Choice of Pension Discount Rate in Financial Accounting and Stock Prices

Transcription

1 CIRJE-F-82 Choice of Pension Discoun Rae in Financial Accouning and Sock Prices Takashi Obinaa Universiy of Tokyo July 2000 Discussion Papers are a series of manuscrips in heir draf form. They are no inended for circulaion or disribuion excep as indicaed by he auhor. For ha reason Discussion Papers may no be reproduced or disribued wihou he wrien consen of he auhor.

2 Choice of Pension Discoun Rae in Financial Accouning and Sock Prices Takashi Obinaa Universiy of Tokyo July 2000 Absrac Under presen financial accouning sandards, in Japan and in U.S., firms can choose pension discoun raes, which hey use for earnings measuremen, a heir discreion. Firs, his paper invesigaes, wha facors affec heir choice of pension discoun raes. The sample firms in his paper reduced heir discoun raes, when marke ineres rae declined in 1990s, more slowly han he endency in he marke. We analyze he causes of his delay by logi model. Regression resuls show ha, given declining ineres rae, he significan facor affecing he firms choice is no leverage, bu profiabiliy (reurn on equiy: ROE). Second, his research invesigaes empirically how sock prices reflec he size of pension discoun rae. Boh unamorized pension obligaions and pension expenses in he year are posiively associaed wih sock prices. However, he coefficiens for he firms, whose discoun raes are higher han median, are smaller han hose for he firms choosing lower raes. Those coefficiens are no significanly differen from zero. These resuls indicae ha he firms, which chose lower pension discoun raes, are valued higher. This asymmeric resul is consisen wih he firs poin in his paper, concerning he firms moives for pension discoun rae choice. Key Words: pension accouning, discoun rae, earnings response coefficiens (ERC), capial marke, Japan

3 1. Inroducion The accouning sandard for a defined benefis pension plan was revised in Japan in By his revision, pension expenses will be measured by he mehod like as U.S. sandard (SFAS No.87). The pension expenses in he year consiss of wo componens. One is he allocaed amoun o each period, discouning fuure pension benefis by he discoun rae ha a firm chooses. Anoher is operaing profis (and losses) from pension asses, and his is deduced from pension expenses. The acuarial mehod, which uses discoun calculaion in allocaing fuure benefis o each fiscal year, is drawn neiher deducively nor consisenly from he exising body of accouning sandards. Moreover, here is no heoreical necessiy ha i should be specified so. Alhough a sligh difference in discoun rae would change dramaically pension expenses and pension liabiliies on he balance shee, he accouning sandard vaguely specifies he discoun rae. Tha is, a firm can choose a discoun rae arbirarily, referring o he ineres rae of a long-erm governmen bond or a long-erm superior corporae bond. A his poin, here is no essenial difference beween he Japanese and he U.S. accouning sandards. Since discouning is neiher inuiive nor naural, and no concepually obvious, we canno deermine he pension discoun rae uniquely. Neverheless, firms can discreionally choose heir pension discoun raes. Under his condiion, wha crierion will firms base on in choosing discoun raes? I is he firs subjec in his paper. Of course, even if firms chose pension discoun raes for he purpose of earnings managemen, invesors would no be misleaded sysemaically. If he capial marke is efficien and invesors are sophisicaed, as far as pension discoun raes are disclosed and he moives of earnings managemen are publicly known, invesors could see hrough he manipulaed numbers and evaluae he firms, assuming he possibiliy of earnings managemen. Then, he second and main subjec in his paper is o confirm how firms choice of pension discoun raes is evaluaed in he capial marke. We invesigae he relaion beween he size of pension discoun rae and sock prices based on a fundamenal-link model. The sample in his sudy consiss of 24 Japanese companies, which prepare he consolidaed financial saemens based on SEC rule and disclose hem in Japan. The sample period sars in he 1990 fiscal year when he applicaion of SFAS No.87 sared, and ends in he 1998 fiscal year (March, 1999). Since he size of sample is quie small, his empirical analysis may have serious resricions. However, even in he U.S., only a few sudies are direcing is aenion o he choice of pension discoun rae, or is evaluaion in he capial marke. Therefore, even if he size of sample is small, he imporance of analyzing for Japanese companies is by no means small. Page 1

4 This paper invesigaed, firs, wha facors affeced heir choice of pension discoun raes. The sample firms reduced heir discoun raes, when marke ineres rae declined in 1990s, more slowly han he endency in he marke. We analyzed he causes of his delay by logi model. Regression resuls show ha, given declining ineres rae, he significan facor affecing he firms choice was no leverage, bu profiabiliy (reurn on equiy: ROE). Second, his research invesigaed empirically how sock prices reflec he size of pension discoun rae. Boh unamorized pension obligaions and pension expenses in he year are posiively associaed wih sock prices. However, he coefficiens for he firms, whose discoun raes are higher han median, are smaller han hose for he firms choosing he lower raes. Those coefficiens are no significanly differen from zero. These resuls indicae ha he firms, which chose lower pension discoun raes, are valued higher. This asymmeric resul is consisen wih he firs poin in his paper, concerning he firms moives for pension discoun rae choice. This paper is organized as follows. Secion 2 describes curren condiions in respec of discoun raes in pension accouning and reviews relaed accouning lieraure o provide a conex for our sudy and a basis for he empirical model we es. Secion 3 analyzes he moives o choose a pension discoun rae, using logi model. In Secions 4 and 5, we invesigae how sock prices reflec he discoun rae by OLS regression. In Secion 4, he book value of equiy model is examined. On he oher hand, in Secion 5, he earnings model is examined. Secion 6 invesigaes he robusness of regression resuls in foregoing wo Secions. Conclusions follow in Secion Curren Condiions and Prior Sudies 2.1 U.S. SITUATIONS AND SITUATION IN JAPAN The pension accouning sandard prescribes ha firms can choose pension discoun raes wih referring o a marke ineres rae. As for he reference rae, however, he accouning sandard SFAS No. 87 only illusraes wo or more examples: a governmen bond rae, a superior corporae bond rae, and an official announcemen rae of Pension Benefi Guarany Corporaion (PBGC). In choosing pension discoun raes, some discreion is given o firms. According o AICPA s invesigaion, here is remarkable variaion in discoun rae among companies as shown in Panel A of Table 1. Alhough he range decreased in 1994 and aferwards, here will sill be a difference by 2.5% beween highes and lowes in Of course, if he kinds of marke ineres rae referred o differ, i is naural ha discoun raes differ, bu he range of variaion is no he quesion here. In his period, wha level were he marke ineres raes ha he accouning sandard Page 2

5 illusraes? CM10 in Panel B of Table 1 denoes he yield of he governmen bond for en years, and CM30 is he yield of he governmen bond for 30 years. AAA denoes he yield of corporae bond of AAA rank by Moody's and AUTL is he yield of he public uiliy bond of A rank. Rae 1 in Panel C is he PBGC rae which calculaes VBO (vesed benefi obligaions) of he pension plan, and Rae 2 is he liquidaion rae by PBGC. When a firm sops a pension program and hands over pension obligaions and pension asses o PBGC, Rae 2 said here will be used for calculaion of pension obligaions. Besides, he informaion on Rae 1 before 1987 has no come o hand. A he ime he ineres rae declined in 1993, he pension discoun raes were also reduced in response o i as presened in Panels A, B and C. I seems ha, however, pension discoun raes were downward rigid a lile, and he reducion widh in discoun rae was smaller whereas various kinds of ineres raes declined sharply. Alhough he same endency as above was observed also in 1996, i is noiceable ha he rend of liquidaion rae of PBGC has deviaed from he rend of bond ineress. While he yield of he U.S. governmen bond declined 0.5% or more and reached 6% in 1996, he liquidaion rae of PBGC rose 1.5% reversely in he meanime, and reached 8% in Thus, when he ineres rae in bond marke declined, neiher cases where a firm did no reduce he pension discoun rae, nor cases where a firm pulled up i on he conrary were agains he accouning sandard. The siuaion, which did no agree wih common sense, has arisen. On he oher hand, how was he siuaion of Japanese firms? Table 2 is he siuaions of hose companies prepare he consolidaed financial saemens based on U.S. SEC rules, and end fiscal year a March 31. In Japan, a conraced ineres rae in pension plan has been fixed o 5.5% for many years by law. Alhough in financial accouning a firm can choose he differen discoun rae form he conraced rae, many firms have chosen 5.5% as a discoun rae a he applicaion sar ime of SFAS No.87. GBI in Table 2 is he yield of he governmen bond for en years, and all numbers are resuls a he end of March (source from Bank of Japan). In Japan, since here is very lile reliable daa on a long-erm ineres rae, we will compare his yield wih pension discoun raes. The firms were relucan o reduce discoun raes, alhough a marke ineres rae declined in his period. Afer a long-erm ineres rae in he marke declined considerably, he firms have reduced heir discoun raes a las. Table 2 provides he evidence ha, even as he ineres rae declined, many firms hesiaed o reduce discoun raes. A his poin, here is no essenial difference beween Japan and U.S. Since he presen condiion became clear, we need o examine why firms are unwilling o reduce discoun raes. Tha is, he moive of a firm is he quesion. Furhermore, we need o examine how sock prices reflec he size of pension discoun rae. Page 3

6 The purpose of his sudy is o examine above wo subjecs. 2.2 PRIOR STUDIES In he earlier empirical sudy, analysis has been concenraed on he issue wheher pension obligaions on acuarial basis are he same obligaions as usual (legal) ones, he issued corporae bond or he borrowing from bank, ec. For example, Marin and Henderson [1983], Maher [1987], Reier [1991, 1992], and Maher and Kez [1993] invesigaed he valuaion of pension obligaions in corporae bond marke by analyzing bond raings or bond premiums. Moreover, he influence ha underfoundings (unfounded pension obligaions) had on firm value has been also research subjec. Feldsein and Seligman [1981], Livna [1984], Dhaliwal [1986], Landsman [1986], Bulow, Mφrck and Summers [1987], Kemp [1987], Gopalakrishnan and Sugrue [1990, 1992, 1993, 1995], and Barh [1991] analyzed he valuaion of pension obligaions on acuarial basis or he valuaion of pension liabiliies on accouning basis in he sock marke. Furhermore, Barh, Beaver and Landsman [1992] invesigaed he relaion beween pension expenses and sock prices. Among he sudies analyzing he influence ha pension obligaions have on sock prices, here are a few sudies ha direced is aenion especially o he variaion in pension discoun rae; Feldsein and Mφrck [1983], Ippolio [1986], Reier [1991], Gopalakrishnan and Sugrue [1990, 1992], and Kwon [1994]. In hose sudies, he consrucive pension obligaions, which could be compued by using he average acuarial assumpions, was chosen as an independen variable. Those sudies invesigaed wheher invesors would adjus he pension liabiliy informaion on accouning basis in such a manner for firm valuaion. However, why should invesors use such a unique mehod? The adjusing mehod is no suppored in heory ye. 1 On he oher hand, research on selecion of accouning policy --- so-called posiive accouning heory --- is anoher main research domain. By analyzing managers ineress prescribed by he conracs, he moives for accouning policy choice and heir changes have been invesigaed in various phases. The deb covenans hypohesis, he managerial compensaion hypohesis, and he regulaion evasion hypohesis are he well-known behavioral hypoheses. Alhough here are no so many empirical sudies focusing on he moives underlying pension discoun rae choice, he following sudies belong o he school of hose research series. Ghicas [1990], Kwon [1994], and Godwin, Goldberg and Duchac [1996] are 1 Francis and Reier [1987] invesigaed he moive of pension founding policy by regression analysis ha chose, as a dependen variable, he pension obligaions sandardized by he official announcemen rae of PBGC. The problem remains on he choice of a dependen variable. And Thomas [1988] includes he same problem. Page 4

7 as such sudies. In addiion, Amir and Gordon [1996] analyzed he discoun rae in measuring pos reiremen benefis oher han pensions. 2 In Ghicas [1990], he moives o change he calculaion mehod of pension expenses, before SFAS No.87 era, was analyzed by logi model. According o Ghicas [1990], he endency was observed, ha he firms whose profiabiliy go worse changed he calculaion mehod ino he new one, which would cu down pension expenses, and chose higher pension discoun raes and made pension obligaions smaller nominally. However, many of independen variables in his research are he resuls (ex pos) brough by changes in pension discoun rae. The condiions (ex ane) a he ime of decision-making wheher o change are no chosen as independen variables. Therefore, in exploring he moives for changes in discoun rae, his model has a miss-specificaion. Kwon [1994] chose he discoun raes he dependen variable, and analyzed he moives for discoun rae choice by OLS regression. Independen variables are he size of oal asses, he deb o equiy raio, he ineres coverage raio, he manager's sock holding raio, and he founding raio. Afer SFAS No.87 era, he deb o equiy raio and he founding raio were saisically significan. However, his research did no choose a marke ineres rae he independen variable, hus oher condiions could no be sufficienly conrolled. Since firms are obliged o choose pension discoun raes wih referring o a marke ineres rae by he accouning sandard, a marke ineres rae should be chosen as an independen variable as a conrol facor in empirical sudy. Moreover, in Kwon [1994] i is ambiguous wheher ineres coverage raio is a proxy for profiabiliy or for financial condiion. As a resul, his conclusion is no clear. On he oher hand, Godwin, Goldberg and Duchac [1996] (GGD hereafer) chose six variables as he independen variables: he changes in marke ineres rae, he rae of reurn, he financial margin of funds for dividends, cash flows, he deb o equiy raio, he exisence of corporae ax paymens. The fac ha he discoun rae was no changed and he magniude of he effec on earnings by he changes in discoun rae were ransformed ino an order measure, and i was chosen as he dependen variable of Ordered Logi Regression. They repored ha he rae of reurn, he financial margin of funds for dividends, he deb o equiy 2 Francis [1987] analyzed he characerisics of he firms ha lobbied agains he proposal of new pension accouning sandard in Sone and Ingram [1988], Sami and Lipka [1989], Seneney and Srawser [1990], Sco [1991], Ali and Kumar [1993, 1994], Brozovsky, Murray and Selo [1993], Harper and Srawser [1993], Langer and Lev [1993], and Espahbodi and Hamer [1996] analyzed he relaion beween he saring year of applicaion of SFAS No.87 and he firm characerisics. The case where he firms can choose he applicaion saring ime of a new accouning sandard is he siuaion ha flexibiliy is high, in he meaning ha pas selecion is no resrained, compared wih he case where he firm would change he once adoped discoun rae. Therefore, i is a suiable maerial o invesigae he incenives of he firm. Besides, here are sudies reaing he change of acuarial assumpions, for example, Morris and Nichols [1984], Rollins [1993], and Blankley and Swanson [1995]. Page 5

8 raio, and he exisence of corporae ax paymens were significan. Alhough he official announcemen rae of PBGC was chosen as a marke ineres rae, i is repored ha changes in ineres rae was no significan saisically. However, he marke ineres raes, such as a governmen bond yield, are no included in independen variables. Therefore, his research does no necessarily show wheher firms had compleely ignored he marke ineres rae movemen when hey chose pension discoun raes. The independen variables, which GGD chose, consised of no only wha was given (ex ane) in choosing a discoun rae, bu also he variable realized as a resul (ex pos). The problem exiss in he choice of independen variables. Moreover, he dependen variable GGD chose was, no he discoun raes, bu he effec brough by he choice of discoun rae. Therefore, GGD has no clarified he moives o choose he discoun rae. In order o analyze he moives, we need o choose he firms choice iself he dependen variable and choose he variables, which firms would face and ook ino consideraion in choosing discoun raes, as he independen variables. We will pay careful aenion o his poin in Secion 3. Amir and Gordon [1996] (AG hereafer), alhough i is reaing pos reiremen benefis oher han pensions (SFAS No.106), analyzed he moives o choose wo facors, he increasing rae of benefis and he discoun rae. AG applied OLS regression, ransforming he combinaion of wo facors ino he sandardized presen value of obligaions, and chose his presen value he dependen variable. In heir resul, only he deb o equiy raio was significan. However, here is no heoreical conclusive mehod for sandardizing pension obligaions. Moreover, he nominal measure, which he combinaions of firm size (oal amoun of marke value of equiy) and price-earnings raio (PER) were ransformed ino, was chosen as he independen variable. The experienial meaning of his variable is very quesionable. Consequenly, in AG, i is no clear wha hypohesis hey assumed. Furhermore, in AG, he consrucive benefi obligaions of each firm was compued by he median of increasing rae of benefis and he median of discoun raes in he sample. Using sock prices as he dependen variable and using he esimaed consrucive obligaions as an independen variable, AG analyzed wheher invesors would correc and value he repored benefi obligaions. However, he meaningful resul was no obained. Two issues are included in his kind of research. The firs issue on he correcion mehod was explained earlier. The second issue is he sock valuaion model. The causal relaion beween he amoun of benefis and sock prices was no sufficienly discussed in AG. In his paper, we emphasize and reconfirm he radiional paradigm in accouning heory: earnings informaion in financial accouning would be used for he predicion of fuure cash flows by invesors, and invesors would ranslae i ino he presen value using he cos of Page 6

9 capial and hen use i for firm valuaion. Considering his paradigm, in his paper, we will build a simple model and use i consisenly as a framework. Based on i, his paper examines how invesors value he pension discoun raes chosen by firms. This paper focuses especially on he foundaion of sock valuaion model, i.e. he heoreical relaion (fundamenal -link) beween accouning informaion and sock prices. Unlike precedence sudies, we place heavy weigh on fundamenals in his paper, and pay careful aenion o ransposing he heoreical relaion o a linear regression model. I is expeced by doing so ha a more precious resul will be obained raher han choosing he asses and liabiliies on he balance shee as independen variables arbirarily. 3. Moives for Pension Discoun Rae Choice 3.1 SAMPLES AND DATA A firm can choose a pension discoun rae, used when calculaing pension obligaions (projeced benefi obligaions: PBO) and pension expenses, a he firm s discreion, as already menioned. This paper does no invesigae how he level of a discoun rae is decided, bu wha crierion firms base on when hey change heir discoun raes. Sample consiss of 24 Japanese companies, which are preparing he consolidaed financial saemens wih he applicaion of SEC sandards in Japan. They adoped SFAS No.87 from The firms ha did no disclose pension discoun raes coninuously were eliminaed. In addiion, he firs applicaion year of SFAS No.87, and he irregular fiscal year in which he accouning period was changed were eliminaed. Finally, he oal sample is 199 firm-years from February 1990 o March We will es his sample hroughou his paper. All accouning daa was hand-colleced from he financial saemens in legal form (YOUKASHOUKEN-HOUKOKUSYO-SOURAN), and pension-relaed daa was aken ou from he foonoes in financial saemens. Since he amoun of pension expenses in he year is no clear in he main par of financial saemens, i is assumed ha he disclosed amoun in foonoe would be he pension expenses in he year. 3.2 EFFECTS BY REDUCTION OF DISCOUNT RATE A firm would choose a pension discoun rae, afer considering he effec on accouning numbers. In general, a firm compares he siuaion where he discoun rae would be changed and he siuaion no changed, and will choose he more favorable one. If so, he quesion on he moives for discoun rae choice would be explained only afer clarifying he anicipaed 3 However, hree companies have applied SFAS No.87 from Page 7

10 unfavorable influences. Since he observable phenomena acually are he resuls and he effecs ha he firm permied, i is difficul o guess, from he realized accouning numbers, he anicipaed unfavorable influences ha were avoided in fac. However, i will be sill useful o confirm he effecs ex pos by he reducion of discoun rae, before solving he moives. In his paper, we compared he various kinds of fundamenal financial raios from wo angles. One is he comparison beween he preceding years and he changed years of he firms ha reduced heir discoun raes. Anoher is he comparison, in he pooled daa, beween he firm-years ha changed discoun raes and he firm-years no changed. The compared financial raios are divided roughly ino hree groups, group I is deb concern, group II is earnings concern, and group III is cash flows concern. The deb-relaed raios are (1) long-erm deb (including pension liabiliies) o equiy (ne-asses) raio, (2) oal deb (including pension liabiliies) o equiy raio, (3) pension liabiliies o equiy raio, (4) off-balanced pension obligaions o equiy raio. Earnings-relaed raios are (5) ne income o equiy raio, (6) earnings before ineress and axes (EBIT) o equiy raio, (7) pension expenses o EBIT raio, and (8) EBIT o ineres paymens raio. Cash flows-relaed raios are (9) cash flows form operaing aciviies o equiy in he previous year raio, (10) cash flows form invesing aciviies o equiy in he previous year raio. Table 3 summarizes he resuls by comparaive analyses. Panel A is he resuls when comparing he financial raios in he preceding year wih hose in he changed year when firms reduced discoun raes. The Z value in he righ end column is Wilcoxon's saisics of signed rank sum es (p-value in parenhesis). Only wo variables, (3) pension liabiliies o equiy raio and (4) off-balanced pension obligaions o equiy raio, show he saisically significan change. Alhough pension liabiliies and pension obligaions increase when he pension discoun raes decrease, he financial condiion and profiabiliy of he firms are no geing worse significanly. However, his resul does no necessarily sugges ha he reducion of discoun rae generally has only small effec on accouning numbers. Possibly he firms, which reduced discoun raes, migh have aken in advance he proecive acions so ha each raio would no be worse. Panel B is he resuls when comparing he firm-years ha reduced discoun raes wih he firm-years no reduced. The Z value in he righ end column is he saisics of Mann- Whiney's U es (p-value in he parenhesis). The cash flows (index (9)) from operaing aciviies in he firm-years, which reduced discoun raes, were more pleniful. This suggess ha he firms reduced discoun raes when heir profiabiliy was good. However, parly because he pension expenses increases as discoun raes decrease, he good performance a he Page 8

11 ime may no be appearing in earnings (index (5)). On he oher hand, i seems ha he reducion of discoun rae was unrelaed o he level of deb o equiy raio. Alhough he reducion brings a seep increase in pension obligaions (indices (3) and (4) in Panels A and B), is effec on leverage is small (indices (1) and (2) in Panel A). Moreover, i seems ha here is no causal relaion beween he deb o equiy raio and he reducion of discoun rae (indices (1) and (2) in Panel B). Public media in Japan are reporing pessimisically, he expeced siuaion a he sar ime of new pension accouning sandard, especially he effec, which pension liabiliies will have on he financial condiion of he firm. However, in he phase of he change of discoun rae, as above-menioned analysis resuls show, he issues on financial condiions seem o be no relevan quesions. We need o examine wha condiion firms are considering when hey change heir pension discoun raes. We will proceed o his subjec in he nex clause. 3.3 LOGIT MODEL If a firm follows he pension accouning sandard mechanically, when a marke ineres rae declines, according o i, he firm should reduce he pension discoun rae quickly. However, he kind of reference rae is no defined uniquely, hus he firm s discreion in choosing he discoun rae is allowed in he accouning sandard. In addiion, firms should no necessarily refer o he ineres rae of he same kind every year. 4 Therefore, i is also possible for a firm no o reduce he pension discoun rae, even if various kinds of marke ineres raes decline, excusing ha he firm has changed he reference rae o he higher one. Probably, i may be a superior plan ha a firm does no reduce he discoun rae a discreion, even if a marke ineres rae would decline, because boh pension expenses and pension liabiliies would increase when a discoun rae would decreases. However, firms may reduce pension discoun raes ineviably, if he difference beween he marke ineres rae and he pension discoun rae exceeds a cerain hreshold. If a firm does no reduce he discoun rae, only considering accouning performance even when a marke ineres rae declines, hose cosmeic aciviies may be negaively valued and he firm value may decrease in he capial marke. If he marke discipline would funcion, no firm may have unresriced freedom in choosing he discoun rae. When analyzing he facors affecing he binary choice, he logi model is used in many cases. 5 Logi model analysis is he mehod of esimaing he coefficiens on a facor by he 4 5 In pension accouning sandard, a firm is no obliged o disclose he kind of marke ineres rae ha i refer. Maddala [1991], Sone and Rasp [1991], Kennedy [1992], and Barniv and McDonald [1999] explain he applicaion and is conroversial poin of he logi model in accouning research. Page 9

12 maximum-likelihood esimaion, assigning a discree nominal measure o he choice of a firm, and using as a dependen variable he raio of he probabiliy in which each choice occurs (Log Odds Raio, LR). In his paper, he firm-years ha did no reduce pension discoun raes are se o 0 (N= 121), he firm-years ha reduced discoun raes 0.5% or less in a fiscal year are se o 1 (N= 52), and he firm-years ha reduced discoun raes more han 0.5% in a fiscal year are se o 2 (N= 26). The reason for seing 0.5% as he boundary line is ha 0.5% is he median and he mode of reducion widh. In our logi analysis, he following model is esimaed: LR α + β Ineres + β Leverage+ Profiabiliy + u (1) = 1 2 β3 where Ineres is changes in marke ineres rae in he year. From a definiion of he dependen variable above-menioned, if he declined size of marke ineres rae is larger and a firm reduces he discoun rae so more grealy, hen β will be negaive. In addiion, if a 1 firm chooses he pension discoun rae only in consideraion of he marke ineres rae principally, and he firm is no aking ino consideraion a all he effecs, which changes in discoun rae has on accouning numbers, hen boh β and 2 β 3 in equaion (1) will be zero. Therefore, he following hypohesis can be assumed. H 1 A firm chooses a pension discoun rae only in consideraion of he changes in marke ineres rae. When he marke ineres rae declines, a firm reduces he pension discoun rae according o i. ( β < 1 0, β 0 2 = β 3 = ) On he oher hand, in he research domain concerning he accouning policy, wo hypoheses, which are no muually exclusive, have been discussed and examined repeaedly. One hypohesis is ha firms avoid he sae ha worsens he deb o equiy raio (leverage). In many cases, mainaining he deb o equiy raio imposed under deb covenans is seemed as an imporan incenive. If he resricive financial covenan is broken, he cos of capial will rise in he fuure. Then, in order o make he possibiliy of echnical defaul lower and o make he cos of capial lower, firms avoid he acions ha would raise he deb o equiy raio. Anoher hypohesis is ha firms ac in order o increase earnings or raise he rae of reurn on equiy. As for he direc moive, he managerial compensaion plan linked wih accouning earnings has ofen been sudied. Moreover, mainaining he reained earnings for smoohing he dividends is said as an indirec moive. Of course, exisence of he resricive financial covenan above-menioned may also serve as he moive for nominal increase in earnings. Page 10

13 However, our research does no specify he moives of firms any more. The reason for adoping his research plan is ha he main saus was no ye given in disclosure sysem o he consolidaed financial saemens in his sample period. Thus, i is no clear how he consolidaed financial saemens were used in a privae conrac. Below, his sudy concenraes on presuming he moives of a firm in he dimension of accouning indices principally, no direcly specifying he moives. We examine in his paper wheher firms has chosen heir discoun raes in consideraion of eiher leverage or profiabiliy. Specifically, we will examine he following wo hypoheses ha are no muually exclusive. H 2a H 2b The firm whose deb o equiy raio is relaively high does no reduce he discoun rae, in order o avoid he much more deerioraion. Conversely, he firm whose deb o equiy raio is low reduces he discoun rae wihou considering i. ( β < 2 0 ) The firm whose profiabiliy (reurn on equiy; ROE) is relaively low does no reduce he discoun rae, in order o avoid he much more deerioraion. Conversely, he firm whose profiabiliy is high reduces he discoun rae wihou considering i. ( β > 3 0 ) Besides, in order o reserve he cash ha a firm can use for business and mainain he pension-founding raio, when cash flows in he year are no pleniful, he firm may no reduce he discoun rae as o conrol he level of pension obligaions. On he conrary, if cash flows are so pleniful ha a firm can increase he pension conribuion corresponding o he increase in pension obligaions, he firm can reduce he pension discoun rae. To es his cash reservaion hypohesis, i is necessary o know he amoun of conribuion o he pension fund each year. However, before 1997, hey were no disclosed. Therefore, we canno es his hypohesis in our sample period. We concenrae on examining above hree hypoheses. 3.4 DEFINITION OF INDEPENDENT VARIABLES Firs issue is wha kind of rae we should adop as a marke ineres rae in he above-shown equaion (1). In general, which marke ineres rae ha firms would refer o in choosing pension discoun raes? In financial accouning heory, he answer is no obvious ye. Moreover, in Japan, he ineres rae of various kinds is used as an index for a long-erm ineres rae in he field of empirical research. In his paper, we chose he yield of he governmen bond for en years ha are raded mos acively as he proxy for a marke ineres rae. Daa was hand-colleced from Financial and Economic Saisics Monhly issued by Page 11

14 Insiue for Moneary and Economic Sudies, Bank of Japan. The following raios are used in his paper as he deb o equiy raio. The numeraor is hree variables respecively, shor-erm deb (SL), long-erm deb (no including pension liabiliies) (LL), and he sum oals of SL and LL (TL). The deflaor is he adjused equiy, which we will explain laer in more deail. The numeraor of profiabiliy index is earnings (ne income) compued as if here has been no pension expenses (ADJNI), EBIT compued as if here has been no pension expenses (ADJEBIT), and oher earnings (OTHER), respecively. The deflaor is he adjused equiy. All variables oher han OTHER are defined as posiive value. As for OTHER, profis and gains are defined as posiive value and expenses and losses are defined as negaive value. Earnings on as if basis is compued in consideraion of ax effec, assuming ha an effecive ax rae is 50%. The adjused equiy here is he equiy compued as if here has been no pension expenses in he year and he oher comprehensive income has no been gained in he year. 6 In shor, all independen variables are creaed, no by he numbers influenced by choice of discoun rae, bu by he numbers given a he ime when firms chose heir discoun raes. The descripive saisics of each variable are summarized in Panel A of Table 4, and he correlaion coefficiens are summarized in Panel B. Alhough he correlaion beween ADJEBIT and OTHER is high, since hese are obained by dividing ne income linearly, we will pu hem ogeher ino regression. However, cauions are required for he high correlaion beween he debs and OTHER. The high correlaion may be caused by he fac ha ineres paymens are included in OTHER. We should judge carefully wheher he resul is sable, since here is a doub of mulicolinearliy. 3.5 ANALYSIS RESULTS The coefficiens of he firm-years ha reduced discoun raes 0.5% or less are presened in Panel A1 of Table 5 and he coefficiens of he firm-years ha reduced discoun raes more han 0.5% are presened in Panel A2. Boh are esimaed a he same ime by logi regression. GBI in he able is he absolue level (%) of governmen bond yield a end-of-year, and GBI is is change in he year. As shown in Panels A1 and A2, he coefficien on GBI is no saisically significan. I is hard o say ha he firms reduced pension discoun raes wih following he decline of marke ineres rae in he year. The hiing raio (percenage of correc esimaion) when he model predicing he sae 1 is a mos abou 30%. On he oher hand, he coefficien on GBI 6 The oher comprehensive income here is only he par of he minimum pension liabiliies. The holding gains and losses of securiies (available for sale) and he foreign currency ranslaion adjusmen accoun are as i is. Page 12

15 is saisically significan a 1%, and he hiing raio is abou 79%. I seems ha he firms had chosen heir discoun raes in consideraion of only marke ineres level, no he change in ineres rae, especially in he cases where he discoun raes were reduced more han 0.5% a once. The resul suggess ha afer he marke ineres rae had declined considerably, he firms reduced pension discoun raes relucanly. Since he firms reaced he ineres rae movemen in he marke neiher mechanically nor imely, he hypohesis H 1 is no suppored. When he deb o equiy raio ogeher wih ADJEBIT and OTHER is included in independen variables, since he correlaion coefficiens beween he deb o equiy raio and OTHER were quie high, he regression resul is quie unsable. Alhough here are a few cases in which significan resuls were obained ((7), (9) and (11) in Panel A1), in hose cases, he sign of coefficiens is conrary o prediced in hypohesis H 2a. Generally, he hypohesis H 2a is no suppored. The firms, when he deb o equiy raio was high, did no reduce heir discoun raes no o worsen leverage. The firms migh have reduced discoun raes regardless of he deb o equiy raio. 7 In conras, in many cases, he coefficiens on he rae of reurn on equiy are significan and he sign is he same as prediced in hypohesis H 2b. As for profiabiliy, he prediced resuls are observed significanly a 5% ((4) and (6) in Panel A1). When EBIT and ohers are separaed, he significan resuls are obained a 5% level ((7), (9), (10) and (11) in Panel A1). These resuls are supporing hypohesis H 2b. Tha is, while he firms did no reduce discoun raes when he rae of reurn on equiy was low, hey had reduced discoun raes when i was high. These resuls indicae ha firms had chosen pension discoun raes, aaching greaer imporance o he rae of reurn on equiy more han o he deb o equiy raio. However, we can say ha firms chose discoun raes in consideraion of he rae of reurn on equiy only if we compare he firm-years where he discoun raes were no reduced wih he firm-years reduced 0.5% or less. As shown in he column of Finess, i seems ha analysis by logi model was no necessarily successful. Then, we assigned measure 1 o he firm-years where he discoun raes were reduced and measure 0 o he firm-years no reduced, and we ried he logi regression once again. The resuls (no repored here) did no differ from hose in Panel A2 of Table 5. The firm-years ha have reduced discoun raes more han 0.5% a a srech migh be quie noisy samples. Thus, we eliminaed he firm-years ha have reduced discoun raes more han 0.5% and ried logi analysis again. The resul is Panel B of Table 5. In order o concenrae on verificaion of hypohesis H 2a and H 2b, i shows only he resuls, choosing he level of marke 7 Even if he pension liabiliies a he end of he previous year was chosen as he independen variable, he resul was unaffeced. Page 13

16 ineres rae as he independen variable. 8 Firs, i urns ou ha limiaion of a sample has conribued o he rise of he finess grealy. Second, he resuls rejec hypohesis H 2a and suppors hypohesis H 2b here oo. In sum, he firms reduced pension discoun raes, considering he rae of reurn on equiy insead of he deb o equiy raio. We can ge from above resuls he answer o he issue on fundamenal concep. I seems an error o see ha choosing a pension discoun rae is an issue of valuing pension liabiliies on he balance shee from asse-liabiliy view. From he radiional maching concep or revenue-expensive view, choosing a pension discoun rae is jus he issue on income measuremen, including he mehod of making he discoun rae zero --- he mehod of allocaion wihou discouning. In fac, he firms had chosen he allocaion paern ha could avoid a sharp decline of profiabiliy hrough manipulaing pension discoun raes. However, as menioned earlier, by he research design in his paper, we canno clarify he conens of moives why firms hink profiabiliy as imporan. An invesigaion of economic incenives is he subjec lef behind o fuure research. 4. Evaluaion in he Capial Marke --- Par One: Book Value of Equiy Model 4.1 CONTROVERSIAL POINTS OF PRIOR STUDIES In many prior sudies, i has been ofen discussed wheher pension liabiliies on he balance shee is negaively associaed wih firm value, assuming ha he asses and liabiliies on he balance shee is he proxy for economic worh of capial sock of a firm. We should check he conroversial poin ha prior discussion holds, before explaining he model we es. Suppose, he oal amoun of marke value of common equiy a he end of year is MVE, and he firm value (oal worh of he invesed projecs in he firm) is amoun of marke value of he deb following equaion is well known. V, and he oal D. The relaion among hese hree variables in he V = D + MVE (2) In he prior research, i is assumed ha he firm value value A in accouning, and in he same manner, he aggregaed marke value of deb be expressed using deb value each relaion as follows. V can be expressed using asses D can L in accouning. Specifically, we can express exclusively 8 The hiing raio was abou 35%, when he change in ineres rae was chosen as an independen variable, and when he level of he ineres rae was chosen as an independen variable, i was abou 77%. Page 14

17 V D = α + β ( β > 0 ) (3) A = α + β ( β > 0 ) (4) L If hese equaions (3) and (4) are subsiued for equaion (2), we can ge he following equaion. MVE = ( α α ) + βa β L (5) I has been invesigaed based on his equaion (5) wheher he coefficien on deb L is negaive. In his way, if he coefficien on pension liabiliies on accouning basis is negaive, some researchers assume ha invesors would regard pension liabiliies and oher debs (legal obligaions) in he same rank, and invesors would negaively value pension liabiliies on accouning basis for firm valuaion. However, he valuaion models like as equaions (3) and (4) are very quesionable concepually. The firs conroversial poin is here. Suppose he following equaions in he form ha is more general. V = a + ba + cl (3 ) D = a + b A + c L (4 ) If above wo equaions are rue, hen equaion (5) should be ransformed as follows. MVE = ( a a ) + ( b b ) A + ( c c ) L (5 ) A his ime, here is nohing o be able o say beforehand wih regard o he sign and size of coefficien on accouning deb L. I is obvious ha he hypohesis of (5') is so-called join hypohesis of (3') and (4'). Since he asses value in accouning generally is no equal o he capialized worh of fuure cash flows, i is an error o assume ha b b = 1 in equaion (5'). A he same ime, since deb is no usually valued by fair values in financial accouning, i is also an error o assume ha c c = 1. 9 Originally, income measuremen is he main objec in financial accouning; herefore, he amoun of asses and liabiliies is no necessarily deermined based 9 However, when he grea porion of asses and liabiliies consis of he so-called financial insrumens and he goodwill of a firm can be ignored saisically (for example, a par of financial indusry), analysis by equaion (5') may have some validiy. Page 15

18 on economic worh saically. Boh asses and liabiliies are jus he dynamic resuls by income measuremen, and invesors would use he disclosed earnings informaion for firm valuaion. The model, which connecs he asses and liabiliies on he balance shee o he firm value direcly, does no have he concepual and heoreical background based on he radiional paradigm, because i ignores he goodwill ha is he source of fuure excess earnings. Neverheless, in he prior research, equaion (5) is handled more inricaely. The asses and liabiliies are usually divided ino some blocks. Here, we will divide ino wo pars, 1 and 2, for convenience. Then equaion (5) is rewrien ino he following equaion (6). ( β A β L ) + ( β A L ) = α (6) MVE ( α ) β 2 2 Suppose ha in par 1, ne deb ne-asses NL 1 ( L A 0) is calculaed on accouning basis, while = 1 1 > NA 2 ( A L 0 ) is calculaed on accouning basis in par 2. The main = 2 2 > purpose for condensing ino ne amoun by each par is o avoid he problem of mulicolinearliy in regression analysis. By hese arrangemens, equaion (6) is furher ransformed ino he following srange equaion (7). MVE γ 0 + γ 1NL1 + γ 2NA2 = (7) Typically, pension liabiliies on he balance shee would be subsiued for he ne deb NL 1, he book value of oher ne-asses will be subsiued for NA 2, hen i has been esed wheher he coefficien γ on pension liabiliies is negaive or no The second conroversial poin exiss here and i is comparaively clear. The quesion is how he projecs in he firm can be divided linearly like equaion (6). Consider a firm has wo invesmen projecs; one is he financial invesmen (invesmen of free cash) and he oher is business invesmen. Since wo projecs are independen (separable) muually, when valuing he firm, wha is necessary is jus o evaluae wo projecs independenly and o sum hem simply. However, he conribuion o he pension fund and he invesmen in pension asses are, for a firm, no a pure financial invesmen. Pension conribuions and pension invesmens are conneced wih wo or more business invesmens closely indivisible; hus, we canno linearly divide ino pension liabiliies and oher ne-asses. In addiion, if ne worh (presen value) of a cerain invesmen projec is negaive, here is primarily no economic 10 For example, Landsman [1986], Barh [1991], Gopalakrishnan and Sugrue [1993, 1995]. Page 16

19 raionaliy o coninue i. Therefore, o assume ha he coefficien on any projec in equaion (7) will be negaive lacks economic raionaliy. There remains a serious problem in discussing he obligaion naure of pension liabiliies on accouning basis by esing he sign of coefficiens in equaion (7). I is imporan o noe he lineariy and economic raionaliy of he regression model. 4.2 HYOPOTHESES AND MODEL Many of prior sudies have no necessarily placed enough weigh on he economical meaning of above-menioned radiional paradigm, bu buil regression models raher easily, and he prior sudies have direced aenion only o saisical inerpreaions of regression resuls. If empirical analysis would be aemped wihou concepual background, common work wih heoreical and empirical sudy would no be aained. From such a viewpoin, Ohlson [1995] model aracs grea concern in recen years. Ohlson divided he sream of accouning earnings ino normal earnings and excess earnings, and he noiced especially ha he former could be expressed as a funcion of he book value of equiy. Moreover, giving he proof ha he cash flow discoun model, he dividend discoun model, and he excess earnings (residual income) discoun model are equivalen, Ohlson formulaed he firm valuaion model using he book value of equiy. 11 This paper builds a regression model by referring o he essence of Ohlson model. Firs, we begin wih he discoun model of permanen earnings. Permanen earnings is defined as he consan earnings ha a firm could produce perpeually. Below, we denoe i as π. I p is assumed in he radiional paradigm ha invesors will predic permanen earnings from accouning informaion disclosed every year and evaluae common equiy, discouning permanen earnings by he cos of capial r. Thus, he following equaion (8) can be assumed. MVE E (π p ) = (8) r The cos of capial r is given and no he quesion here. Second, we assume he relaion beween permanen earnings and he book value of Equiy. According o Ohlson, normal earnings is equal o he produc of he book value of equiy and he cos of capial. If an assumpion analogous o i is applied here, we can express permanen earnings as follows: 11 See Ohlson [1995], Felham and Ohlson [1995], Penman [1998], Penman and Sugiannis [1998]. Page 17

20 E (π p ) = r BVE (9) where rae r. This capial sock BVE is he expeced capial sock ha could produce permanen earnings by annual BVE does no always agree wih he acual book value of equiy BVE. The acual book value of equiy could be influenced by various accouning allocaion mehods of cash flows and managerial assumpion and discreion. Then a cerain kind of noise is included in acual book value. Therefore, when invesors acually use accouning informaion, hey would assume in he following form. BVE = a + bbve (10) Third, we formulae he regression model. If above equaion (10) is subsiued for equaion (9) and is furher subsiued for equaion (8), he following equaion (11) will be obained. 1 MVE = r( a + bbve ) = a + bbve r (11) This equaion (11) is he foundaion of he model in his paper. I is imporan o noe ha coefficien b on he book value of equiy is independenly deermined from he size of he cos of capial r here. Coefficien b implies he raio of he capial sock, which is expeced o produce permanen earnings, o he acual book value of equiy. Of course, in he sae where a firm always produces earnings equal o he muliplied amoun of he book value of equiy by he cos of capial, neiher he noise nor he deviaion of book value exiss and so he coefficien b will be one. However, here is no guaranee ha he regression coefficien will be one, acually. The coefficien may exceeds one or no, so boh cases canno be logically denied eiher. 12 Finally, we inroduce he arge variable relaed wih pension accouning ino equaion (11). Since unamorized pension obligaions UPO is poenially he adjusing iem o he book value of equiy, i can be included in equaion (11). Then, we can esimae he following regression model. 12 One of he reasons is because he goodwill exiss in business invesmen of he firm and excess earnings will be produced. Page 18

21 MVE = 2 + α + β 1 BVE + β UPO u (12) I is imporan poin ha he linear relaion assumpion ha β 2 = β is no rejeced formally 1 here as a calculaion rule. If all pension obligaions were amorized a once, he book value of equiy would decrease necessarily as much amoun as UPO addiionally. However, we do no suppose ha a firm should amorize pension obligaions quickly as soon as possible; because i is repeaedly confirmed in empirical sudies ha leveling-amorizaion (smoohing) is generally useful for invesors o predic fuure cash flows. Raher, on he assumpion ha UPO will no be amorized a once, his paper invesigaes he difference in he informaion conens beween he book value of equiy and unamorized pension obligaions. In his paper, while unamorized pension obligaions is chosen as he independen variable, he oal amouns of pension obligaions, he underfoundings amoun, and pension liabiliies on he balance shee are all no chosen as independen variables. I is he disincive poin where his model differs from ha of he prior sudy. As menioned above, he basic srucure of accouning gives he basis for he linear combinaion in equaion (12). On he conrary, neiher he oal amoun of pension obligaions nor pension liabiliies on he balance shee can be included in independen variables of a linear regression model like equaion (12). There is no necessiy ha i will be a linear model when hey and he book value of equiy are a he same ime included in independen variables. If hey are he same as ha of legal obligaions, such as borrowing, i is sill more so as explained earlier. In his way, he greaes feaure of our model is he poin of including only unamorized pension obligaions. Nex, le us consider he sign of coefficien β 2 on unamorized pension obligaions UPO in equaion (12). I is a wrong expecaion ha he sign will be negaive only by he reason ha he paymens would occur in he fuure. From he premise of our model, he sign would be deermined depending on he relaion beween UPO and permanen earnings. Originally, pension obligaions has wo aspecs. One is he unpaid wages for labor services consumed in he pas, so his aspec has no relaion wih fuure permanen earnings. Anoher is he incenive cos for buonholing he compeen employee o he firm and pulling ou much more abiliy. The laer is an invesmen in human resources ha is usually off-balanced in financial accouning, and i will be reasonable o suppose ha cash inflows will exceed he invesmen. This aspec may have he meaningful relaion wih permanen earnings. Therefore, he coefficien β will be posiive. 2 H 3a When sock price is chosen as a dependen variable and he book value of equiy and unamorized pension obligaions are chosen as independen variables, he laer Page 19