STABLE BOOK-TAX DIFFERENCES, PRIOR EARNINGS, AND EARNINGS PERSISTENCE. Joshua C. Racca. Dissertation Prepared for Degree of DOCTOR OF PHILOSOPHY

Transcription

1 STABLE BOOK-TAX DIFFERENCES, PRIOR EARNINGS, AND EARNINGS PERSISTENCE Joshua C. Racca Disseraion Prepared for Degree of DOCTOR OF PHILOSOPHY UNIVERSITY OF NORTH TEXAS Augus 0 APPROVED: Teresa Conover, Major Professor Mazhar Siddiqi, Minor Professor Carol Ann Fros, Commiee Member Paul Huchison, Commiee Member Govind Iyer, Commiee Member Don Finn, Professor and Chair of he Deparmen of Accouning Finley Graves, Dean of he College of Business James D. Meernik, Acing Dean of he Toulouse Graduae School

2 Racca, Joshua C. Sable Book-Tax erences, Prior Earnings and Earnings Persisence. Docor of Philosophy (Accouning), Augus 0, 69 pp., ables, 7 illusraions, references, 7 iles. This sudy resolves divergen prior findings relaing book-ax differences o fuure earnings, deermines wheher prior lieraure has missed relaionships beween differen ypes of book-ax differences and pre-ax and/or afer-ax income, and invesigaes prior earnings as a facor conribuing o he observed relaionships. As pas research has found ha some firms have large book-ax differences over several years, his sudy separaes hese firms wih large sable book-ax differences from ohers wih large book-ax differences (non-sable) when invesigaing he link beween large book-ax differences and fuure earnings. Finally, his sudy invesigaes wheher he relaionship beween book-ax differences and fuure earnings reflecs informaion abou prior earnings and finds ha prior earnings growh explains much of he lower persisence found for firms wih large book-ax differences.

3 Copyrigh 0 By Joshua C. Racca ii

4 TABLE OF CONTENTS INTRODUCTION... PRIOR RESEARCH AND HYPOTHESIS DEVELOPMENT... 6 SAMPLE AND METHODOLOGY... RESULTS... 9 CONCLUSION... APPENDIX A TIME SERIES PROPERTIES OF BOOK-TAX DIFFERENCES... 6 APPENDIX B ADDITIONAL ANALYSIS... 9 REFERENCES iii

5 INTRODUCTION Prior research has found ha firms wih large differences beween repored ne income and axable income (book-ax differences) exhibi lower earnings persisence, lower earnings growh, higher earnings managemen, fraudulenly over-saed earnings, and lower earnings qualiy, when compared o firms wih average levels of book-ax differences. Oher sudies have found ha some firms mainain large book-ax differences over muliple years and aribued his o economic fundamenals, firm characerisics, or ax aggressiveness, raher han low earnings qualiy. This sudy combines hese sreams of research and invesigaes he relaionship beween large book-ax differences and earnings persisence, conrolling for firms wih large differences over muliple years, and finds ha differen conclusions are reached when firms wih large sable book-ax differences are separaed from firms wih large differences which are no sable in models of book-ax differences and earnings persisence. This sudy also looks a prior earnings paerns and deermines ha prior earnings of firms wih large book-ax differences explains some of he lower earnings persisence exhibied by firms wih large book-ax differences. There is lile heory addressing wha drives he observed relaionship beween large book-ax differences and lower earnings persisence. Evidence of boh earnings managemen and ax aggressiveness has been found; however, oher economic facors have no been horoughly examined. This sudy improves he conclusions ha can be made abou he relaionship beween large book-ax differences. By conrasing he relaionship beween book-ax differences and earnings persisence for firms wih large sable book-ax differences and oher firms wih large bu no sable book-ax differences i becomes easier o idenify he facors driving he

6 relaionship beween large book-ax differences and low earnings persisence. For he purposes of his sudy a firm is considered o have a large book-ax difference if he difference scaled by asses is wihin he op weny percen for he indusry in a given year. A firm is considered o have a large sable difference if he firm has a large difference of a given ype (emporary or permanen) for hree consecuive years. Figure shows example paerns of book-ax differences, wih a firm having large sable book-ax differences shown a he op, and oher firms wih large book-ax differences in year, bu no in all prior years below. Alernae ime periods are examined and he resuls are no sensiive o differen ime period specificaions for large sable differences.

7 FIGURE Examples of Book Tax erences Over Time Large Sable Book Tax erences Book Tax erence 4 Large Book-Tax erences in Year T, bu no Large in all Prior Included Time Periods Book Tax erence 4 Year Book Tax erence 4 Year Book Tax erence 4 Year Book Tax erence 4 Year A finding ha firms wih large sable differences are sysemaically differen from oher firms wih large book-ax differences would sugges ha models linking book-ax differences and fuure earnings should incorporae his disincion. I find ha firms wih large sable book-ax differences are differen han oher firms wih large book-ax differences, and ha his conras is imporan for undersanding he relaionship beween book-ax differences and fuure earnings. This sudy examines wheher book-ax differences are linked o fuure earnings persisence, because book-ax differences reflec

8 variabiliy in prior earnings or differences in prior earnings growh and finds differences in boh prior earnings variabiliy and growh, bu finds ha only prior earnings growh is useful for explaining he lower earnings persisence for firms wih large book-ax differences. This sudy answers he call in recen reviews of ax research for sudies ino why book-ax differences convey informaion abou fuure earnings and wheher permanen and emporary book-ax differences each lead o differen conclusions abou fuure earnings (Hanlon and Heizman 009; Graham e al. 00). For he purposes of his sudy, boh emporary and permanen book-ax differences refer o aggregae differences esimaed from financial saemens. Companies wih emporary differences ofen experience large and/or increasing differences over ime. While individual emporary differences evenually reverse, new emporary differences are creaed and he overall level may increase, decrease, or remain consan. Permanen differences do no reverse as wih emporary differences, bu companies wih permanen differences can have new permanen differences each year, hus mainaining large permanen differences over many years. Large aggregae book-ax differences have led o calls for book-ax conformiy, osensibly o minimize he incenives for earnings managemen and/or increase ax receips. This sudy deermines ha large book-ax differences resul boh behaviors such as earnings managemen and ax aggressiveness, bu can also be explained by prior earnings as suggesed by he finding ha book-ax differences are srongly linked o economic cycles (McClelland and Mills 007). As changes in prior earnings (due o economic cycles or oher reasons) explain some of he relaionship beween book-ax differences and earnings persisence, hen concerns ha large book-ax differences reflec aggressive earnings or ax managemen (or even ax fraud) 4

9 may be misplaced. This finding also suggess ha fuure research aemping o use large bookax differences as a proxy for earnings qualiy or managemen should consider boh wheher he large differences are sable and he prior earnings growh of hose firms. Low earnings qualiy (and poenial earnings managemen) is mos likely o be found where book-ax differences are large bu no sable and where low persisence canno be explained by higher prior earnings growh. This may lead fuure researchers o beer use book-ax differences as a ool o sudy earnings managemen by isolaing hose more suspec firms. This research conribues o he lieraure by linking wo sreams of research, namely, sudies ha find relaionships beween book-ax differences and fuure earnings, and sudies ha show ha firms wih large book-ax differences over long ime periods have differen characerisics han oher firms wih large book-ax differences. I also uses he findings ha firms wih large sable book-ax differences differ from oher firms wih large book-ax differences o deermine ha in some cases he relaionship beween book-ax and earnings persisence reflec pass earnings growh. The remainder of his paper is organized as follows. Secion II describes prior lieraure and develops he hypoheses. Secion III describes he proposed daa sample and research mehodology. Secion IV presens he resuls and Secion V discusses he significance of he findings and concludes. The governmen sill has an incenive o increase revenue by requiring book and ax earnings o conform. This issue is unlikely o be seled by any evidence abou causes and effecs of book-ax differences. Oher firms wih large book-ax differences in year may include firms wih many differen possible levels of book-ax differences. Firms wih relaively small differences ha suddenly increase, firms wih large differences in prior years, bu a drop in one or more years reurning o a large difference in year, or a myriad of oher combinaions. These differences may be hough of as unsable, bu he degree of insabiliy would vary grealy. For his reason, he erm non-sable refers o firms wih large book-ax differences which do no mee he crieria o be considered sable (op 0% wihin indusry for hree consecuive years). 5

10 PRIOR RESEARCH AND HYPOTHESIS DEVELOPMENT Boh emporary and permanen book-ax differences are linked o lower fuure earnings. Hanlon (005) finds ha firms wih large emporary book ax differences have lower pre-ax earnings persisence han firms wih smaller emporary book-ax differences. 4 Lev and Nissim (004) find ha firms wih large oal (emporary and permanen) book-ax differences have lower fuure afer-ax earnings growh han firms wih smaller differences. Hanlon s and Lev and Nissim s sudies are moivaed wih poenial earnings qualiy issues, bu use differen measures of boh book-ax differences and differen measures of fuure earnings. Graham, Ready, and Shackelford (00) call for furher sudy of hese findings by using separae measures of emporary and permanen book-ax differences. This proposed sudy does so and separaes firms wih large sable book-ax differences when examining he link beween book-ax differences and fuure earnings and poenial explanaions for his relaionship. 5 Jackson (009) aemps o reconcile he differences in he measure of book-ax differences and earnings, and finds ha he relaionship beween permanen book-ax differences and afer-ax earnings is relaed o changes in fuure ax expense, while emporary book-ax differences are relaed o pre-ax earnings. This sudy separaes firms wih large sable book-ax differences and re-examines wheher a differen conclusion can be reached abou he relaionship beween each ype of book-ax difference and boh pre-ax and afer-ax earnings persisence. Oher research has suggesed ha firms manage earnings using ax expense (Schmid 006; Dhaliwal e al. 004; Holland and Jackson 004). Firms manage ax expense downward o 4 Hanlon finds ha firms wih large negaive book-ax differences and large posiive book-ax differences boh have lower pre-ax earnings persisence han firms wih average levels of book-ax differences. In his sudy large bookax differences refer o insances where repored earnings are higher han esimaed axable income. 5 Firms wih large sable book-ax differences are hose firms which have large differences for he curren year and each of he prior wo years. 6

11 mee afer-ax earnings arges. If firms manage earnings downward using permanen differences, his suggess a leas a negaive correlaion beween curren pre-ax earnings and permanen book-ax differences. This sudy separaes firms wih large sable book-ax differences from oher firms wih large book-ax differences which provides a beer undersanding of how permanen differences are relaed o pre-ax earnings persisence; as well as afer-ax earnings hrough ax expense as suggesed by Jackson (009). Oher research suggess ha some firms wih large book-ax differences or low effecive ax raes mainain hese large differences over muliple years. 6 Dyreng, Hanlon, and Maydew (008) find ha some firms mainain low cash effecive ax raes for as long as en years. 7 Oher sudies use he cash effecive ax rae measure used in Dyreng e al. (008) o idenify firms ha are ax aggressive. 8 Blaylock, Shevlin, and Wilson (009) use he cash effecive ax rae o idenify firms as ax aggressive, and separae ax aggressive firms from firms wih high earnings managemen incenives. They find ha ax aggressive firms do no have lower earnings persisence, while firms wih high earnings managemen incenives do have lower earnings persisence. 6 Low average effecive ax raes occur when firms have comparaively low axes relaive o book income, which corresponds o firms having large book-ax differences. 7 The cash effecive ax rae uses cash axes paid divided by preax book income less special iems. CashTaxesPaid CASHETR Pr eax Book Income SpecialIems In he case of muliple year (N) measures, he cash ETR is calculaed as he sum of cash axes paid over a number of years divided by he sum of preax book income less special iems over a number of years. CASTETR N N i N i CashTaxesPaid Pr eaxbook Income Special Iems i i 8 Firms wih low cash effecive ax raes are considered ax aggressive as heir acual cash oulays for axes are relaively low for long periods of ime. 7 i

12 Dhaliwal, Huber, Lee and Pincus (008) find ha while he cos of capial is relaed o variabiliy in book-ax differences, he size of he book-ax differences is he mos imporan facor. This research builds on he findings of Dhaliwal e al. by examining wheher variabiliy in book-ax differences is imporan when examining he relaionship beween large book-ax differences and earnings persisence. Separaing firms wih large sable book-ax differences from firms wih large non-sable book-ax differences provides more insigh ino he relaionship beween book-ax differences and earnings persisence. This sudy s firs se of hypoheses direcly addresses wheher firms wih large sable book-ax differences have higher earnings persisence han oher firms wih large book-ax differences. 9 If book-ax differences reflec prior earnings, firms wih large sable book-ax differences can be expeced o have higher earnings persisence han oher firms wih large bookax differences. Boh emporary and permanen differences are examined. Hypoheses one hrough four are saed in alernae form below, hypoheses one and wo examine emporary differences and hypoheses hree and four examine permanen differences H: Firms wih large posiive sable emporary book-ax differences exhibi higher pre-ax earnings persisence han oher firms wih large posiive emporary book-ax differences in year. H: Firms wih large posiive sable emporary book-ax differences exhibi higher afer-ax earnings persisence han oher firms wih large posiive emporary book-ax differences in year. 9 Earnings persisence is defined as he coefficien of curren earnings in a regression of year ahead earnings. The following is he regression equaion for pre-ax book income persisence. PTBI PTBI In his equaion, β represens pre-ax earnings persisence, and PTBI is pre-ax book income. 8

13 H: Firms wih large posiive sable permanen book-ax differences exhibi higher pre-ax earnings persisence han oher firms wih large posiive permanen book-ax differences in year. H4: Firms wih large posiive sable permanen book-ax differences exhibi higher afer-ax earnings persisence han oher firms wih large posiive permanen book-ax differences in year. There are many sudies ha provide evidence ha book-ax differences are linked o earnings managemen, earnings qualiy, and fraud. Book-ax differences have been found o be incremenally useful o accrual-based measures in idenifying firms wih fraudulenly over-saed earnings (Eredge e al. 008). Oher sudies have also shown ha firms manage earnings using ax expense o mee arges; avoid losses; and smooh earnings, and ha he marke seems o discoun such managemen (Dhaliwal e al. 004; Schmid 006; Frank and Rego 006; Tang 007). A few recen sudies have also aemped o direcly address he quesion of wheher he book-ax gap can be explained by earnings managemen or ax aggressiveness (Blaylock e al. 009; Seidman 00). McClelland and Mills (007) sugges addiional causes, finding ha bookax difference are relaed o macro-economic condiions, wih aggregae differences increasing during economic expansion and decrease during economic conracions. By applying his finding on a micro level o companies, Hypoheses 5 and 6 invesigae wheher companies wih large book-ax differences have differen levels of prior earnings growh han firms wihou large book-ax differences. Boh Lev and Nissim (004) and Jackson (004) find ha firms wih large differences have lower fuure earnings growh. This sudy invesigaes wheher his may be relaed o differences in prior earnings growh and provides evidence ha he lower fuure 9

14 earnings growh is relaed o prior earnings and no necessarily only earnings managemen or ax aggressiveness. Hypoheses five and six are no direcional since eiher high or low prior earnings can indicae lower fuure earnings. Higher prior earnings growh may represen an unsusainable level of earnings and i may be ha fuure earnings will be lower, while lower prior growh may represen a rend ha coninues ino he fuure. 0 The relaionship may be differen for firms wih emporary han for firms wih permanen differences, which migh provide evidence abou he inferences which can be made abou firms wih each ype of difference. Hypoheses 5 and 6 are saed below. H5: Firms wih large emporary book-ax differences have eiher higher or lower prior earnings growh when compared o firms wihou large emporary book-ax differences. H6: Firms wih large posiive book-ax differences have eiher higher or lower prior earnings growh when compared o firms wihou large permanen book-ax differences. Hypoheses 7 and 8 predic differences in prior earnings variabiliy among firms wih large bookax differences. As wih earnings growh, variabiliy in prior earnings is a facor ha may conribue o lower earnings persisence for firms wih large book-ax differences. Eiher higher or lower variabiliy in prior earnings may be relaed o he observed differences in earnings persisence; hus, Hypoheses 7 and 8 are non-direcional. As he mechanism for emporary and permanen differences may be differen; each is esed separaely. Hypoheses 7 and 8 are saed below. 0 Lower prior earnings level may represen an abnormally low level ha will increase in he fuure, and higher prior earnings growh may also represen a rend ha will coninue ino he fuure; however, prior research has shown ha firms wih large book-ax differences have lower fuure earnings growh. 0

15 H7: Firms wih large emporary book-ax differences have eiher higher or lower prior earnings variabiliy when compared o firms wihou large emporary book-ax differences. H8: Firms wih large permanen book-ax differences have eiher higher or lower prior earnings variabiliy when compared o firms wihou large permanen book-ax differences. In addiion o he esing of hypoheses, addiional analysis was conduced o deermine wheher here is a link beween earnings persisence and pas earnings growh and/or variabiliy and wheher he relaionships beween book-ax differences and earnings can be explained by differences in prior earnings growh and/or variabiliy for firms wih large book-ax differences.

16 SAMPLE AND METHODOLOGY The daa o be used in his sudy are derived from financial saemens issued by publicly raded companies and obained from Compusa. To compare only firm-years using a consisen sandard in accouning for income axes, he sudy begins wih 994, when Saemen of Financial Accouning Sandards 09, Accouning for Income Taxes (SFAS 09) was implemened. The implemenaion of SFAS 09 has been shown o posiively affec invesors abiliy o inerpre informaion in book-ax differences (Lev and Nissim 004). To be included in he primary sample, each firm mus have daa available for curren ax expense, deferred axes, pre-ax book income, ne operaing loss carry-forwards, and ne income in each year. As in prior research, financial services firms and uiliies are excluded from he sample since hey have differen incenives influencing repored earnings han oher firms. Firms wih large sable book-ax differences are hose ha have large book-ax differences in he curren year (year ) and he preceding wo years. A sensiiviy analysis examines he impac of differen specificaions of muliple year book-ax differences. Only U.S.-based firms are included in he sudy, and firm-years wih any of he following: negaive ax expense, negaive ne income, or a ne operaing loss carry-forward in year is excluded. Addiionally, sensiiviy analysis examines he effec of excluding firms wih large amouns of non-us earnings, as hose firms have differen opporuniies o manage axable income and heir income is subjec o differen ax raes. To deermine which firm-years have large differences (boh emporary and permanen), firms are ranked for each year for each of hese differences and he op weny percen in each caegory (emporary and permanen) are considered o have large differences. Temporary differences are calculaed as deferred axes deflaed by average oal asses. Permanen

17 differences are esimaed as he difference beween esimaed ax expense a he sauory ax rae less deferred ax expense and are also deflaed by average oal asses. Quinile rankings are performed for each year wihin each indusry using Fama-French indusry definiions. The model used o es Hypohesis is: PTBI β 4 α β SableTemp LargeTemp β 5 PTBI β PTBI SableTemp β PTBI LargeTemp () Where for firm i and year : PTBI is pre-ax book income in year ; PTBI + is pre-ax book income in year +; LargeTemp is a binary variable ha equals one if firm i has a large emporary book-ax difference in year and zero oherwise; and SableTemp - is a binary variable ha equals one if firm i has a large emporary difference for all hree years saring wih - and coninuing hrough year (years -, -, and ). All variables are deflaed by oal asses. Table presens he inerpreaions and expeced values of coefficiens esimaed from equaion prediced by hypohesis : There is large variaion in levels of book-ax differences wihin all indusries and years. In every indusry and for every year levels of book-ax differences range from firms wih higher axable income han book income o firms wih high book-income and lile or no axable income. The LargeTempGroup includes he SableTemp group. This differs from he groups used in esing hypoheses hrough 6 where he groups do no overlap. Care should be aken when comparing regression resuls o -es resuls.

18 TABLE Inerpreaions and Predicions for Coefficiens for Equaion Coefficien Inerpreaion Expeced Value β Pre-ax earnings persisence of firms wihou large emporary differences (+) β Pre-ax earnings persisence of firms wih large emporary differences in year, relaive o firms wihou large emporary (-) differences β 5 Pre-ax earnings perisence of firms wih large emporary differences in years, -, and -, relaive o all firms wih large emporary differences in year (+) A posiive coefficien for β5 indicaes suppor for H Pre-ax earnings persisence higher for firms wih large sable differences relaive o oher firms wih large differences β β +β β +β +β 5 Pre-ax earnings persisence of firms wihou large emporary differences Pre-ax earnings persisence of firms wih large emporary differences, no-sable Pre-ax earnings persisence of firms wih large sable emporary differences The model used o es Hypohesis is equaion below: IBEI β 4 α β SableTemp LargeTemp β 5 IBEI β IBEI β SableTemp IBEI LargeTemp () Where pre-ax book income (PTBI) is replaced wih IBEI; IBEI is income before exraordinary iems. Table presens he inerpreaions and expeced values of coefficiens esimaed from equaion prediced by hypohesis : 4

19 TABLE Inerpreaions and Predicions for Coefficiens for Equaion Coefficien Inerpreaion Expeced Value β Afer-ax earnings persisence of firms wihou large emporary differences (+) β Afer-ax earnings persisence of firms wih large emporary differences in year, relaive o firms wihou large emporary (-) differences β 5 Afer-ax earnings perisence of firms wih large ermporary differences in years, -, and -, relaive o all firms wih large emporary differences in year (+) A posiive coefficien for β5 indicaes suppor for H Afer-ax earnings persisence higher for firms wih large sable differences relaive o oher firms wih large differences β β +β β +β +β 5 Afer-ax earnings persisence of firms wihou large emporary differences Afer-ax earnings persisence of firms wih large emporary differences, no-sable Afer-ax earnings persisence of firms wih large sable emporary differences The models used for esing H and H4 are similar o he models used for esing H and H, wih permanen differences replacing emporary differences in he model. The model for esing hypohesis is equaion below: PTBI β 4 α β SablePerm LargePerm β 5 PTBI β PTBI SablePerm β PTBI LargePerm () Where previously used variables are defined as above and: LargePerm is a binary variable ha equals one if firm i has a large emporary book-ax difference in year and zero oherwise; and SablePerm - is a binary variable ha equals one if firm i has a large permanen difference for all hree years saring wih - and coninuing unil ending in year (years -, -, and ). Table presens he inerpreaions and expeced values of coefficiens esimaed from equaion prediced by hypohesis : 5

20 Table Inerpreaions and Predicions for Coefficiens for Equaion Coefficien Inerpreaion Expeced Value β Pre-ax earnings persisence of firms wihou large permanen differences (+) β Pre-ax earnings persisence of firms wih large permanen differences in year, relaive o firms wihou large permanen (-) differences β 5 Pre-ax earnings perisence of firms wih large permanen differences in years, -, and -, relaive o all firms wih large permanen differences in year (+) A posiive coefficien for β5 indicaes suppor for H Pre-ax earnings persisence higher for firms wih large sable differences relaive o oher firms wih large differences β β +β β +β +β 5 Pre-ax earnings persisence of firms wihou large permanen differences Pre-ax earnings persisence of firms wih large permanen differences, no-sable Pre-ax earnings persisence of firms wih large sable permanen differences The model for esing hypohesis 4 is equaion 4 below, which is idenical o equaion excep pre-ax book income (PTBI) replaced wih income before exraordinary iems (IBEI). IBEI β 4 α β SablePerm LargePerm β 5 IBEI β IBEI β SablePerm IBEI LargePerm (4) Table 4 presens he inerpreaion and expeced values of coefficiens esimaed from equaion 4 prediced by hypohesis 4. 6

21 TABLE 4 Inerpreaions and Predicions for Coefficiens for Equaion 4 Coefficien Inerpreaion Expeced Value β Afer-ax earnings persisence of firms wihou large permanen differences (+) β Afer-ax earnings persisence of firms wih large permanen differences in year, relaive o firms wihou large permanen (-) differences β 5 Afer-ax earnings perisence of firms wih large permanen differences in years, -, and -, relaive o all firms wih large permanen differences in year (+) A posiive coefficien for β5 indicaes suppor for H4 Afer-ax earnings persisence higher for firms wih large sable differences relaive o oher firms wih large differences β β +β β +β +β 5 Afer-ax earnings persisence of firms wihou large permanen differences Afer-ax earnings persisence of firms wih large permanen differences, no-sable Afer-ax earnings persisence of firms wih large sable permanen differences For hypoheses 5 hrough 8, he sample is divided ino hree groups for permanen differences and hree groups for emporary differences. For emporary differences; group (Small Temporary erence) consiss of firm-years which do no have large emporary differences in year ; group (Large Temporary erence) consiss of firm-years ha have large emporary differences in year ha do no also have large differences in year - and -; and group (Large Sable Temporary erence) consiss of firm-years which have large differences in year, -, and -. For permanen differences hree groups are also consruced similarly o he groups for emporary differences: Group (Small Permanen erence), Group (Large Permanen erence), and Group (Large Sable Permanen erence). Hypoheses 5 and 6 examine he means of prior earnings growh and compare means beween groups. Hypohesis 5 compares means of prior earnings growh beween he hree emporary difference groups. Hypohesis 6 compares means beween he hree permanen difference groups. Prior earnings growh is defined similarly o fuure earnings growh in Lev and Nissim (004), however using pre-ax book income, see equaion 5 below: 7

22 PTBI PTBI PTBI PTBI (5) Where PTBI n is pre-ax book income scaled by average asses in year n. Hypoheses 7 and 8 examine he variance of prior earnings and compare means beween groups, wih hypohesis 5 comparing mean variance beween he hree emporary difference groups and hypohesis 6 comparing mean variance beween he hree permanen difference groups. The variance of prior earnings is calculaed as he variance of he pre-ax income divided by average asses over he hree years from - hrough year. 8

23 RESULTS The sample of firms used for he regression and oher analyses is creaed from all firms in he Compusa daabase wih he necessary daa, as described in he previous secion. The final number of firm-years in he sample is 7,44. Table 5 shows summary saisics and he composiion of he samples for boh emporary and permanen differences. TABLE 5 Descripive Saisics for Temporary and Permanen erence Subsamples Descripive Saisics for Subsamples separaed by Temporary Book-Tax erence Behavior No Large Temporary erence Firms wihou large emporary difference in year Firms wihou large permanen difference in year Variable N Mean Median 5h Pcl 75h Pcl Variable N Mean Median 5h Pcl 75h Pcl PTBI PTBI PTBI PTBI IBEI IBEI IBEI IBEI DTE DTE AvgToAsses AvgToAsses Large Temporary erences Large Permanen erences Firms wih large emporary differences in year, Firms wih large permanen differences in year, bu no par of he sable group bu no par of he sable group Variable N Mean Median 5h Pcl 75h Pcl Variable N Mean Median 5h Pcl 75h Pcl PTBI PTBI PTBI PTBI IBEI IBEI IBEI IBEI DTE DTE AvgToAsses AvgToAsses Large Sable Temporary erences Large Sable Permanen erences Firms wih large emporary differences in year, year -, and year - Descripive Saisics for Subsamples separaed by Permanen Book-Tax erence Behavior No Large Permanen erence Firms wih large permanen differences in year, year -, and year - Variable N Mean Median 5h Pcl 75h Pcl Variable N Mean Median 5h Pcl 75h Pcl PTBI PTBI PTBI PTBI IBEI IBEI IBEI IBEI DTE DTE AvgToAsses AvgToAsses Variable Definiions PTBI + Pre-ax book income in year + PTBI Pre-ax book income in year All variables are scaled by average oal asses, IBEI + Income before exraordinary iems in year + wih he excepion of average oal asses IBEI Income before exraordinary iems in year DTE Deferred ax expense scaled by average oal asses AvgToAsses Average oal asses Table 6 shows Pearson correlaions for seleced variables esed in he hypoheses. 9

24 TABLE 6 Correlaions Among Seleced Variables AT Temporary Permanen Earnings Pre Tax Prior yr Prior yr Variance PTBI Variance EBEI erences erences Before EI Book Income PTBI Growh EBEI Growh Prior years Prior years Asses * 0.05* <.000 < Temporary * 0.0* erences <.000 < Permanen * 0.880* 0.846* * 0.788* erences <.000 <.000 < <.000 <.000 Earnings 0.07* 0.0* 0.880* 0.990* * 0.78* Before EI <.000 <.000 <.000 < <.000 <.000 Pre Tax 0.05* * 0.990* * 0.76* Book Income < <.000 < <.000 <.000 Prior yr PTBI Growh < Prior yr EBEI Growh < Variance PTBI * 0.78* 0.76* * Prior years <.000 <.000 < <.000 Variance EBEI * 0.78* 0.76* * Prior years <.000 <.000 < <.000 * denoes saisical significance a 5% or beer level Table 7 shows ha 644 of he firm years wih large emporary differences had large sable differences and 4,69 firm years had large emporary differences ha are no considered sable. This equaes o % of he observaions wih large emporary differences having large sable emporary differences. For large permanen differences,08 firm years had large sable differences, while,99 firm-years had large differences, equaing o.6% of he large permanen differences being sable. There is no a grea deal of overlap beween he observaions ha have large emporary differences and hose ha have large permanen differences wih only,9 observaions having boh large emporary and large permanen differences. Table 7 shows he breakdown of observaions. Of he 7,44 observaions, 8,4 (66%) have neiher large emporary nor large permanen differences. 0

25 T e m p o r a r y D i f f e r e n c e G r o u p s TABLE 7 Overlap of Temporary and Permanen erence Groups No Large Large & No Sable Large & Sable Permanen erence Groups No Large Large & No Row Sums Large & Sable Sable % 0.6%.7% 80.5% %.45% 0.9% 7.% % 0.% 0.8%.5% Column Sums % 4.0% 4.4% Regression resuls for equaion, esing hypohesis a, can be found below in Table 8: PTBI 4 Table 8 Resuls of OLS Regression of Pre-Tax Earnings Persisence and Temporary Book-Tax erences α β LargeTemp β PTBI β PTBI LargeTemp β SableTemp 5 β PTBI SableTemp Variables Esimae T-Sa Prob(T) Adj R α < β <.000 β <.000 β <.000 β β Hypohesis is suppored, as β 5 is posiive and significan, indicaing ha pre-ax earnings persisence is higher for firms wih large sable emporary differences compared o firms wih

26 large emporary differences ha are no in he large sable group. The persisence of he large emporary differences group (β ) is lower han oher firms as found in prior research, while he persisence of all firms is similar in magniude as well (Hanlon 005). Hypohesis is esed using he regression model in equaion, he resuls of his regression model are presened in able 9: TABLE 9 Resuls of OLS Regression of Afer-Tax Earnings Persisence and Temporary Book-Tax erences IBEI α β LargeTemp β IBEI β IBEI LargeTemp β 4 SableTemp β 5 IBEI Variables Esimae T-Sa Prob(T) Adj R α < β β <.000 β <.000 β β SableTemp The esimae for he coefficien for β 5 is posiive, bu only significan a he 0% level. Hypohesis a is suppored, while hypohesis b is no suppored a he 5% level. Since he firs model ess pre-ax earnings persisence and he second model ess afer-ax earnings, a reasonable explanaion for he difference is he change in ax expense. Firms wih large emporary differences have increased ax expense relaive o firms wih large sable emporary differences. Hypoheses and 4 examine earnings persisence for firms wih large permanen bookax differences. Hypohesis predics higher pre-ax earnings persisence for firms wih large sable permanen differences when compared o firms wih large differences. The resuls of he regression model esing hypohesis using equaion are in Table 0 below:

27 PTBI β 4 TABLE 0 Resuls of OLS Regression of Pre-Tax Earnings Persisence and Permanen Book-Tax erences α β SablePerm LargePermD β 5 iff PTBI β PTBI SablePerm β Variables Esimae T-Sa Prob(T) Adj R α < β <.000 β <.000 β <.000 β <.000 β <.000 PTBI LargePermD iff The resuls of he regression indicae suppor for hypohesis. Addiionally, he resuls indicae ha earnings persisence for firms wih large sable permanen differences is greaer han he remaining sample of firms, as β 5 + β > 0. The finding ha pre-ax earnings persisence is lower for firms wih large differences ha are no sable (β < 0) conrass wih he findings of Jackson (009), who finds ha permanen book-ax differences are relaed only wih afer-ax earnings hrough ax expense. Running he regression model wih large permanen differences bu wihou he disincion beween sable and oher differences produces he resuls shown in able below: TABLE Resuls of OLS Regression of Pre-Tax Earnings Persisence and Permanen Book-Tax erences Sable book-ax differences included in Permanen group bu no separaely in model PTBI α β LargePermD iff β PTBI β PTBI LargePermD iff Variables Esimae T-Sa Prob(T) Adj R α < β β <.000 β <.000 Comparing he resuls from Table o he resuls from Table 0, i is apparen ha a differen conclusion is reached when large sable differences are no included in he model (excep as par of he large differences group). In his model, pre-ax book income persisence is higher for firms

28 wih large permanen differences. This provides srong evidence ha he sable group is imporan when considering book-ax differences and earnings persisence. Hypohesis 4 predics ha afer-ax earnings persisence is higher for firms wih large sable permanen differences, compared o firms wih large differences which are no large and sable. Table presens he resuls of he regression for equaion 5. IBEI β 4 TABLE Resuls of OLS Regression of Afer-Tax Earnings Persisence and Permanen Book-Tax erences α β SablePerm LargePermD β 5 iff IBEI β IBEI SablePerm β Variables Esimae T-Sa Prob(T) Adj R IBEI α < β <.000 β <.000 β <.000 β <.000 β <.000 LargePermD iff Hypohesis 4 is suppored by he resuls of he regression model presened in Table, as β 5 > 0 and saisically significan. Again, he resuls show ha no only do firms wih large sable bookax have higher earnings persisence han firms wih large differences ha are no sable; hey have higher average persisence han he sample. From he resuls of esing hypoheses and, i is clear ha no all book-ax differences have he same implicaions for earnings persisence. Hypoheses hree hrough six predic differences in pas earnings for firms wih large book-ax differences. Hypohesis hree predics ha firms wih large emporary book-ax differences have differen prior earnings growh han This sample has some characerisics which make comparisons wih oher sudies more challenging. Loss firms are generally excluded in year in oher sudies, however hey are excluded in years, -, and - in his sudy. This makes direc comparisons difficul and suggess he possibiliy ha resuls in prior sudies are parially driven by firms wih losses in prior years. Hayn (995) shows ha firms wih prior losses are significanly more likely o have fuure losses. Losses are also significan o book-ax research as hey generae ax deducions increasing book-ax differences. Loss firms are excluded in his sudy o avoid poenial confounds wih loss firm behavior. This limis he generalizaions ha can be made from his sudy in favor of inernal validiy. 4

29 firms ha do no have large emporary book-ax differences. Figure shows he earnings growh for he differen classificaions of emporary book-ax differences. Figure shows ha he earnings growh measured eiher as pre-ax (PTBI) or afer-ax (IBEI) earnings are significanly differen only for he large sable group. This does no suppor eiher lower or higher prior earnings growh as a facor in he lower persisence of firms wih book-ax differences. The large sable group does have lower earnings growh; however, i does no have lower earnings persisence which suggess ha prior earnings may play a role in he relaionship beween book-ax differences and earnings persisence if prior earnings growh and persisence is negaively correlaed. This is wha is found, which is shown in he following resuls. H5 is suppored, as he large sable emporary difference group has lower prior earnings. Earnings growh does no appear o be a driving facor in he relaionship beween large permanen differences and lower earnings persisence, bu may explain why he group wih large sable 5

30 emporary difference has higher earnings persisence han oher firms wih large emporary differences. The comparison for firms wih permanen differences yields differen resuls, as he large permanen group is he group wih differen prior earnings growh. Figure shows he prior earnings growh for he groups of firms wih large permanen differences. The firms wih large permanen differences (bu no in he sable group) prior earnings growh is larger han he small permanen difference group and large sable permanen differences group. This suppors hypohesis 4 and provides evidence ha prior earnings may be a facor in lower earnings persisence for firms wih large permanen differences. Furher invesigaion of he relaionship beween prior earnings growh and earnings persisence shows ha firms wih he highes prior earnings growh have he lowes earnings persisence. Boh earnings growh and persisence are exremely non-linear, suffering from small denominaor issues and disribuion ha is highly skewed wih a large degree of kurosis making 6

31 spearman rank correlaions appropriae. Table shows spearman correlaions for earnings persisence and prior earnings growh. TABLE Spearman Correlaions beween Prior Earnings Growh and Persisence Prior yr Pre-ax Earnings growh Pre-ax earnings Persisence Prior yr Afer-ax Earnings growh Afer-ax earning Persisence Prior yr Pre-ax Earnings growh Pre-ax earnings Persisence Prior yr Afer-ax Earnings growh Afer-ax earning Persisence All values significan a % of greaer level Table shows ha prior earnings growh is negaively relaed o earnings persisence, boh for afer-ax and pre-ax measures of growh and persisence, he bold numbers in he able show he negaive correlaion beween pre-ax earnings growh and pre-ax income persisence and beween afer-ax earnings growh and afer-ax earnings persisence. 4 Firms wih permanen differences have higher levels of prior earnings growh and lower earnings persisence, suggesing ha for permanen differences prior earnings are a driving facor for he relaionship beween large permanen differences and lower earnings persisence. Hypoheses 7 and 8 predic ha firms wih large book-ax differences have differen variabiliy in prior earnings. Figure 4 shows he means of he variance of earnings for he prior hree years for he hree emporary difference groups. 4 The able presens he correlaions for he sample. The correlaions for each of he permanen difference groups are similar in sign, magniude, and saisical significance. In each case, he firms wih larger prior earnings growh have lower earnings persisence, indicaing ha even wihin book-ax difference groups he relaionship beween prior earnings growh and persisence is significan. 7

32 Hypohesis 7 is suppored as boh he large emporary difference group and he large sable emporary difference groups have significanly lower prior earnings variance han he small emporary difference group. The large difference group and large sable difference groups do no exhibi significan differences in variance of prior earnings, however he large sable difference group does exhibi higher earnings persisence. There is no clear link beween variance of prior earnings and persisence based on he comparison of means. 5 Hypohesis 8 predics differences beween he variance of prior earnings for permanen difference groups. Figure 5 shows he means of he variance of earnings for he prior hree years for he hree permanen difference groups. 5 Addiional analysis shows ha here is no significan correlaion beween variance of prior earnings and persisence. 8

33 Figure 5 shows ha he variances of he permanen difference groups are significanly differen, wih he large permanen difference groups having higher variance han he small permanen difference group. The large sable difference group has he highes variance of he hree groups. Despie differences in prior earnings variance, i can be concluded ha he relaionship beween book-ax differences and earnings persisence is no driven by book-ax differences reflecing differences in prior earnings variabiliy as no significan relaionship beween variance of prior earnings and earnings persisence. Boh large permanen differences and large emporary differences have lower earnings persisence, however, hose firm-years in he large permanen group have higher variance of prior earnings, while hose in he large emporary difference group have lower variance of prior earnings. This is also significan o he research design as he sable difference groups do no have more sable prior earnings, eliminaing he possibiliy ha he 9

34 disincion beween firms wih large differences and large sable differences is merely a funcion of more sable prior earnings. Prior earnings growh is found o be a significan facor and may be he more fruiful avenue o explore when looking for facors driving he relaionship beween book-ax differences and earnings persisence. However, even when conrolling for prior earnings growh, some bookax differences are negaively relaed o earnings persisence. Adding earnings growh ino models of earnings persisence for emporary book-ax differences causes he large sable emporary difference group o lose significance, suggesing ha he reason he large sable emporary differences group is differen from he large difference group is lower prior earnings. Adding earnings growh ino models of earnings persisence for permanen book-ax differences causes he large permanen group o lose significance, leading o he conclusion ha he relaionship beween permanen book-ax differences is driven by higher earnings growh among firms wih large permanen differences. 0

35 CONCLUSION This research makes hree conribuions o knowledge abou he relaionship beween book-ax differences and earnings persisence. The firs conribuion is ha large sable book-ax differences have differen implicaions for earnings persisence han large non-sable differences. While firms wih large book-ax differences have lower earnings persisence han oher firms in he sample, hose firms wih large sable book-ax differences have higher earnings persisence han hose firms wih large differences ha are no sable. 6 While variance of prior earnings is shown o be differen among he difference book-ax groups, i is no significanly relaed o earnings persisence and no evidence can be found ha i is a facor in he book-ax earnings persisence relaionship. Addiionally, he sable book-ax difference groups eiher have similar or higher prior earnings variance han he oher large difference groups, suggesing ha sable book-ax differences is no merely a proxy for sabiliy of prior earnings. Second, his sudy shows a relaionship beween pre-ax earnings persisence and permanen differences. Prior research has found a relaionship beween oal differences and afer-ax earnings, emporary differences and pre-ax earnings, and ha permanen differences relae o afer ax earnings hrough ax expense changes. Using a model which includes sable differences separae from oher differences, his sudy finds ha firms wih large permanen difference have lower pre-ax earnings persisence han firms which do no have large permanen differences. I is also shown ha models no accouning for sable differences would reach differen conclusions. 6 To he exen ha some measures of ax aggressiveness use long erm measures (such as cash effecive ax raes measured over 5 years) his is consisen wih he idea ha ax aggressive firms do no have lower earnings persisence (Blaylock e al. 009).

36 Finally, his sudy shows ha prior earnings growh is a significan facor driving he relaionship beween book-ax differences and earnings persisence. The sable emporary differences group has lower prior earnings growh and higher persisence, while he large permanen difference group (no including large sable differences) has higher prior earnings growh and lower persisence. When prior earnings growh is pu ino models, he relaionship beween earnings persisence and hese groups becomes insignifican. Overall, he relaionship beween permanen differences and earnings persisence can be explained largely by prior earnings growh, while he differences beween he large emporary difference group and he large sable emporary difference group can also be explained by prior earnings growh. The separae consideraion of sable differences is significan in he search for prior earnings growh as a facor in he relaionship beween book-ax differences and earnings persisence. Figure 6 and 7 show ess of differences in means beween firms wih large differences of each ype and hose wih small differences.

37 Significan differences in prior earnings growh among groups can be found when sable differences are a separae group, he differences in prior earnings growh are much weaker when sable differences are no considered separaely. Fuure research ino book-ax differences and earnings should consider prior earnings growh as an explanaory facor and conrol for sable book-ax differences. On a macro-economic level prior research shows ha increasing earnings increases book-ax differences (McClelland and Mills 007), and his sudy shows ha prior earnings is negaively relaed wih earnings persisence, suggesing ha in addiion o ax aggressiveness, earnings managemen, and fraud; prior earnings growh should be considered a conribuing facor o he relaionship beween book-ax differences and fuure earnings and considered as par of fuure models. Those firms wih above average prior earnings growh end o have boh large book-ax differences, as well as lower fuure earnings as he earnings paern exhibis reversion o he mean.