allowance for bad debts requires professional judgment to deter-mine its balance, suggesting that managers have a substantial amount ofdiscretion over this accrual account . While auditors likely scrutinize financialstatement accounts involving managers' subjective judgment, it is unlikely thatthey fully counteract intentional or unintentional bias in those accounts . Rather,auditors are likely to develop a range of acceptable values for judgmentallydetermined accounts and cannot insist upon a particular point estimate withinthat range (Arens & Loebbecke, 1996 ; Pany & Whittington, 1997 ; Robertson,1996). If managers of IPO firms establish allowances that tend toward the lowerend of that range, intentional or unintentional bias is likely to persist despiteauditors' efforts to counteract it .

The results of this study are summarized as follows . In the two periodsadjacent to IPOs, the evidence reveals that the discretionary component of theallowance for bad debts is negative, suggesting that the allowance is under-stated. Not only is the understatement statistically significant, but it iseconomically significant in relation to the recorded balance in the allowanceaccount. It is estimated that the mean (median) IPO firm understates theallowance for bad debts by approximately 40% (75%) and 35% (60%) of itsrecorded balance in the year before and year after IPO, respectively . Thisfinding is consistent with the claim that managers of IPO firms use theflexibility afforded by GAAP to bolster earnings and assets, and that auditorsof IPO firms do not fully counterbalance intentional or unintentional bias inthe allowance for bad debts . In addition, the evidence does not suggest thathigh quality auditors mitigate the general tendency of IPO firms to understatethe allowance for bad debts. This finding suggests that the intensivemonitoring provided by high quality auditors has little incremental effect onthe accounts of IPO firms that involve highly subjective evidence . Furthermore,supplemental analyses reveal that the results are robust with respect to modelspecification, industry fineness, and the financial profile of existing publiccompanies used to develop proxies for the non-discretionary component of theallowance for bad debts of IPO firms .

Readers should note that economic significance is measured relative to therecorded balance in the allowance account. Economic significance could alsobe measured relative to total assets, in which case the understatement of theallowance account would be about one-half of 1% of total assets (see Panel Aof Table 4, DALL2) . 3 This observation does not mean that the results reportedin this paper are uninformative about whether earnings and assets of IPO firmsare manipulated. Indeed, when the evidence reported in this study is consid-ered in conjunction with the evidence reported in related studies, it is reasonableto conclude that earnings and assets are manipulated and that the allowance for

92


bad debts is one instrument through which managers achieve their financialreporting objectives. Nonetheless, readers should be cautious about concludingthat managers' discretion over the allowance account alone has a material effecton earnings and assets .

The remainder of this study is organized as follows . Section 2 reviewsrelevant prior research on earnings management . Section 3 discusses themotivations to manage accruals in the periods adjacent to IPOs and developsthe hypotheses that are tested in this study . Section 4 provides the models usedto estimate the discretionary component of the allowance for bad debts . Section5 reports sample selection procedures, descriptive statistics, primary results, andsupplemental analyses . Section 6 contains concluding remarks and discussessome limitations of this study .

2. PRIOR RESEARCH ON EARNINGSMANAGEMENT BY IPO FIRMS

Studies on earnings management by IPO firms include Friedlan (1994), Teohet al. (1998a, b), and Aharony et al . (1993) . The results of Friedlan (1994)suggest that IPO firms record positive discretionary accruals in the period(interim or annual, whichever was the last reported) immediately precedingIPOs, and the results of Teoh et al . (1998a) suggest that IPO firms recordpositive discretionary accruals in the annual period following IPOs . Teohet al. (1998a) also document an inverse relation between discretionary accrualsin the period following IPOs and subsequent earnings, which suggests thatdiscretionary accruals are opportunistic . Teoh et al. (1998b) find that the IPOfirms which most aggressively record discretionary accruals in the periodfollowing IPOs are the ones that experience the most severe post-IPO stockprice underperformance . Finally, Aharony et al . (1993) find weak evidenceof earnings management by IPO firms in the annual period immediatelypreceding IPOs .

Several studies on earnings management have also focused on a single accrualaccount. McNichols and Wilson (1988) focus on the provision for bad debts ofindustrial firms in three industries. Beaver and Engel (1996) focus on theallowance for loan losses in the banking industry. As part of a larger study onaccruals, Teoh et al . (1998a) perform a limited analysis of the allowance forbad debts of IPO firms.' Guidry et al . (1999) focus on the inventory reserveaccount in their study on earnings management by business unit managers . Allof these studies find evidence of discretionary behavior with respect to theaccount examined.


93

3. HYPOTHESES DEVELOPMENT

3.1. Motivations to Manage Accounting Information

The primary source of information about an IPO firm is its offering prospectus .This document contains audited financial statements which are consistentlymentioned as being useful in IPO pricing decisions (Richardson, 1976 ; Perez,1984; Bartlett, 1988 ; Weiss, 1988 ; Bloch, 1989 ; Blowers et al ., 1995) . Underthe plausible assumptions that: (1) managers are able to manipulate accountinginformation in the periods preceding stock offerings,' (2) IPOs are valued, inpart, by reference to information contained in the financial statements, and (3)underwriters and other investors do not fully adjust accounting numbers forbias or manipulation,6 it seems reasonable that accounting choices could havean effect on the offering proceeds received by the firm and its entrepreneurs .Thus, a primary motivation for firms to manage accounting information in thelast annual period preceding IPOs (referred to as year 0 in the remainder ofthis paper) is to influence the perceptions of investors about firm value in orderto obtain higher offering proceeds . Figure 1 illustrates the periods examined inthis study and the relation between those periods and the IPO date .

With respect to the first annual period after IPOs are consummated (referredto as year I in the remainder of this paper), Teoh et al . (1998a) discuss threereasons why firms might manage earnings . First, managers of IPO firms maybe under pressure from underwriters and investors to meet verbal earnings fore-casts (which are perhaps optimistic) made when marketing new issues . Bymeeting these forecasts, managers of IPO firms develop reputations withinvestors for reliability and potentially avoid lawsuits by disgruntled share-holders . Second, because there is a lock-up period for 180 days or longer afterthe offering date during which entrepreneurs agree not to sell their shares,managers may try to report higher earnings and asset values until the lock-upperiod expires to enhance their personal wealth. Third, underwriters practice

Last annual

IPO

F rst annualperiod before IPO

period after IPO

Year 0

Year I

Fig. 1 . Periods Examined in this Study .

94

SCOTT B . JACKSON, WILLIAM E. WILCOX AND JOEL M. STRONG

what is called "price stabilization" whereby they purchase stock of the IPO firmin the open market, often at the original issue price, to prevent or retard adecline in its open market price (Hanley et al ., 1993). Since this activity iscostly, underwriters might pressure IPO firms to manage accounting informa-tion after the IPO is consummated to support the stock price .

3.2. Understatement of the Allowance for Bad Debts

When credit is extended to a firm's customers, there is uncertainty about whethercollection will occur in the future . If uncollectible receivables are both prob-able and estimable, an allowance for bad debts must be recognized in accordancewith Statement of Financial Accounting Standards (SFAS) No. 5 and thematching principle . Unfortunately, there is a substantial amount of inexactnessand ambiguity inherent in the allowance for bad debts . As a result, this accountposes special problems for both management and auditors because subjectiveevidence is used to establish its balance .

Management is responsible for making the accounting estimates reflected inthe financial statements and auditors are responsible for evaluating the reason-ableness of those estimates . Auditors may gain satisfaction that the allowancefor bad debts is adequate by, among other things, confirming accounts, evalu-ating the client's credit policies, examining customer credit files, analyzingwrite-off experience, and examining cash collections . Despite a variety ofauditing techniques and procedures, auditors must nonetheless rely to someextent on managements' representations because the allowance for bad debtsinvolves subjective evidence. As a result, auditors cannot entirely eliminatediscretion in accounting estimates because subjective evidence is difficult orimpossible to verify . In addition, Kreutzfeldt and Wallace (1986) document thataccounts receivable is prone to error and that judgmental errors are more preva-lent in this account than in other current asset accounts. Judgmental errors areconsistent with attempts by management to manipulate earnings (DeFond &Jiambalvo, 1994) .

Moreover, auditors can only develop a range of acceptable values for anaccounting estimate and usually cannot insist upon a particular point estimatewithin that range (Arens & Loebbecke, 1996 ; Pany & Whittington, 1997 ;Robertson, 1996) . Indeed, auditors are likely to identify unreasonable accountingestimates but may not isolate instances where management introduces inten-tional or unintentional bias into accounting estimates that is not of an egregiousor erroneous nature . For example, management may shade the allowance forbad debts towards the lower bound of a range of reasonable values . In suchinstances, auditors may find it difficult to support the position that the allowance


95

for bad debts is materially understated. Moreover, auditors are not expected tosubstitute their judgment for that of management when auditing accountsinvolving subjective evidence .

Auditing standards recognize the subjectivity inherent in many accountingestimates and the risks associated with them. For example, Statement onAuditing Standards (SAS) No . 57 suggests that the risk that management wouldor could misstate an account balance increases with the subjectivity involvedin determining that balance . When accounting numbers involve judgment andsubjective evidence, reasonable individuals can come to justifiably differentconclusions given identical information . In addition, IPO firms pose specialproblems because they commonly have short operating histories and largeincreases in sales and accounts receivable, which exacerbate the conventionalproblems associated with auditing the allowance for bad debts. As a result,intentional or unintentional bias in the allowance for bad debts is likely topersist despite auditors' efforts to counteract it .

Collectively, the discussion in this section indicates that managers of IPOfirms have the incentives, opportunity, and ability to manage the allowance forbad debts. This discussion gives rise to the following testable hypothesis (statedin alternative form) :

Hypothesis 1 : The allowance for bad debts of IPO firms is understated (i .e .low relative to a benchmark for what the allowance for bad debts should beabsent managerial discretion) in years 0 and 1 .

3.3. Understatement of the Allowance and High Quality Auditors

Researchers often argue that Big Five accounting firms provide higher qualityaudits than their non-Big Five counterparts, because Big Five firms have incen-tives to protect their investments in reputation capital (DeAngelo, 1981 ; Francis& Wilson, 1988) . Consistent with the desire to protect their reputations, BigFive accounting firms are more likely to: (1) use audit tests and procedurescapable of identifying understatements of the allowance for bad debts, (2)develop a narrower range of acceptable values for the allowance for bad debts,thereby reducing intentional or unintentional bias, and (3) report disagreementsif the client fails to make necessary adjustments to the allowance for bad debts .Thus, although hypothesis 1 predicts that the allowance for bad debts will beunderstated in years 0 and 1, we expect that the understatement is less for firmsaudited by Big Five accounting firms than for firms audited by non-Big Fiveaccounting firms . 7 This discussion gives rise to the following testable hypoth-esis (stated in alternative form) :

96


Hypothesis 2 : The understatement of the allowance for bad debts of IPOfirms is less negative in years 0 and 1 for firms audited by Big Five audi-tors than for firms audited by non-Big Five auditors .

4. RESEARCH DESIGN

4.1. Measurement of the Discretionary Component of the Allowance

Conceptually, the allowance for bad debts (ALL) can be partitioned into adiscretionary (DALL) and non-discretionary (NALL) component :

ALL = DALL + NALL. 8

(1)

Because both DALL and NALL are unobservable, an estimate of one of thetwo components is needed . We develop two expectation models for the non-discretionary component of the allowance for bad debts (NALL1 and NALL2)and use them along with ALL to derive the corresponding estimates of thediscretionary component of the allowance for bad debts (DALL1 and DALL2) .

In developing the empirical models to estimate the discretionary componentof the allowance for bad debts, we adopt a balance sheet emphasis becauseprior research (McNichols & Wilson, 1988) and auditing texts (Arens &Loebbecke, 1996; Pany & Whittington, 1997 ; Robertson, 1996) indicate thatmanagers and auditors are more concerned with proper balance sheet valuationthan with matching the provision to current revenues . However, as discussedin Section 5 .4, we perform additional tests which incorporate income statementinformation into the models to assess the robustness of the results .

The first expectation model assumes that the non-discretionary component ofthe allowance for bad debts of IPO firm i in year t is equal to the mean allowancefor bad debts (stated as a percentage of gross trade accounts receivable (AR))of existing public companies in the same industry (same two-digit SIC code),but excluding the IPO firm and all other firms that went public in the previousfive years."' The first expectation model is expressed as follows :

NALL1 A = mean(ALL,/AR t ),

(2)

where j is a firm index for the industry estimation sample. Two-digit SIC codematching is used if the number of firms in the industry estimation sample issix or greater, otherwise one-digit SIC code matching is used . I' Prediction errorsrepresent the discretionary component of the allowance for bad debts, DALL1,and are defined as follows :

DALLl it = ALL,t/AR1t - NALL% .

(3)


97

The second expectation model assumes that the non-discretionary componentof the allowance for bad debts of IPO firm i in year t is a linear function ofgross trade accounts receivable and the year-to-year change in gross tradeaccounts receivable (DAR) . To estimate this relationship, we use existing publiccompanies in the same industry (same two-digit SIC code) as the IPO firm, butexclude the IPO firm and all other firms that went public in the previous fiveyears, to estimate the following cross-sectional regression :

ALL~ t. /TA . t = aOt(l/TAJt.) + aIt

~(AR .t

~t/TA.) + a 2t

~t

~t(DAR. /TA .) + its . ,

(4)~where TA is total assets . 12 Ordinary least squares is used to obtain estimatesaot, a lt , and alt of aot , a lt , and a lt , respectively. Equation (4) is scaled by contem-poraneous total assets in an attempt to reduce heteroskedasticity . 13 Two-digitSIC code matching is used if the number of firms in the industry estimationsample is 15 or greater, otherwise one-digit SIC code matching is used ." Thenon-discretionary component of the allowance for bad debts for IPO firm i inyear t is defined as follows :

NALL2,t = aot(l/TAit) + a it(AR)TAit) + a2t(DAR t/TA, t ) .

(5)

Prediction errors represent the discretionary component of the allowance forbad debts, DALL2, and are defined as follows :

DALL2 it = ALLIt/TA it - NALL2 i

(6)

This model assumes that the fitted allowance for bad debts is the amount neces-sary to state gross trade accounts receivable at its net realizable value and thatthe prediction errors primarily reflect managerial accounting discretion .

In Eq. (4) the coefficient on AR is expected to be positive because a largerbalance in gross trade accounts receivable should require a larger balance in theallowance for bad debts . The coefficient on LAR is expected to be negative . Ourintuition is that managers may not fully and immediately adjust the allowance forbad debts in response to year-to-year changes in accounts receivable . In otherwords, increments and decrements to accounts receivable are expected to have asmaller impact on the allowance for bad debts than the beginning balance inaccounts receivable . This expectation is supported by research in psychologyrelated to the anchor and adjustment heuristic (Tversky & Kahneman, 1974) .Specifically, the anchor is the prior year allowance for bad debts and the adjustmentis the increment or decrement to the opening balance of the allowance account .Research suggests that the anchor takes on great psychological importance in manydecision contexts and that the adjustment to the anchor is often inadequate(Tversky & Kahneman, 1974), which supports our expectation that the year-to-year change in accounts receivable will have a negative coefficient .

98


By including both AR and DAR in Eq . (4), we attempt to remove the non-discretionary component of the allowance for bad debts and isolate that portionof the allowance which reflects managerial discretion . As discussed later, theregression results support our expectations concerning the signs on AR andAAR.'s

4.2. Statistical Tests of the Discretionary Component of the Allowance

Tests of significance are computed using standardized (DALL2) and unstan-dardized (DALL1 and DALL2) prediction errors and percentage predictionerrors from Eqs (3) and (6) . Following Defond and Jiambalvo (1994), stan-dardized prediction errors are computed as

Vit = DALL21/s(ej. t ),

(7)

where s(e,. t ) is the standard deviation of the error term from the cross-sectionalregression estimated using existing public companies in the same industry asIPO firm i in year t . Parametric significance tests of the standardized predic-tion errors are computed as

ZVt = Y.V;t/[(YI; - k)/(11 - (k + 2))]u2 (8)

where Ii is the number of firms in the estimation portfolio for IPO firm i andk is the number of parameter estimates in the model . Both parametric and non-parametric tests of standardized and unstandardized prediction errors arereported. The unstandardized prediction errors measure the discretionary compo-nent of the allowance for bad debts as a percentage of either gross trade accountsreceivable (DALL1) or total assets (DALL2) . Percentage prediction errorsmeasure the discretionary component of the allowance for bad debts as apercentage of the recorded balance in that account . They are defined for model1 (PERI) and model 2 (PER2) as follows :

Percentage prediction errors provide a convenient way to assess the economicsignificance of the discretionary component of the allowance for bad debtsbecause they express the magnitude of the understatement in relation to therecorded balance in the allowance for bad debts."

PERI« = (ALL 1t/AR1t - NALLI 1t)/ALL I,/AR, t , (9)

PER21t = (ALLit1TA1t - NALL2, t)/ALL1t/TA1t . 16 (10)


99

5. SAMPLE SELECTION, DESCRIPTIVE STATISTICS,AND RESULTS

5.1. Sample Selection

A list of all IPOs occurring from 1980 through 1984 (n = 2,396) was obtainedfrom Jay Ritter's IPO database . 18 Firms on this list had to be covered byCompustat in either year 0 or year 1 to be included in this study . In order toconduct the analyses relating to model 2 in year 0, we had to obtain financialstatement information related to year -1 . As Compustat begins coverage of IPOfirms in year 0, data had to be manually collected from prospectuses . Microfichecopies of prospectuses were obtained from other researchers and the Universityof Texas-Austin library . Table 1 summarizes the sample selection proceduresand the resulting sample sizes for the statistical tests .

Table 1 . Description of Sample Selection .

° Expectation models 1 and 2 are described in Section 4 .1 of the text.Year 0 is the last annual accounting period before an IPO occurs and year 1 is the first annual

accounting period after an IPO is consummated .In order to conduct the analyses in year 0 for model 2, financial statement data for year -1 is

required . Since Compustat does not begin coverage of IPO firms until year 0, microfiche copiesof prospectuses had to be obtained in order to obtain the required data items .I Auditor type (Big Five/non-Big Five) was obtained from Ritter's IPO database . Some firms hadmissing data for this variable so it was manually collected from microfiche copies of availableprospectuses. Auditor type could not be determined for some sample firms because prospectuseswere not available .

Model 1 1 Model 2

Year 0' Year I Year 0 Year 1

Number of IPOs occurring from 1980through 1984 listed in Ritter's IPO database 2,394 2,394 2,394 2,394

Less : Number of firms either not covered byCompustat or not reporting required data items 1,827 1,692 1,871 1,842

Less: Number of firms for which microfiche copiesof prospectuses could not be obtained 223`

Number of firms used to test hypothesis 1 567 702 305 552

Less : Number of firms for which the auditorcould not be determined ° 8 10 8

Number of firms used to test hypothesis 2 559 692 305 544

100


5.2. Descriptive Statistics

Table 2 reports descriptive statistics for receivables-related variables of samplefirms. The information in Table 2 is partitioned by empirical model (model 1or model 2) and year (year 0 or 1), since a different number of firms is usedfor each model year . The mean (median) ratio of allowance for bad debts totrade accounts receivable is approximately 4% (2.4%) and 4.8% (2 .7%) in years0 and 1, respectively, for both models . The mean (median) allowance for baddebts as a percentage of total assets ranges from a low of 1 .03% (0.53%) to ahigh of 1 .24% (0.72%), depending on the year and model . The mean (median)ratio of trade accounts receivable to total assets in year 0 for both models isapproximately 32% (29%) and the mean (median) ratio of trade accounts receiv-able to total assets in year 1 for both models is approximately 24% (21%) .With respect to the latter two ratios, the cause of their decline between years0 and 1 is the inclusion of IPO proceeds in total assets in year 1 .

Table 3 reports summary statistics for Eqs (2) and (4). With respect to mean(ALUAR) (Eq. (2)), it is estimated using all firms in the same two-digit SIC codeas the IPO firms. There are 567 and 702 sample firms in years 0 and 1 (see Table

Table 2 . Descriptive Statistics for Receivables-Related Variables ofSample Firms .

a Expectation models 1 and 2 are described in Section 4 .1 of the text.b Year 0 is the last annual accounting period before an IPO occurs and year 1 is the first annualaccounting period after an IPO is consummated .Components of these ratios are defined as follows : allowance is the allowance for bad debts ;

receivables is gross trade accounts receivable ; assets is total assets .

Model l a Model 2

Year 0b Year 1 Year 0 Year 1

Allowance/Receivables`Mean 0.0401 0.0478 0.0409 0 .0470Median 0.0242 0.0273 0.0265 0 .0281Standard deviation 0.0577 0.0752 0 .0543 0 .0657

Allowance/AssetsMean 0.0124 0.0103 0.0117 0 .0108Median 0.0063 0.0053 0 .0072 0 .0056Standard deviation 0.0238 0.0182 0 .0164 0 .0192

Receivables/AssetsMean 0.3200 0.2361 0.3208 0.2449Median 0.2934 0.2120 0.2952 0 .2184Standard deviation 0.2320 0.1597 0.1766 0.1611

Number of observations 567 702 305 552


101

1), respectively, resulting in 1,269 industry observations . 19 Panel A reveals that themean (median) fraction of trade accounts receivable reserved at the two-digit SICcode level is 5.55% (4.64%). With respect to matched(ALL/AR), it is estimatedfor each IPO firm by averaging ALL/AR of the two four-digit SIC code matches .The motivation for computing matched(ALL/AR) and using it to estimate thediscretionary component of the allowance for bad debts of IPO firms is discussedin Section 5 .4 (supplemental analyses) . Panel A reveals that the mean (median)fraction of trade accounts receivable reserved at the four-digit SIC code level forindustry matches is 6 .04% (4.02%). A comparison of the descriptive statisticsreported in Tables 2 and 3 reveals that IPO firms reserve a smaller fraction of theirgross trade accounts receivable than their publicly-owned counterparts .

Table 3. Summary Statistics for Models I and 2 .

I Variables are defined as follows : ALL is the allowance for bad debts ; AR is gross trade accountsreceivable ; TA is total assets ; AAR is year-to-year change in AR ; j is a firm index for the industryestimation sample ; t is a time index .b Mean (ALL/AR) and matched (ALL/AR) represent the mean non-discretionary component of theallowance for bad debts of all IPO firms in years 0 and 1 combined . With respect to mean (ALL/AR)for a particular IPO firm, it is estimated using all firms in the same two-digit SIC code as the IPOfirm . See Section 4.1 of the text . With respect to matched (ALL/AR) for a particular IPO firm, itis estimated by averaging two four-digit SIC code matched firms . See Section 5 .4 of the text .` Amounts are obtained by averaging regression statistics from industry-by-industry regressions inyears 0 and 1. The number of observations represents the average number of firms used toestimate the industry-by-industry regressions .

Variable' Mean Std. Dev . 1st

Quartiles

2nd 3rd

Panel A : Summary Statistics for Model I (Eq. (2)) and Modified Model P

Mean(ALL /AR,) 0.0555 0.0264 0.0395 0.0464 0.0625Number of observations 1,269Matched(ALL,/AR,)b 0.0604 0 .0644 0.0243 0.0402 0.0693Number of observations 1,269

Panel B: Summary Statistics for Model 2 (Eq . (4))'

-0.0001 0.0070 0.0209a01 (1ITAi1) 0 .0156 0 .0357t-statistic 2 .26 6 .82 0.01 1 .29 3 .07a ir (AR,/TA,.,) 0.0513 0 .0339 0.0337 0.0412 0.0607t-statistic 7 .85 3 .91 4.95 7 .45 10 .24a2 (AAR,/TA,,) -0.0317 0 .1056 -0.0532 -0.0296 -0.0059t-statistic -2.22 3 .31 -3 .36 -1 .78 -0.30Adjusted R 2 0 .59 0 .17 0.48 0.60 0 .70Number of observations 93 74 34 64 131

102


Panel B of Table 3 also reports summary regression statistics related toEq. (4) . These cross-sectional regression statistics were calculated by averagingrelevant Eq. (4) amounts across industries represented in the sample . Asexpected, the average coefficient on AR/TA is positive and highly significant(average t-statistic 7 .85). Also as expected, the average coefficient on DAR/TAis negative and significant (average t-statistic -2.22) . The independent variablesin Eq. (4) explain much of the variation in the allowance for bad debts, asrevealed by the average R 2 of 59%. The strength of these results suggest thatthe independent variables in Eq . (4) control for factors driving the allowancefor bad debts and that the difference between the observed allowance for baddebts of IPO firms and their fitted allowance for bad debts approximately isolatesthe discretionary component of the allowance for bad debts .

5.3. Primary Results

Hypothesis 1 predicts that the allowance for bad debts of IPO firms is under-stated (i .e . the discretionary component of the allowance for bad debts isnegative) in years 0 and 1 . Prediction errors from model 1 (DALL1) and model2 (DALL2) proxy for the discretionary component of the allowance for baddebts. Table 4 contains an analysis of prediction errors (Panel A) and percentageprediction errors (Panel B) relating to models 1 and 2 in years 0 and 1 . Thefirst two columns report prediction errors for model 1, the middle 2 columnsreport unstandardized prediction errors for model 2, and the last two columnsreport standardized prediction errors for model 2 . The bottom two rows in eachpanel of Table 4 report parametric (t-test) and non-parametric (Wilcoxon test)p-values (one-tailed) for tests of whether the prediction errors are significantlynegative .

The results reported in Panel A of Table 4 reveal that the mean predictionerrors are significantly negative in years 0 (-0 .0155, p < 0.0001) and 1 (-0 .0076,p < 0.0001) for model 1 . Panel A also reveals that the mean unstandardized pre-diction errors are significantly negative in years 0 (-0 .0083, p < 0.0006) and 1(-0.0039, p < 0 .0001) for model 2 . Similar results are reported for meanstandardized prediction errors in Panel A . Notice that the minimum andmaximum values reported in Panel A are of a sufficient magnitude that they couldhave an undue influence on mean prediction errors . As a result, the non-parametric Wilcoxon test is used to assess whether median prediction errors aresignificantly negative . This test reveals that median prediction errors aresignificantly negative in years 0 (-0 .0245, p < 0.0001) and 1 (-0 .0227,p < 0.0001) for model 1 . Panel A of Table 4 also reveals that median unstan-dardized prediction errors are significantly negative in years 0 (-0 .0042,


103

Table 4. Analysis of Prediction Errors From Models 1 and 2 inYears 0 and 1 .

Prediction errors are computed using the procedures described in Section 4.1 of the text.Standardized prediction errors and percentage prediction errors are computed as described in Section4 .2 of the text .' Year 0 is the last annual accounting period before an IPO occurs and year 1 is the first annualaccounting period after an IPO is consummated .Percentage prediction errors are not computed for these values because there is no obvious inter-

pretation of the percentages .a The parametric p-values for unstandardized prediction errors and percentage prediction errors areone-tailed t-tests. The parametric p-values for standardized prediction errors are also one-tailed andare derived as described in Section 4 .2 of the text .The non-parametric p-values are obtained from one-tailed Wilcoxon tests .Negative prediction errors and negative percentage prediction errors suggest that the allowance

for bad debts is understated. Prediction errors and percentage prediction errors are defined inSections 4 .1 and 4.2 of the text, respectively .

Prediction Errors

Model 1 (DALL1)Model 2 (DALL2,Unstandardized)

Model 2 (DALL2,Standardized)

Year 0' Year 1 Year 0 Year 1 Year 0 Year 1

Panel A : Prediction Errors

Mean

-0.0155 -0.0076 -0.0083 -0.0039 -0.3619 -0.0969Median

-0.0245 -0.0227 -0.0042 -0.0026 -0.3526 -0.1878Standard deviation

0.0604 0.0765 0.0437 0.0221 2 .0265 1 .5101Minimum

-0.4247 -0.2148 -0.6464 -0.1691 -17.5441 -8.8983Maximum

0.5775 0.8858 0 .1136 0.2216 10 .3299 17 .2969Number positive

143 186 88 191 88 191Number negative

424 516 217 361 217 361Parametric p-value°

0.0001 0.0044 0.0006 0.0001 0.0001 0.0125Non-parametric p-value''

0.0001 0.0001 0.0001 0.0001 0 .0001 0.0001

Panel B: Percentage Prediction Errors'

-0.4330 -0.3561Mean

-0.5197 -0.4812Median

-0.8800 -0.8216 -0.7699 -0.6078Standard deviation

0.5983 0.6210 0 .6642 0 .6976Minimum

-1.0000 -1.0000 - 1 .0000 -1 .0000Maximum

0.9545 0 .9628 0.9500 1 .0000Number positive

143 186 88 191Number negative

424 516 217 361Parametric p-value'

0.0001 0.0001 0 .0001 0.0001Non-parametric p-value ,

0.0001 0.0001 0.0001 0.0001

104


p < 0.0001) and 1 (-0.0026, p < 0.0001). Similar results are reported for medianstandardized prediction errors . Consistent with expectations, the results reportedin Panel A of Table 4 strongly support the view that the allowance for bad debtsis understated in years 0 and 1 .

While the discretionary component of the allowance for bad debts is statis-tically significant, it may not be economically significant . To assess economicsignificance, Panel B of Table 4 provides percentage prediction errors . Thesevalues express the discretionary component of the allowance for bad debts asa percentage of its recorded ending balance . The mean (median) percentageprediction error for model 1 is approximately -52% (-88%) and -48%(-82%) in years 0 and 1, respectively . Similarly, the mean (median) percentageprediction error for model 2 is approximately -43% (-77%) and -36%(-61%) in years 0 and 1, respectively. The mean and median percentageprediction errors are significantly negative (p < 0 .0001) for both models in bothyears .2o

It is reassuring to note that both models yield plausible and similar estimatesof the percentage by which the allowance for bad debts is understated . However,it is probably inappropriate to view percentage prediction errors as preciseestimates. In addition, the percentage estimates from model 2 are arguably morereliable than those from model 1 because model 2 better controls for non-discretionary factors that determine the allowance for bad debts . With theseconsiderations in mind, we estimate that the mean (median) IPO firmunderstates the allowance for bad debts by approximately 40% (75%) and 35%(60%) of its recorded balance in the year before and year after the IPO,respectively . This discussion supports the contention that the allowance for baddebts of IPO firms is materially understated in the periods adjacent to IPOs .

Hypothesis 2 predicts that the understatement of the allowance for bad debtsof IPO firms is less negative for firms audited by Big Five auditors than forfirms audited by non-Big Five auditors . To test this hypothesis, the analysisreported in Table 5 was conducted. The first two columns of Table 5 reportand analyze prediction errors in years 0 and 1 of IPO firms audited by Big Fiveauditors, and the second two columns of Table 5 report and analyze predictionerrors in years 0 and 1 of IPO firms audited by non-Big Five auditors . The lasttwo columns of Table 5 report p-values for differences between mean andmedian prediction errors of IPO firms audited by Big Five and non-Big Fivefirms in years 0 and 1 .

Panels A and B of Table 5 report the tests of hypothesis 2 using predictionerrors and percentage prediction errors, respectively, from model 1 (DALL1) .In Panel A, the difference between means is insignificant in years 0 (p = 0 .7429)and 1 (p = 0.985 1) . The difference between medians is marginally significant


105

in year 0 (p = 0.0771) and insignificant in year 1 (p = 0 .7594) . Similar resultshold for differences in percentage prediction errors reported in Panel B . PanelC and D test hypothesis 2 using prediction errors and percentage predictionerrors, respectively, from model 2 . In Panel C, the difference between means

Table 5 . Analysis of Prediction Errors From Models 1 and 2 in Years 0and 1 Partitioned by Auditor Type

Panel B : Model 1 Percentage Prediction Errors (DALLI) Partitioned by Auditor Type'

Panel C: Model 2 Prediction Errors (DALL2, Unstandardized) Partitioned by Auditor Type

Prediction Errors P-value for Diff. in

Big Five Non-Big Five Means/Medians,Year 0b

Year 1 Year 0

Year 1 Year 0 Year 1

Panel A : Model 1 Prediction Errors (DALLI) Partitioned by Auditor Type

Mean -0.0168 -0.0138 -0.0123

0.0036 0.7429 0 .9851Median -0.0226 -0.0231 -0.0296

-0.0220 0.0771 0.7594Standard deviation 0.0515

0.0505 0.0791

0.1102Minimum -0.4245 -0.1969 -0.1015

-0.2148Maximum 0.2448

0.2205 0.5775

0.8858Number positive 104

123 35

58Number negative 297

359 123

152Parametric p-value` 0 .0001

0.0001 0.0259

0.6358Non-parametric p-value'' 0 .0001

0.0001 0.0001

0.0001

Mean -0.5087 -0.5042 -0.5575 -0.4444 0.1923 0 .8789Median -0.8003 -0.8306 -0.9056 -0.7911 0.0881 0 .7753Standard deviation 0.5918 0.6008 0.6089 0.6547Minimum -1 .0000 -1.0000 -1.0000 -1 .0000Maximum 0.8103 0.8327 0.9450 0.9628Number positive 104 123 35 58Number negative 297 359 123 152Parametric p-value` 0 .0001 0.0001 0.0001 0 .0001Non-parametric p-value'' 0 .0001 0.0001 0.0001 0 .0001

Mean -0.0081 -0.0032 -0.0090 -0.0061 0.4157 0 .1065Median -0.0041 -0.0026 -0.0043 -0.0028 0.2412 0 .1771Standard deviation 0.0482 0.0207 0.0207 0.0256Minimum -0.6464 -0.1001 -0.0930 -0.1691Maximum 0.1136 0.2216 0.0268 0.1107Number positive 73 141 15 46Number negative 166 249 51 108Parametric p-value° 0 .0051 0 .0013 0.0004 0.0019Non-parametric p-value" 0.0001 0 .0001 0.0001 0 .0001

106

SCOTT B . JACKSON, WILLIAM E. WILCOX AND JOEL M . STRONG

Table 5. Continued

Prediction Errors'

P-value for Diff. inBig Five

Non-Big Five

Means/Medians'Year 0b

Year 1

Year 0

Year 1

Year 0

Year 1

Panel D: Model 2 Percentage Prediction Errors (DALL2, Unstandardized) Partitioned byAuditor Typet

Panel E.• Model 2 Prediction Errors (DALL2, Standardized) Partitioned by Auditor Type

a Prediction errors are computed using the procedures described in Section 4.1 of the text .Standardized prediction errors and percentage prediction errors are computed as described in Section4 .2 of the text .b Year 0 is the last annual accounting period before an IPO occurs and year 1 is the first annualaccounting period after an IPO is consummated .The parametric p-values for unstandardized prediction errors are derived using one-tailed t-tests .

The parametric p-values for standardized prediction errors are one-tailed and are derived using theprocedures described in Section 4 .2 of the text .d The non-parametric p-values are obtained from one-tailed Wilcoxon tests.I The p-values for differences between means are obtained using one-tailed t-tests . The p-valuesfor differences between medians are obtained using one-tailed Wilcoxon tests .t Negative prediction errors and negative percentage prediction errors suggest that the allowancefor bad debts is understated. Prediction errors and percentage prediction errors are defined inSections 4.1 and 4 .2 of the text, respectively .

Mean -0.4168 -0.3402 -0.4919 -0.4169 0.2084 0 .1241Median -0.7001 -0.4665 -0.9035 -0.8083 0.2236 0 .1183Standard deviation 0.6727 0 .6791 0.6336 0 .6966Minimum -1 .0000 -1 .0000 -1 .0000 -1 .0000Maximum 1 .0000 1 .0000 1 .0000 1 .0000Number positive 73 141 15 46Number negative 166 249 51 108Parametric p-value 0.0001 0.0001 0 .0001 0.0001Non-parametric p-value° 0.0001 0.0001 0 .0001 0.0001

Mean -0.2567 -0.0305 -0.7431 -0.2788 0.0671 0 .0310Median -0.3315 -0.1655 -0.4415 -0.2282 0.1140 0 .0793Standard deviation 1 .8971 1 .5926 2 .4164 1 .3055Minimum -14.8246 -4.2643 -17.5441 -8.8983Maximum 10.3299 17 .2969 3 .8903 3.9637Number positive 73 141 15 46Number negative 166 249 51 108Parametric p-value` 0.0001 0.2776 0.0001 0 .0003Non-parametric p-valued 0.0001 0.0001 0 .0001 0 .0001


1 07

is insignificant in years 0 (p = 0.4157) and I (p = 0 .1065). The differencebetween medians is also insignificant in years 0 (p = 0 .2412) and 1 (p = 0.1771) .Qualitatively similar results hold for percentage prediction errors reported inPanel D. Finally, Panel E of Table 5 tests hypothesis 2 using standardizedprediction errors from model 2. The difference between means is marginallysignificant in year 0 (p = 0 .067 1) and significant in year 1 (p = 0 .03 10) . Thedifference between medians is insignificant in year 0 (p = 0 .1140) and margin-ally significant in year 1 (p = 0 .0793). Taken together, the evidence reported inall panels of Table 5 provides little support for the hypothesis that Big Fiveauditors mitigate the general tendency of IPO firms to understate the allowancefor bad debts .

5.4. Supplemental Analyses

The primary analyses reported in this paper are based on the two models devel-oped in Section 4 .1 . This section discusses two important assumptions of thosemodels and analyzes whether the inferences of this study are sensitive to them .First, we assume that industry can be accurately defined at the two-digit SICcode level, although there is significant diversity across firms within thatindustry definition. Second, we assume that Eq . (4) is properly specified,although the allowance for bad debts may be jointly determined by both accountsreceivable and sales . Having highlighted these assumptions, the remainder ofthis section is devoted to analyzing whether the main results of this study aresensitive to them .

We begin the analysis by selecting two industry matched firms for each IPOfirm. These firms are chosen by identifying two existing public companies inthe same four-digit SIC code as the IPO firm that have the closest ratio of tradeaccounts receivable to total assets (referred to as the AR ratio hereafter) . Thefirst public firm in these sets has an AR ratio just above the IPO firm's ARratio and the second public firm has an AR ratio just below the IPO firm'sAR ratio ." This matching procedure is implemented in both years 0 and 1 andis implemented without replacement . In addition, unlike the main analysis whichrequires that existing public companies be at least five years old, the industrymatched firms must only be at least two years old . This should help mitigateproblems associated with using relatively mature public firms to estimate thenon-discretionary component of the allowance for bad debts of IPO firms .

With respect to the first expectation model, the non-discretionary (NALL1, seeEq. (2)) and discretionary (DALLI, see Eq . (3)) components of the allowancefor bad debts were re-computed using the sets of industry matched firms ratherthan industry means . The benefits of this analysis are twofold . First, because we

108

SCOTT B . JACKSON, WILLIAM E . WILCOX AND JOEL M. STRONG

define industry at the four-digit level instead of the two-digit level, there is greatercomparability between IPO firms and existing public companies . Second,because we match on the AR ratio, only existing public companies withfinancial profiles similar to that of IPO firms are used to estimate the non-discretionary component of the allowance for bad debts of IPO firms .

The results of the above analysis are reported in Table 6 . Panel A analyzesmodel 1 prediction errors in years 0 and 1 . The mean (median) prediction erroris -0.0185 (-0 .0134) and -0 .0140 (-0 .0135) in years 0 and 1, respectively .The mean (median) percentage prediction error is -40.76% (-70.16%)and -35.88% (-61 .55%) in years 0 and 1, respectively . Not only are theprediction errors and percentage prediction errors in Table 6 significant in allcases (p < 0 .0001), but they are comparable to those reported in Table 4 . Withrespect to the main results discussed in Section 5.3 and reported in Table 4,we therefore conclude that they are unaffected by industry fineness and thefinancial profile of existing public companies used to estimate the non-discretionary component of the allowance for bad debts .

With respect to the second expectation model, we estimate several cross-sectional regressions similar to Eq . (4) to evaluate whether the main results aresensitive to alternative model specifications . These regressions contain both IPOfirms and the sets of industry matched firms selected according to theprocedures described above . As shown in Table 7, these regressions includedifferent combinations of trade accounts receivable and net sales and year-to-year changes in those accounts . The regressions also contain timefixed-effects (not reported) and a test variable (IPOD) which is coded as 1 forIPO firms and 0 for industry matched firms . The coefficient on IPOD is expectedto be negative since IPO firms are expected to have lower fractions of theallowance for bad debts reserved than their publicly-owned counterparts .

Panels A and B of Table 7 report the results of pooled cross-sectionalregressions for years 0 and 1, respectively. Regressions in both panels are highlysignificant (p < 0 .0001) and explain a substantial amount of the variation in thedependent variable . As expected, the coefficient on IPOD is significantlynegative (p < 0 .01, one-tailed) in year 0 for all specifications . Similar resultsare reported in Panel B of Table 7, although the coefficient on IPOD in thesecond and sixth regressions is only significant at about the 0 .06 level(one-tailed). The coefficient on IPOD in the remaining regressions issignificant at the 0 .025 level (one-tailed) or better . These results suggest thatalternative specifications of Eq. (4) are unlikely to yield results that arematerially different from those reported in Table 4 .

Finally, we examine whether firm and offering characteristics of IPO firmsare associated with the discretionary component of the allowance for bad debts


Table 6. Analysis of Model 1 (DALL1) Prediction Errors Using IndustryMatched Firms .'

109

° Industry matched firms are chosen by identifying two existing public companies in the same four-digit SIC code as the IPO firm that have the closest ratio of trade accounts receivables to totalassets (referred to as the AR ratio). The first public firm in these sets has an AR ratio just abovethe IPO firm's AR ratio and the second public firm has an AR ratio just below the IPO firm'sratio . Using these sets of industry matched firms, mean (ALL/AR) is calculated for each IPO firm,which represents the non-discretionary component of the allowance for bad debts . See Section 5 .3of the text .I Year 0 is the last annual accounting period before an IPO occurs and year 1 is the first annualaccounting period after an IPO is consummated.The parametric p-values for prediction errors are derived using one-tailed t-tests.

I The non-parametric p-values are obtained from one-tailed Wilcoxon tests .Negative prediction errors and negative percentage prediction errors suggest that the allowance

for bad debts is understated. Prediction errors and percentage prediction errors are defined inSections 4.1 and 4 .2 of the text, respectively .

Year 01 Year 1

Panel A : Prediction Errors

Mean -0.0185 -0.0140Median -0.0134 -0.0135Standard deviation 0.0761 0 .0954Minimum -0.5398 -0.4962Maximum 0.5817 0 .8534Number positive 186 246Number negative 381 456Parametric p-value` 0.0001 0 .0001Non-parametric p-value`' 0.0001 0 .0001

Panel B : Percentage Prediction Errors'

Mean -0.4076 -0.3588Median -0.7016 -0.6155Standard deviation 0.6556 0 .6857Minimum -1.0000 -1.0000Maximum 0.9481 0.9837Number positive 186 246Number negative 381 456Parametric p-value' 0.0001 0.0001Non-parametric p-value" 0.0001 0.0001

110

SCOTT B. JACKSON, WILLIAM E . WILCOX AND JOEL M . STRONG

Table 7. Regression Results for Alternative Model 2 Specificationsa ,b

Regression Coefficients`

ALLJA.1 = a0(1/TA i1) + a, (AR,/TA,,) + a2(AAR,/TA1,) + a 5 (SALE,,/TA. 1) + a4(ASALE,,/TA,,)

ao

(X I

a2

a3

a4

a5 Adj . R2 F-Stat .

Panel A : Regression Results for Year 0 (n = 915)'

ALLA/TA j, = ao(I/TA,,) + a,(AR,/TA,,) + a2IPOD., + sit

Coefficient

0.0075 0.0054 -0 .0070 0 .35 63 .59t-statistic

1.26

4.44

-2.66

0.35 56.83

ALL.,/TA.,= a o(1/TAj ,) + a i (AR .,/TA,,) + a 2(DAR,/TA.,) + a3IPOD .i + s,,

Coefficient 0.0075 0.0570 -0.0090-0.0063t-statistic 1 .59 4 .59 -1 .30 -2.42

ALL 1I/TA . i = a,,(1/TA.,) + ai(SALE,,/TA, t) + a 2IPOD,E + e, t

0 .26 42 .21Coefficient 0.0082 -0.0001 -0.0047t-statistic 5 .77 -0.98 -2.74

ALL, t/TA. 1 = a o (1/TA.i) + a I (SALE it/TA.,) + a2(ASALEi /TAi,) + a3IPOD., + sit

0 .29 42 .16Coefficient 0.0074 0.0001 0.0075 -0.0089t-statistic 5 .27 -0.81 5 .54 -4.78

ALL/rA,, = (Y0(1/TA.,) + a,(AR,t/TA,,) + a 2(ASALE, t/TA,,) + a3IPOD., + sit

0 .37 60 .99Coefficient 0.0068 0.0527 0.0065 -0 .0105t-statistic 1 .30 4.72 1 .71 -2.50

+ a 5IPOD, 1 + s,,

Coefficient 0.0069 0.0520 0.0020 -0 .0001 0.0067 -0.0107 0 .37 49 .87t-statistic 5 .25

10.01

0.31

-0.64 4.88 -5.58

Panel B: Regression Results for Year I (n = 1,656)'

ALL.,/TA,, = a0(1/TA ji ) + a,(AR1/TA.1) + a2IPOD ., + s it

0.34 108 .29Coefficient

0.0085 0.0583 -0 .0042t-statistic

2.80

5.12

-3.28

0.35 101 .62

ALLi,/TA,1 = ao(1/TA i ,) + a, (AR,/TA,t) + a2(AARj/TA j1) + a 3IPOD. i + s it

Coefficient

0.0074 0.0657 -0.0312-0.0022t-statistic

2.32 5.61 -2.31 -1.50

ALL i,/TA .i = a o,(1/TA . 1) + a,(SALE/rA) + a 2IPOD. i + s it

0 .23 63 .21Coefficient

0.0105 0.0010 -0.0025t-statistic

6.81 2.82 -2.01

Do Initial Public Offering Finns Understate the Allowance for Bad Debts?

M

Table 7. Continued

Regression Coefficients`

ALL)FA it = aJ1/TAi ,) + a i (AR,/TA,,) + a,(AAR,/TA,,) + 0 3 (SALE,,/TA,,) + a 4(ASALE .,/TA)+ a 3 IPOD,, + e~,

Coefficient

0.0066 0.0693 -0.0339 -0 .0010-0.0001 -0 .0022 0 .38

93 .27t-statistic

2.25

5.23

-2.48

-1.09

-30.11 -1 .54

° Regressions are estimated using IPO firms and their industry matches . Industry matches are chosenby identifying two existing public companies in the same four-digit SIC code as the IPO firm thathave the closest ratio of trade accounts receivables to total assets (referred to as the AR ratio) . Thefirst public firm in these sets has an AR ratio just above the IPO firm's AR ratio and the secondpublic firm has an AR ratio just below the IPO firm's ratio . See Section 5 .4 of the text .I Regression coefficients are estimated using ordinary least squares . When the null hypothesis ofhomoskedasticity is rejected (p<0 .10), t-statistics are computed using the heteroskedasticity-consistent covariance matrix (White, 1980) . Variables are defined as follows : ALL is the allowancefor bad debts ; TA is total assets ; AR is gross trade accounts receivable ; IPOD is a dummyvariable coded as 1 for IPO firms and 0 otherwise; SALE is net sales ; DAR is the year-to-yearchange in AR ; OSALE is the year-to-year change in SALE ; i is a firm index ; t is a time index .Regressions also include time fixed-effects (not reported) . The coefficients on the year dummy vari-

ables are jointly significant in the third and fourth regressions of both panels A and B (p < 0 .01) .Year 0 is the last annual accounting period before an IPO occurs and year I is the first annual

accounting period after an IPO is consummated .

(DALLI and DALL2). The firm and offering characteristics considered areretained ownership, age, risk, and offering size . Retained ownership is thepercentage of equity retained after the offering by all previous shareholders .Age is the number of months from the date of incorporation to the offeringdate. Risk is the number of risk factors listed in the prospectus. Offering sizeis the ratio of total IPO proceeds to the firm's book value .

Retained ownership is expected to have a positive relationship with DALLIand DALL2 because managers that sell larger fractions of their ownershipconcurrent with the IPO may be more likely to make accounting choices that

a0

a i

a2

a

a4

(X5 Adj . R2 F-Stat.

Panel B: Continued

ALLi,/TA,1 = a 11 (1/TA) + a i (SALE/TA,) + a,(OSALE, t/TA-,) + a3IPOD i, + e, t

Coefficient

0.0097 0.0011 -0.0001 -0.0027 0 .26 64.00t-statistic

6.35

3.13

-7.39 -2.20

ALL/17A t = a 0 (1/TAi ,) + a i (AR,/TA 44) + a,(ASALE,,/TA,,) + a 3IPOD, + e i ,

0 .36 106 .39Coefficient 0.0078 0.0582 -0.0001 -0.0044t-statistic 2 .75 5 .10 -30.78 -3 .49

1 1 2


bolster earnings at the time of IPO . Age is expected to have a positive relationwith DALL1 and DALL2 because younger firms have short operatinghistories, making it more difficult for auditors to determine an appropriatebalance for the allowance account . Risk is expected to have a negative relationwith DALL1 and DALL2 because higher risk firms may be concerned withportraying strong financial performance to counterbalance investors' perceptionsof risk . Offering size is expected to have a negative relation with DALL1 andDALL2 because firms making relatively large offerings may be more relianton the offering proceeds and may feel greater pressure to make income-increasing accounting choices .

To examine whether firm and offering characteristics of IPO firms areassociated with the discretionary component of the allowance for bad debts, weregressed retained ownership, age, risk, and offering size on DALL1 andDALL2 in years 0 and 1 . The only variable that was significantly associatedwith DALL1 and DALL2 in any of the regressions was risk (p = 0 .05) . Thisfinding suggests that the allowance for bad debts of risky IPO firms may beunderstated to a greater extent than IPO firms in general .

6. CONCLUDING REMARKS AND LIMITATIONS

This study adds to a growing body of empirical accounting research which indi-cates that managers of firms respond to external stimuli (i .e . bonus plans, debtcovenant violations, political scrutiny) by exercising discretion over reportedaccounting numbers . The results of this study are consistent with the view thatmanagers respond to the incentives arising in connection with IPOs by makingaccounting choices that bolster earnings and assets . This study also documentsthat managers' accounting response is not only statistically significant, but thatit is economically significant in relation to the account examined . Interestingly,despite the incremental monitoring role commonly ascribed to high quality audi-tors, the evidence does not indicate that Big Five auditors mitigate the generaltendency of IPO firms to understate the allowance for bad debts .

Prior studies on earnings management by IPO firms have generally concludedthat firm managers exercise discretion over accounting information reported tothe investing public . These studies, however, must be interpreted with somecaution because their methodologies are subject to some important limitations .The methodology used in this study, however, differs from those used in priorstudies in that it focuses on a single accrual account of IPO firms rather thantotal accruals or accounting method choices . Importantly, the accrual accountexamined in this study was chosen because it is inherently subjective and isarguably representative of managers' accounting responses to stock offerings


1 13

The results of this study are reassuring in that they are consistent with priorresearch, yet they are based on a methodology that is significantly differentfrom that used in prior research .

One benefit of focusing on one accrual account rather than total accruals isthat researchers can specifically identify accounts over which managersexercise accounting discretion . As a result, this study should be of interest topracticing auditors since it provides evidence that IPO firms tend to understatethe allowance for bad debts. Such knowledge could influence how auditors allo-cate their audit effort when examining the financial statements of IPO firms ."

This study should also be of interest to standard setters because it could helpthem assess the ". . . pervasiveness of earnings management and the overallintegrity of financial reporting" (Healy & Wahlen, 1999) . Further, the resultsof this study bring indirect evidence to bear on the merits of discretion versusuniformity . Should the accounting profession continue to afford managerssubstantial discretion over determining the appropriate balance in the allowancefor bad debts or should it establish more stringent guidelines that prescribe tosome extent how the allowance should be determined?

Finally, the results of this study should be interpreted with the followinglimitations in mind. First, managers of IPO firms may initiate stock offeringsduring periods in which their firms are performing particularly well, suggestingthat the allowance account of IPO firms is justifiably below that of their industrypeers. On the other hand, Boyajian (1994) suggests that the allowance fordoubtful accounts of IPO firms should be somewhat comparable to that of theirindustry peers . Further, anecdotal evidence in the financial press suggests thatIPO firms may relax their credit policies prior to stock offerings in an effortto bolster sales (Hall & Renner, 1988 ; Khalaf, 1992) . If the latter contentionis correct, the allowance of IPO firms should actually exceed that of theirindustry peers, which biases our tests against finding that IPO firms understatethe allowance . Nonetheless, the results of this study should be interpreted withsome caution because we cannot disentangle the "firm performance" explana-tion for our findings from the "earnings management" explanation .

Second, since the allowance for bad debts is only one accrual account overwhich managers have discretion, earnings management by IPO firms might existbut be targeted at other accrual accounts. DeAngelo (1988) points out that onecould observe no unusual patterns in the discretionary component of theprovision for bad debts when in reality earnings manipulation has taken placevia total accruals . Conversely, one could observe unusual patterns in the discre-tionary component of the provision for bad debts and erroneously conclude thatearnings manipulation has taken place because the discretionary component ofother accrual accounts were not incorporated into the research design ."

114


However, when the results of this study are considered along with the resultsof related studies on earnings management by IPO firms, this concern does notseem to be particularly compelling .

NOTES

1 . Throughout this paper, the term "allowance for bad debts" refers to the contra-accounts receivable account in the balance sheet . The term "provision for bad debts"refers to the expense account included in the income statement . Accounting textbooksuse a variety of synonymous names for the allowance for bad debts and provision forbad debts such as "allowance for uncollectible accounts" and "uncollectible accountsexpense," respectively .

2 . It should be noted that Teoh et al . (1998a) directed most of their attention towardsexamining total accruals and secondarily examined the allowance for bad debts . It wasnot their intention to perform an in-depth analysis like the one performed in this study .

3. Economic significance could also be measured relative to earnings . However,many IPO firms report negative earnings or earnings that are around zero making itdifficult to interpret economic significance when the understatement is evaluatedrelative to earnings .

4. Section 1 discusses the analyses performed by Teoh et al . (1998a) and describeshow this study extends and is distinguished from Teoh et al . (1998a) .

5. The results and discussion contained in Teoh et al. (1998a and 1998b) stronglysupport this assumption . In particular, see Appendix B of Teoh et al . (1998b) for adetailed description of how companies can manage earnings .

6. The assumption that outsiders do not adjust accounting numbers for bias ormanipulation does not seem unreasonable . For example, the results of Dechow et al .(1996) indicate that investors may not see through even the most aggressive forms ofearnings management . In their study of firms subject to SEC scrutiny, the market initiallyvalued the earnings of sample firms normally and only recognized aggressive reportingwhen the SEC pointed out the overstatement of earnings .

7. We do not suggest that Big Five accounting firms counteract all bias in theallowance for bad debts . Rather, we suggest that they mitigate a general tendency ofIPO firms to understate this accrual account .

8. Variables related to Eq. (1) are discussed here in their unscaled form forconvenience. As discussed later, these variables are scaled by either accounts receivable(expectation model 1) or total assets (expectation model 2) .

9. Firms that went public within five years preceding the formation of industryestimation samples were excluded to ensure that they had no incentives to understatethe allowance for bad debts in connection with IPOs .

10. Both models developed in this section are cross-sectional . This is because IPOfirms do not report a sufficient number of yearly observations to estimate time-seriesmodels . Most IPO firms report two years of balance sheet information and three yearsof income statement information .

11 . Admittedly, the requirement that at least six firms be included in every two-digitSIC code is somewhat arbitrary . The rationale for imposing this minimum requirementis that an outlying firm in a particular industry could have an undue influence on NALL1if there is a small number of firms included in that industry .


115

12. To evaluate the predictive ability of this model, we performed the followinganalyses . First, we selected the five industries (28XX, 35XX, 36XX, 38XX, 73XX)with the greatest representation in our IPO sample (IPO firms are not involved in theanalyses). Second, we partitioned firms in each industry into an estimation sample anda holdout sample . The holdout sample consisted of every fifth firm in an industry whilethe estimation sample consisted of the remaining firms . Third, we estimated Eq . (4) onan industry-by-industry basis using firms in the estimation samples . Fourth, wecomputed and analyzed the prediction errors, as defined in Eq . (6), of the holdoutsample . Our central concern is whether Eq . (4) yields unbiased prediction errors forthe holdout sample . If Eq. (4) is unbiased, prediction errors for a sample of existingpublic companies will have mean/median values that do not differ from zero . For eachof the five industries, parametric and non-parametric tests indicate that the predictionerrors have mean and median values that do not differ significantly from zero(all p-values exceed 0 .18) . This analysis provides evidence that the results reported inthis study are not induced by model misspecification . In addition to evaluating whetherthe prediction errors are unbiased, we also evaluated the absolute percentageprediction errors, as defined in Section 4 .2. The mean and median absolutepercentage prediction errors are approximately 40%, which is consistent with somediversity across firms in their credit and collection policies . Given evidence that theprediction errors are unbiased, the size of the absolute prediction errors means that wemay not detect manipulation when it actually exists . In order to increase confidence inthe results, we performed a variety of supplemental analyses as described in Section5 .4 which rely upon complementary methodology . All analyses yielded similarconclusions .

13. Two potential econometric problems arise in these cross-sectional regressions . Theproblems are biased standard error estimates resulting from heteroskedasticity andautoregressive error terms associated with using observations that are clustered on timeand industry. We use the parameter estimates for predictive purposes rather than testingfor statistical significance so these problems are not a major concern . As noted in Kmenta(1986), parameter estimates are unbiased in the presence of both of these econometricproblems.

14. The rationale for the requirement that at least 15 firms be included in every two-digit SIC code is that using fewer than 15 observations could result in erratic regressioncoefficients . In addition, this requirement is consistent with closely related studies onearnings management .

15. There are two untested assumptions in the expectation models . First, they assumethat the allowance for bad debts is influenced by industry factors . Second, they assumethat the relationship between accounts receivable and the allowance for bad debts variesintertemporally . To test these assumptions, we performed the following analysis . Apooled cross-sectional and time-series regression with the allowance for bad debts (scaledby contemporaneous total assets) as the dependent variable and dummy variables forindustry (two-digit SIC code) and time (year) as the independent variables wasestimated. Both the industry (F-statistic = 5 .79) and time (F-statistic = 23 .61) dummyvariables were significant at the 0.0001 level, thus supporting our assumptions .

16. In cases where ALL/AR (ALL/TA) is zero, PER! (PER2) is set equal to 1 toavoid division by zero . Also, when PER! or PER2 is greater than 1, they are set equalto 1 to avoid allowing extreme values (caused by division by a relatively small number)to have an undue influence on the results .

116


17. The use of percentage prediction errors in statistical tests also overcomes a problemassociated with using the prediction error metrics defined in Eqs (3) and (6) . This problemis illustrated by the following example . Assume that firm A has a predicted scaledallowance of 0 .095 and an actual scaled allowance of 0.085, and firm B has a predictedscaled allowance of 0 .035 and an actual scaled allowance of 0.025. Based on theprediction error metrics defined in Eqs (3) and (6), the understatement of the allowancefor bad debts of firm A is the same as that of firm B (-0 .01 in both cases), despiteobvious differences in the relative magnitude of the prediction errors . However, usingEqs (9) and (10), the percentage prediction error for firm A is - 12% while the percentageprediction error for firm B is -40% . Because the error metrics defined in Eqs (3) and(6) and those defined in Eqs (9) and (10) are complementary, we use both error metricsto test the hypotheses .

18. There have been no changes in GAAP related to accounting for bad debts forindustrial companies over the past 20 years . As a result, we believe that the resultsreported in this study are also representative of more recent time periods .

19 . Note that there are 1,269 firm years in our IPO sample related to model 1 (Eq .(2)) . Thus, there are 1,269 industry level observations used to determine NALLI inEq. (2) for the 1,269 IPO firms.

20. Percentage prediction errors (see Eqs (9) and (10)) are computed by dividingprediction errors by the observed balance in the allowance account. An alternative wayto define percentage prediction errors is to divide prediction errors by the expectedbalance in the allowance account (NALL1 or NALL2) . Because the results suggest thatIPO firms understate the allowance, dividing prediction errors by the expected balancein the allowance account rather than the actual balance will result in smaller percentageprediction errors . The following schedule shows what the percentage prediction errorswould have been if the alternative definition were used. Percentage predictionerrors computed using the alternative definition are all significant at p < 0 .0001 .

Model 1 (DALL1)

Model 2 (DALL2)

21. In cases where four-digit matches could not be found, firms were matched atsuccessively broader industry definitions . In excess of 75% of sample firms were matchedat the four-digit level . For approximately 20% of the sample firms we could not iden-tify one match above and one match below the AR ratio for IPO firms . In suchcircumstances, both matches were either above or below the IPO firms' AR ratio .

22. It should be pointed out that the way we measure economic significance may notcorrespond to the way auditors measure audit significance . We measure economic signif-icance by reference to the recorded balance in the allowance account, while auditorsprobably measure audit significance by reference to net income or total assets . Thus, itis possible that auditors are aware that IPO firms tend to understate the allowance forbad debts, but do not require them to adjust the allowance because the understatementis not material in relation to total assets or net income .

Year 0 Year 1 Year 0 Year 1

Mean (as reported in Table 4) -0.5197 -0.4812 -0.4330 -0.3561Mean (using alternative definition) -0.3385 -0.2805 -0.2536 -0.1721Median (as reported in Table 4) -0.8800 -0.8216 -0.7699 -0.6078Median (using alternative definition) -0.5196 -0.4796 -0.4652 -0.3780


117

23. DeAngelo's (1988) comments concern the provision for bad debts (an incomestatement account), while the focus of this study is the allowance for bad debts (a balancesheet account) . Her comments apply with equal force to this study since the provisionfor bad debts has a direct impact on the allowance for bad debts . Note the followingrelationship: Allowance for bad debts, = Allowance for bad debts t , + Provision for baddebts, - Write-offs of accounts receivable, .

ACKNOWLEDGMENTS

We thank Jay Ritter for giving us access to his IPO database and John Friedlan,Chris James, William Megginson, and Kathleen Weiss Hanley for giving usaccess to some of the prospectuses used in this study . We also thank two anony-mous reviewers, John Barrick, Cheryl Fulkerson, Jim Groff, Elaine Mauldin,Siva Nathan, Marshall Pitman, Jeff Quinn, Robin Radtke, and workshop partic-ipants at the University of Nebraska-Lincoln and the 1999 AAA SoutheastRegion Meeting for providing helpful comments .

REFERENCES

Aharony, J., Lin, C ., & Loeb, M. P. (1993) . Initial public offerings, accounting choices, andearnings management . Contemporary Accounting Research, 10, 61-81 .

American Institute of Certified Public Accountants (AICPA) (1988) . Auditing Accounting Estimates .Statement on Auditing Standards No . 57 . New York : AICPA .

Arens, A . A ., & Loebbecke, J. K . (1996). Auditing : An Integrated Approach (7th ed .) . EnglewoodCliffs, NJ: Prentice-Hall, Inc .

Bartlett, J . W. (1988) . Venture Capital: Law, Business, Strategies, and Investment Planning . NewYork, NY : John Wiley and Sons, Inc .

Beaver, W . H ., & Engel, E. E . (1996) . Discretionary behavior with respect to allowances for loanlosses and the behavior of security prices . Journal ofAccounting and Economics, 22, 177-206 .

Bloch, E . (1989). Inside Investment Banking. Homewood, IL: Dow Jones-Irwin .Blowers, S . C., Ericksen, G . K ., & Milan, T. L . (1995) . The Ernst and Young Guide to Taking Your

Company Public . New York, NY : John Wiley and Sons, Inc .Boyajian, V . H . (1994) . Early planning for your initial public offering . Securities Regulation Law

Journal, 22, 67-77 .Browning, E. S . (1998) . IPOs often come dressed up with best figures, studies say . Wall Street Journal,

(March 10), Cl-C2 .DeAngelo, L . E. (1981) . Auditor size and audit quality . Journal of Accounting and Economics, 3,

183-199 .DeAngelo, L . E . (1988) . Discussion of evidence of earnings management from the provision for

bad debts. Journal of Accounting Research, 26, 32-40 .Dechow, P. M., Sloan, R . G ., & Sweeney, A . P . (1995) . Detecting earnings management . The

Accounting Review, 70, 193-225 .Dechow, P . M ., Sloan, R . G ., & Sweeney, A . P. (1996) . Causes and consequences of earnings

manipulation : An analysis of firms subject to enforcement actions by the SEC . ContemporaryAccounting Research, 13, 1-36.

118

SCOTT B. JACKSON, WILLIAM E. WILCOX AND JOEL M . STRONG

DeFond, M. L., & Jiambalvo, J . (1994) . Debt covenant violation and manipulation of accruals .Journal of Accounting and Economics, 17, 145-176 .

Financial Accounting Standards Board (FASB) (1975) . Accounting for Contingencies. Statement ofFinancial Accounting Standards No. 5 . Stamford, CT : FASB .

Francis, J. R., & Wilson, E . R. (1988) . Auditor changes : A joint test of theories relating to agencycosts and auditor differentiation . The Accounting Review, 63, 663-682 .

Friedlan, J. M. (1994) . Accounting choices of issuers of initial public offerings . ContemporaryAccounting Research, 11, 1-31 .

Guidry, F., Leone, A., & Rock, S . (1999) . Earnings-based bonus plans and earnings managementby business unit managers . Journal of Accounting and Economics, 26, 113-142 .

Hall, W., & Renner, A . (1988) . Lessons that auditors ignore at their own risk . Journal ofAccountancy, 166, 50-59 .

Hanley, K. W., Kumar, A . A., & Seguin, P. J . (1993) . Price stabilization in the market for newissues . Journal of Financial Economics, 34, 177-197 .

Healy, P . M., & Wahlen, J . M . (1999) . A review of the earnings management literature and itsimplications for standard setting . Accounting Horizons, 13, 365-383 .

Jones, J. J . (1991) . Earnings management during import relief investigation . Journal of AccountingResearch, 29, 193-228 .

Khalaf, R . (1992). Buyer, do thy homework . Forbes, (April 13), 47-48 .Kmenta, J. (1986) . Elements of Econometrics (2nd ed .). New York, NY : Macmillan .Kreutzfeldt, R . W., & Wallace, W . A . (1986) . Error characteristics in audit populations : Their

profile and relationship to environmental factors . Auditing : A Journal of Practice and Theory,6, 20-43 .

McNichols, M ., & Wilson, G . P. (1988) . Evidence of earnings management from the provision forbad debts. Journal of Accounting Research, 26, 1-31 .

Pany, K ., & Whittington, R . (1997) . Auditing (2nd ed.) . New York, NY: McGraw Hill, Inc .Perez, R . C . (1984) . Inside Investment Banking . New York, NY: Praeger Publishers .Richardson, M . R . (1976) . Going Public . London: Business Books .Robertson, J . C . (1996) . Auditing (8th ed.). New York, NY: McGraw Hill, Inc.Schipper, K. (1989) . Commentary on earnings management . Accounting Horizons, 3, 91-102 .Schroeder, M . (1994) . The Sherlock Holmes of accounting . Business Week, (September 5), 48-52 .Teoh, S . H ., Wong, T. J ., & Rao, G. R. (1998a). Are accruals during initial public offerings

opportunistic? Review of Accounting Studies, 3, 175-208 .Teoh, S . H ., Welch, I ., & Wong, T . J . (1998b) . Earnings management and the long-run market

performance of initial public offerings . The Journal of Finance, 53, 1935-1974 .Tversky, A ., & Kahneman, D . (1974). Judgment under uncertainty : Heuristics and biases . Science,

185, 1124-1131 .Weiss, M. (1988) . Going Public: How to Make Your Initial Stock Offering Successful . Blue Ridge

Summit, PA: Liberty House .White, H. (1980) . A heteroskedasticity-consistent covariance estimator and a direct test for

heteroskedasticity. Econometrica, 48, 817-838 .

Do initial public offering firms understate the allowance for bad debts?

Documents