Asian Economic and Financial Review

Transcription

1 Asian Economic and Financial Review journal homepage: hp://aessweb.com/journal-deail.php?id=5002 LOSS RESERVE ADJUSTMENT AND ITS DETERMINANTS: EMPIRICAL EVIDENCE FROM THE UNITED KINGDOM GENERAL INSURANCE INDUSTRY Chi-Feng Wang Deparmen of Business Adminisraion, Naional Yunlin Universiy of Science and Technology, Taiwan Yung-Ming Shiu Deparmen of Risk Managemen and Insurance, College of Commerce, Naional Chengchi Universiy, Taiwan Andrew Adams School of Managemen and Languages, Herio-Wa Universiy, Edinburgh, UK Kuei-Ling Ou Deparmen of Finance, Tunghai Universiy, Taiwan ABSTRACT The purpose of his paper is o assess wheher loss reserving adjusmen exiss in he UK general insurance indusry and, if so, wha moivaes managers o modify reserves. Our research shows ha: insurers bias loss reserve accrual o smooh income or o avoid riggering regulaory inervenion; insurers underesimae loss reserves o avoid reporing small losses; insurers curren and expeced fuure performance affec heir loss reserve esimaion; insurers focused on commercial lines of business underesimae loss reserves more han insurers focused on personal lines; and cerain economic and firm-specific facors influence he accuracy of loss reserves. The findings conribue o our growing undersanding of earnings manipulaion in he non-life insurance indusry and should be of ineres o regulaors, invesors and crediors. Keywords: Loss reserve errors, Earnings disribuion, General insurance. INTRODUCTION The loss reserve held for unpaid claims is generally he larges liabiliy on a general insurer s balance shee and is deermined wih significan managerial judgmen. In some cases, he originally repored loss reserve can be quie differen from he ulimaely developed loss reserve. The loss reserve adjusmen may have nohing o do wih manipulaion bu may simply reflec esimaion problems. On he oher hand, he loss reserve adjusmen may occur due o an insurer s 160

2 managemen manipulaing reserves o make he insurer appear financially healhy, o smooh income or o reduce ax paymens. In his paper, we invesigae he exisence and deerminans of loss reserve errors in he Unied Kingdom (UK) general insurance indusry using sauory reurns over he period 1990 o The UK general insurance indusry provides a poenially ineresing seing wihin which o conduc his invesigaion for wo reasons. Firs, unlike is Unied Saes (US) counerpar, he UK insurance secor is relaively less regulaed (Wang, 2002) and so our sudy offers a poenially cleaner empirical invesigaion of he exisence and deerminans of loss reserve errors. Second, he examinaion of loss reserve errors is worhwhile o conduc in he UK general insurance secor because i is an imporan insurance marke ha generaed annual premiums of billion (US$ billion) in 2002, ranking firs in he Europe and hird in he world (Swiss Reinsurance Company., 2003). Moreover, since he changes in legislaions in he UK during he analysis period do no have significan effecs on how insurers operae, he associaed confounding legal effecs are accordingly minimized. Bu wha are he underlying deerminans of loss reserve manipulaion? A wealh of research has invesigaed he underwriing cycle and suggess ha underwriing cycles are influenced by cerain economic variables. As loss reserve accrual is closely linked o underwriing aciviy, we migh also expec loss reserve manipulaion o be influenced by economic variables. We find his o be he case, wih changes in real GDP, changes in he inflaion rae and he shor-erm ineres rae all significan facors. We also find ha ne premiums wrien scaled by oal asses is a significan explanaory variable. The remainder of he paper is organized as follows. Secion 2 reviews previous sudies relaing o reserve manipulaion and economic variables affecing insurers repored earnings. Secion 3 oulines he research quesions and hypoheses. Secion 4 provides deails of he mehodology and daa employed. Secion 5 summarizes he empirical resuls and Secion 6 is he conclusion. LITERATURE REVIEW The relevan lieraure may be caegorized as: reserve adjusmen by non-life insurance companies; economic variables and he underwriing cycle; and evidence on earnings managemen by noninsurance firms. Reserve Adjusmen by Non-Life Insurance Companies Mos sudies of reserve adjusmen by non-life insurance companies focus on he US properycasualy (P&C) insurance indusry. Early research on he influence of economic facors on he smoohing of repored earnings hrough loss reserve managemen can be raced back o Weiss (1985). Using a pooled cross-secion, ime-series approach she finds a negaive relaion beween 161

3 ineres raes/unexpeced inflaion and loss reserving manipulaion. When ineres raes and unanicipaed inflaion increase, insurers end o underesimae loss reserves in repored financial saemens. However, her work focuses enirely on auomobile liabiliy insurance. Peroni (1992) uses weighed leas squares regression o esablish a direc link beween he income smoohing behavior of insurance managers and heir firm s financial condiion. Over 1,000 US P&C insurance firms are sudied from 1979 o IRIS raios are used o measure an insurer s financial condiion. Peroni (1992) repors ha he incenive o undersae insurance reserves is a decreasing funcion of he acual financial condiion of he insurer. Thus, managers of financially weak insurers end o underesimae heir claim loss reserves relaive o financially srong insurers. Unforunaely, he research does no consider wheher he bias is designed inenionally o mee a paricular financial reporing goal. (Beaver e al., 2003) parially address his research gap. They employ a disribuional approach o consider furher he influence of earnings on reserve bias. Their resuls show ha P&C insurers wih small posiive earnings undersae loss reserves o avoid reporing losses relaive o insurers wih small negaive earnings. They also find ha public and muual insurers are more likely o engage in earnings managemen han privae insurers wih he effec insignifican in he case of privae insurers. Researchers have invesigaed wheher earnings manipulaion is associaed wih masking insolvency problems in he P&C insurance indusry. Gaver and Paerson (2004) conend ha managers have an incenive o bias loss reserves in order o avoid violaing cerain regulaory raios. Mayers and Smih (2004) invesigae wheher muual P&C insurers have an incenive o manage he accouning informaion in heir repored saemens when hey are convering o common sock ownership. Their evidence suggess ha such muual insurers are more likely o engage in surplus managemen because policyholders embedded equiy claims are imporan in he conversion process. Economic Variables and he Underwriing Cycle The underwriing cycle is comprised of periodic hard and sof markes. Sof markes are characerized by loose underwriing sandards, low premiums, and unprofiable underwriing. Hard markes are characerized by igh underwriing sandards, high premiums and profiable underwriing. There are various explanaions for he underwriing cycle. One school of hough is based on he irraional behavior of insurance markes, such as compeior-driven pricing, naive rae-making processes, and capaciy consrains. Anoher is called he raional expecaions/insiuional inervenion hypohesis. This suggess ha he underwriing cycle is caused by exernal facors and marke feaures, no irraional behavior. There is now considerable 162

4 evidence o show ha economic facors have a significan effec on he P&C insurers earnings performance and pricing sraegies. Lamm, Tennan and Weiss (1997) use a generalized leas squares model o analyze wheher changes in insurance premium levels are influenced by he facors of he raional expecaions/insiuional inervenion hypohesis. Insurance markes in welve developed counries over he period 1965 o 1987 are considered, and he resuls sugges ha boh changes in he relevan sock marke index and real gross domesic produc (GDP) are significanly relaed o premium changes in mos counries. Grace and Hochkiss (1995) documen a link beween insurance indusry performance and long-run general economic condiions using a coinegraion echnique. They also show ha real GDP is negaively relaed o premium changes. A number of sudies have invesigaed he influence of ineres raes on underwriing profis. Fields and Venezian (1989) use join generalized leas squares o sudy he effec of economic variables on profis in all lines of he P&C insurance indusry. The resuls sugges no only ha differen lines have individual cycles bu also ha ineres raes have a significan influence on he profiabiliy of insurers. There is posiive correlaion beween ineres raes and insurance operaing margins, and hose lines of business for which invesmen income is imporan are more sensiive o unanicipaed changes in ineres raes. The paper also shows ha disaggregaed models wih ineres rae erms perform beer han simple auoregressive models in explaining he behavior of profis. Dohery and Garven (1992) apply a capaciy consrain model o show ha changes in ineres raes simulaneously influence he insurers capaciy srucure and equilibrium underwriing profi. Their sudy reveals ha changes in ineres raes resul in changes in he level of underwriing profi. Thus, changes in ineres raes may influence he insurer s operaing performance and underwriing pricing decisions. Fung e al. (1998) sudy he facors ha influence insurance cycles in differen lines of business using a vecor auoregressive model. Their empirical resuls show ha ineres raes provide significan explanaory power for he premiums of home muliple perils, oher liabiliy, auo liabiliy and workers compensaion. Leng and Meier (2006) use ime series analysis o invesigae wheher underwriing cycles are an inernaional phenomenon or wheher each counry has is own disinc paern. They conclude ha he facors causing underwriing cycles are counry-specific raher han global. Furhermore, hey find ha he loss raio series is no co-inegraed wih he ineres rae series, which is inconsisen wih esablished heory. 163

5 Evidence on Earnings Managemen by Non-Insurance Firms There has been considerable research ineres in he relaionship beween non-financial firms repored earnings and he accouning discreion of managers. Hayn (1995) documens a disconinuiy in he earnings disribuion of companies using he Compusa daabase: oo few firms repor small losses and oo many firms repor small profis. Burgsahler and Dichev (1997) exend his work and discover ha he earnings kink is due o earnings managemen. Dechow e al. (2003) also show ha, for non-financial firms seleced from he Compusa annual daabase, he number of firms reporing small profis is much greaer han he number of firms reporing small losses. There have been numerous sudies concerned wih he deerminans of loan loss provisions (LLP) in he banking indusry. LLP is generally he larges accrual for mos banks and is under managemen conrol should i wish o bias repored earnings. Kanagarenam e al. (2003)examine he hypohesis ha managers in he banking indusry who are concerned abou job securiy, bias heir repored earnings. The empirical evidence shows ha if a bank has poor curren earnings bu good expeced fuure earnings, managers will end o undersae LLP o smooh pre-managed income. They show ha he LLP adjusmen in he banking indusry correlaes closely wih financial posiion. Rivard e al. (2003) also repor ha managers in he banking indusry use LLP as a ool for income smoohing, and ha managers are more likely o engage in income smoohing afer he Basel Accord risk-based capial requiremens were inroduced in (Pain, 2003) invesigaes he influence of macroeconomic variables and bank-specific facors on LLP in he UK. The resuls show ha real GDP has a significan effec on major UK banks provisions, wih mos banks increasing heir provisions in lean years. Bikker and Mezemakers (2005) examine how bank provisioning behavior is relaed o he business cycle, using 8000 bank-year observaions from 29 OECD counries. They conend ha provisioning is subsanially higher when GDP growh is lower, consisen wih he noion of couner cyclicaliy (i.e. banks decrease provisions during an economic expansion and increase hem during an economic recession). So he evidence suggess ha LLPs are negaively correlaed wih he business cycle and are also influenced by firm-specific facors. I seems reasonable o expec ha he same will apply o he non-life insurance indusry. RESEARCH QUESTIONS AND HYPOTHESES Research Quesions We consider wo main research quesions. Firs, does reserve managemen exis in he UK insurance indusry? Second, wha are he main influences on loss reserve managemen? We adop 164

6 he approaches used in prior lieraure o esablish he exisence of reserve adjusmen. Our invesigaion hen looks more deeply ino he facors influencing reserve esimaion manipulaion. Hypoheses We will address six major hypoheses in null form. The firs hypohesis concerns he firs of our main research quesions. H1: Managers in he UK non-life insurance indusry do no manipulae heir loss reserves. Given previous evidence from he insurance, banking and non-financial secors, we can expec ha earnings manipulaion is more likely o occur when an insurer has small pre-managed losses. We herefore es he following hypohesis. H2: Managers in he UK non-life insurance indusry do no adjus he loss reserves o avoid small losses. The naure of loss reserve accrual in he insurance indusry is similar o LLP in he banking indusry. Following he work of Kanagarenam e al. (2003), we herefore es he following hypohesis. H3: Curren and fuure performance have lile o do wih loss reserve manipulaion. A number of research papers have sudied he underwriing cycle for differen lines of business. However, few have looked a loss reserve adjusmen for differen lines of business. We es wheher firms wriing mainly commercial lines of business are more likely o bias heir claims loss reserve han firms wriing mainly personal lines of business. Our fifh hypohesis is: H4: Managers of commercial lines of business do no manage he loss reserve accrual o avoid losses. Assuming loss reserve manipulaion exiss, we now consider he deerminans of loss reserve managemen. These include economic facors and firm-specific facors. H5: Economic variables and firm-specific facors do no bias he accuracy of loss reserve esimaion. METHODOLOGY AND SAMPLE We adop a wo-phase approach. In he firs phase, we use descripive saisics o assess wheher loss reserve errors exis, and examine he relaion beween loss reserve errors and repored earnings. The second phase invesigaes he deerminans of loss reserve errors. Our iniial sample consiss of all UK non-life insurance firms in he SynThesys Non-Life daabase from 1990 hrough The insurers mus have loss reserves subjec o managerial discreion. Insurers ha cede all premiums o oher insurers are excluded since hey do no have reserves. We also exclude observaions wih an original loss reserve esimae ha differs from he revised esimae by more han 50 percen in absolue value. Firms are firs caegorized ino ne income class (Dechow e al., 165

7 2003) o show he disribuion of earnings scaled by oal asses across he indusry. We expec o find a significan disconinuiy in he earnings disribuion of repored ne income scaled by oal asses. We hen calculae descripive saisics for loss reserve adjusmen across he indusry. If he mean and median adjusmens are negaive, his suggess manipulaion of loss reserves o avoid losses. We hen examine he relaionship beween loss reserve adjusmen and he disribuion of repored earnings in he insurance indusry, employing he following model used by Beaver e al. (2003): ADJUST j, NEGATIVE LIA 6 1 BELOW 2 ABOVE 3 POSITIVE 4 (1) AUTO 5 Alhough several sudies in corporae finance have aken he magniude of he disconinuiy in he earnings disribuion a zero as a measure of earnings manipulaion, Beaver e al. (2003) develop he idea furher for he insurance indusry. To es wheher he disconinuiy in he earnings disribuion is caused by reserve manipulaion, hey divide he earnings disribuion ino five pars: ZERO, NEGATIVE, BELOW, ABOVE and POSITIVE, defined in Table 1. In his model, he coefficiens β1, β4 provide esimaes of he difference in he loss reserve for small (big) loss firms compared o small (big) profi firms. The reserve adjusmen of zero earnings firms is given by α. Addiionally, hey conrol for cerain ypes of business, namely auomobile (AUTO) and liabiliy (LIA), each measured as ne premiums earned for ha ype of business as a percenage of oal ne premiums earned. Table-1. Specificaion of he Loss Reserve Adjusmen Equaion ADJUST j, NEGATIVE LIA 6 1 BELOW 2 ABOVE 3 POSITIVE 4 (1) AUTO 5 Variable Variable Definiion ADJUST The difference beween he developed loss reserves in year +5 and originally repored loss reserves in year, divided by he developed loss reserve in year +5. BELOW The firs earnings group below zero. NEGATIVE The oher earnings groups below zero ABOVE The firs earnings group above zero. POSITIVE The oher earnings groups above zero. LIA Ne premiums earned for he liabiliy business as a percenage of oal ne premiums earned. 166

8 AUTO Ne premiums earned for he privae and commercial auomobile business as a percenage of oal ne premiums earned. We employ his model in our sudy of he UK non-life insurance indusry and go furher o assess: a) he effec of curren and expeced fuure performance on reserve manipulaion; b) wheher insurers wriing commercial business are more likely o manipulae reserves han insurers wriing personal business. Alhough much research has been carried ou o show he exisence of loss reserve manipulaion, here has been relaively lile research assessing he deerminans of reserves error. To ascerain he effec of firm-specific and economic variables on he accuracy of loss reserve esimaion, we apply an OLS model and panel model o invesigae he deerminans of loss reserve manipulaion in he UK non-life insurance marke. The explanaory variables considered in his research are defined in Table 2. Table-2. Specificaion of he Deerminans Equaion ADJUST i, PRETA INF CB 5 1 i, 6 CAPTA 2 i, 7 i, INT 3 STOCK GDP i, 4 (2) Variable Variable Definiion ADJUST i, The difference beween he developed loss reserves in year +5 and he originally repored loss reserves in year, divided by he developed loss reserves in year +5 PRETA i, Ne premiums wrien scaled by oal asses for insurer i in year CAPTA i, Capial divided by oal asses for insurer i in year INT Rae of reurn on hree monh Treasury bills ΔGDP ln(real gross domesic produc) - ln(real gross domesic produc) -1 ΔINFLATION CB i,, ln(inflaion Rae) - ln(inflaion Rae) -1 Overall combined raio for insurer i in year ΔSTOCK ln(sock index) - ln(sock index) -1. ε i, Error erm 167

9 Following prior sudies (Peroni, 1992; Gaver and Paerson, 2004), we use a five-year developmen window because mos claims are paid wihin five years since he loss was incurred. The combined raio is he sum of he raio of expenses before axes o premiums wrien and he raio of losses and loss adjusmen expenses o premiums earned. Oher firm-specific facors are ne premiums wrien and capial boh scaled by oal asses. UK macroeconomic daa for changes in real GDP, he shorerm ineres rae, changes in inflaion rae and changes in sock index are obained from Daasream. We use he FTSE 100 sock index, he main index of he Unied Kingdom sock marke, for changes in he sock index. EMPIRICAL RESULTS Figure 1 shows ha he disribuion of earnings scaled by oal asses is disconinuous. Like Beaver e al. (2003), we documen a significan kink in his disribuion near zero. The number of insurers reporing small profis is much greaer han he number of insurers reporing small losses. This finding lends suppor o he noion ha insurers may repor small profis hrough manipulaing loss reserves. Figure-1. The disribuion of repored ne income scaled by oal asses Table 3 provides descripive saisics for loss reserve adjusmen, measured as he difference beween he developed loss reserve and he originally repored loss reserve, scaled by he developed loss reserve. The resuls show ha loss reserves are significanly oversaed a he 25h percenile, bu unadjused a he 75h percenile. Table 3 also suggess ha boh he mean and he median adjusmen are overesimaed. The resuls rejec he hypohesis (H1) ha managers in he UK non-life insurance indusry do no manipulae heir loss reserves. Table-3. Descripive saisics for he loss reserve adjusmen Year N Mean Sd. Dev. Firs quarile Median Third quarile p-value, mean

10 Toal We now look more carefully a he relaionship beween managemen of he loss reserve accrual and he disribuion of earnings. We employ Equaion (1) and examine he sensiiviy of he resuls in he OLS model, fixed effecs model and random effec model. If he Lagrange Muliple (LM) es saisic is oo high o rejec he hypohesis, we replace he homogeneous pooled ordinary leas squares esimae wih he heerogeneous panel daa model. The Hausman es is used o deermine which kind of panel daa model is fied. Table 4 (Panel A) shows ha he fixed effecs model is he mos appropriae model o examine H2: Managers in he UK non-life insurance indusry do no adjus he loss reserves o avoid small losses. In Table 4 (Panel A) he resuls for he fixed effecs model show ha he magniude of reserve developmen is similar across all regions of he earnings disribuion. However, in Panel B of Table 4, we compare he coefficien of ABOVE and BELOW, and he coefficien of POSITIVE and NEGATIVE. We rejec boh he null hypohesis ha ABOVE=BELOW and he null hypohesis ha NEGATIVE=POSITIVE a he 1% level. Even if we conrol for he auomobile and liabiliy lines of business (righ hand side of he able), he resuls for he earnings disribuion indicaors are subsanially he same. We conclude ha he reserve developmen of small profi insurers is significanly higher han ha of small loss insurers, and he reserve developmen of big loss firms is significanly higher han ha of big profi firms. We furher conclude ha he managers of large loss firms are more incenivised o smooh income han hose of large profi firms. The resuls above are all consisen wih Beaver e al. (2003). Table-4. Regression of loss reserve adjusmen on earnings groups ADJUST j, NEGATIVE LIA 6 1 BELOW 2 ABOVE 3 POSITIVE 4 (1) AUTO 5 169

11 Panel-A. Regression summary saisics Panel-B. Tess of coefficien resricions We use he Whie es o check wheher heeroskedasiciy exiss in he models employed in our research. When he es saisic exceeds he criical Chi-square value, we use Whie s heeroskedasiciy correced covariance marix o derive heeroskedasiciy-consisen esimaes o esimae he fix-effecs models. An alernaive research design for idenifying he impac of earnings managemen on he earnings disribuion is o es he pre-managed earnings disribuion. If he disconinuiy in he earnings disribuion is induced by loss reserve adjusmen, we would expec he disribuion of pre-managed earnings o be more dispersed han repored earnings, especially around zero. Gaver and Paerson (2004) define loss reserve manipulaion as he cumulaive ne adjusmen in he reserve accoun five years afer he iniial loss year. The loss reserve manipulaion is hen subraced from repored earnings. Figure 2 shows ha he disribuion of pre-managed earnings is smooher and more sable han he repored earnings disribuion. This suggess ha insurers modify loss reserves o avoid reporing losses. As anicipaed, here is no disconinuiy around zero in he pre-managed earnings. Figure-2. The disribuion of repored earnings (broken line) and pre-managed earnings (shaded), scaled by oal asse

12 Job Securiy and he Disribuion of Repored Earnings Kanagarenam e al. (2003) consider wheher managers in he banking indusry who smooh earnings hrough he manipulaion of LLP are influenced by heir job securiy. They conclude ha managers of banks wih poor curren performance bu expeced good fuure performance (he poor-good group) have greaer incenive o modify LLP han managers of banks wih good curren performance bu expeced poor fuure performance (he good-poor group). To examine wheher reserve adjusmen in he UK non-life insurance indusry is influenced by managers job securiy, we follow he approach of DeFond and Park (1997), which relaes o non-insurance organizaions, o separae insurers ino four groups. Poor (or good) performance is defined as earnings below (or above) he indusry median earnings and nex year s earnings is aken as a reasonable proxy for anicipaed fuure performance. We combine he job securiy approach and disribuional approach o divide our sample of firms. Table 5 shows wheher reserve adjusmen is influenced by job securiy in he four groups. The resuls of he LM and Hausman ess in panel A sugges ha he mos appropriae model is he fixed-effecs model for equaion (1) wih four variables and he random-effecs model is he mos appropriae model for equaion (1) wih six variables. Table 5 (panel A) shows ha managerial discreion applies in he case of poor-poor firms. The coefficiens on all earnings group variables are posiive and saisically significan, suggesing ha when managers face curren poor performance and anicipae ha he performance will remain poor in he fuure, hey underesimae loss reserve accrual. The LM and Hausman es saisics in panel B favor he random-effecs model in he four facor equaion and he fixed-effec model in he six facor equaion. Panel B repors resuls for hose insurers wih good curren performance and expeced good fuure performance. There is no insurer in he lef ail (NEGATIVE) of he earnings disribuion. The coefficiens of POSITIVE and ABOVE are negaive and significan a he 1% level, and he coefficien of BELOW is negaive bu no significan a he 10% level. We conclude ha good-good firms have an incenive o overvalue loss reserves o save income for he fuure. The LM and Hausman es saisics in panel C favor he random-effecs model in he four facor equaion and he fixed-effecs model in he six facor equaion. The resuls in panel C sugges ha insurers in he poor-good group undersae heir loss reserve esimaes bu he resuls are no saisically significan. However, we rejec he null hypohesis ha ABOVE=BELOW and POSITIVE=NEGATIVE a less han he 1% level, so we believe ha here is a difference in reserve developmen beween small profi and small loss insurers for he poor-good group. I may be ha because managers expec good fuure performance hey recognize inadequae loss reserves o avoid riggering regulaory inervenion. In panel D, insurers wih good curren bu expeced poor fuure performance are concenraed on he righ of he earnings disribuion. The resuls 171

13 sugges ha managers smooh income by oversaing loss reserve esimaes bu he resuls are no saisically significan. Table-5. Regression of loss reserve adjusmen on earning porfolios for job securiy ADJUST j, NEGATIVE BELOW ABOVE POSITIVE AUTO LIA (1) 5 Panel-A. Regression summary saisics of poor-poor firms Panel-B. Regression summary saisics of good-good firms Panel-C. Regression summary saisics of poor-good firms 172

14 Panel-D. Regression summary saisics of good-poor firms Line of Business and he Disribuion of Repored Earnings The A. M. Bes Company classifies homeowner and farm owner muliple peril, auomobile liabiliy and physical damage lines as personal lines. All oher lines are classified as commercial. Division of our sample ino personal and commercial firms is hen ineviably subjecive. We regard an insurer as commercial if 70 per cen of premium revenue comes from commercial lines of business and an insurer as personal if 70 per cen or more of premium revenue comes from personal lines of business. We anicipae ha firms wriing commercial lines are more likely o bias reserve esimaion han firms wriing personal lines. Table 6 (panel A) shows ha he loss reserves of firms focused on commercial lines of business are undersaed in all secions of he earnings disribuion and panel C shows ha for commercial firms, he loss reserves of small profi firms are more biased han hose of small loss firms. Table-6. Regression of loss reserve adjusmen on earning porfolios for commercial and personal line of business ADJUST j, NEGATIVE LIA 6 1 BELOW ABOVE 2 3 POSITIVE AUTO 4 (1) 5 Panel-A. Regression summary saisics of commercial firms 173

15 Panel-B. Regression summary saisics of personal firms Panel-C. Tess of Coefficien Resricions of Commercial Firms Unlike he resuls repored in panel a, none of he coefficiens in panel B are saisically significan, so here is lile evidence o sugges ha managers of firms focused on personal lines of business underesimae loss reserves. This is consisen wih he noion ha managers of such firms have limied opporuniies o bias accouning numbers, as repored in he lieraure. Deerminans of Loss Reserve Managemen Wha facors influence managers o engage in reserve managemen? We invesigae his issue using economic variables which have significan power in explaining he underwriing cycle ogeher wih economic facors and firm-specific facors which affec he LLP manipulaion in he banking indusry as repored by Bikker and Mezemakers (2005). However, preliminary invesigaions showed a srong correlaion beween earnings scaled by oal asses and capial scaled by oal asses. The earnings facor is herefore eliminaed from he analysis. The regression resuls in Table 7 show ha premium revenue (scaled by oal asses) is negaively associaed wih reserve adjusmen, suggesing ha premium revenue increases a a ime when insurers overvalue loss reserves o avoid insolvency. Oher firm-specific facors are no significan. Table-7. Regression summary saisics of relaion beween economic variables and reserve adjusmen ADJUST i, PRETA INF CB 5 1 i, CAPTA 6 2 i, 7 i, INT 3 STOCK GDP i, 4 (2) 174

16 The coefficien for ineres rae is posiive and significan, which is consisen wih Weiss (1985). The resul suggess ha a rise in ineres raes, which direcly influences insurers invesmen reurns, leads o a sof insurance marke and loose underwriing sandards. According o underwriing cycle heory, we would expec o see low premiums and unprofiable underwriing in sof markes leading o managers underesimaing loss reserves o offse fuure unfavorable underwriing resuls. The coefficien of change in GDP is posiive and significan. So managers use heir discreion o undersae reserve esimaion in economic expansion condiions. The resuls also sugges ha change in inflaion rae is negaively relaed o reserve manipulaion. The implicaion here is ha when inflaion increases, real invesmen income decreases and managers overvalue loss reserves o mainain solvency. To es for robusness, we calculae he correlaions beween shor erm ineres rae, changes in inflaion and GDP growh rae and conclude ha he correlaions are low. The coefficien of he change in sock index is no significan. Thus, here is no evidence ha performance of he sock marke influences loss reserve adjusmen. CONCLUSIONS Loss reserves represen he larges liabiliy esimaed a managers discreion on non-life insurer balance shees, so i is he primary means of exercising discreion in he managemen of earnings. In his paper, we have esed a number of hypoheses concerning loss reserve manipulaion in he UK non-life insurance indusry. Our empirical resuls reveal several ineresing findings: 1. Reserving manipulaion exiss in he UK non-life insurance indusry, a finding consisen wih he lieraure on US propery-casualy insurers (Peroni, 1992; Beaver e al., 2003; Gaver and Paerson, 2004). 2. Managers of UK non-life insurers adjus loss reserves o avoid small losses, as repored in prior research for he US propery-casualy indusry (Beaver e al., 2003; Dechow e al., 2003). The number of insurers reporing a small profi is greaer han he number reporing small losses. The earnings disribuion is disconinuous around zero bu becomes smooh afer subracing he discreionary loss reserve accrual. Resuls also show ha insurers wih big losses manipulae heir loss reserve accrual o avoid riggering regulaory inervenion. 175

17 3. Small profi insurers in boh poor-poor and poor-good groups are more likely o underesimae reserves compared o small loss insurers. However, we believe ha he underlying reasons are differen for he wo groups. Managers of small profi firms in he poor-poor group adjus loss reserves o avoid riggering a regulaory invesigaion bu managers of small profi firms in he poor-good group adjus loss reserves o smooh earnings. This is he firs sudy o invesigae wheher managers manipulaion of loss reserve manipulaion is influenced by heir job concerns. 4. Firms wriing mainly commercial lines of business are more likely o bias heir claims loss reserve han firms wriing mainly personal lines of business. This is consisen wih he view of Peroni (1992) ha managers of personal lines of business have less opporuniy o influence repored reserves. We find no evidence of reserve adjusmen for firms concenraing on personal lines of business. 5. Boh economic facors and firm-specific facors are associaed wih accuracy of reserve modificaion. The explanaory power of changes in real GDP, changes in he inflaion rae and he shor erm ineres rae is high in each case. Moreover, our resuls sugges ha ne premiums wrien scaled by oal asses helps o explain loss reserve manipulaion. Inadequae loss reserves may imply poenial insolvency problems, hence he need o invesigae reserve manipulaion and is deerminans in he UK non-life insurance indusry. If regulaors do no deec earnings manipulaion in ime, hey are likely o face more serious problems a a laer sage. Our research provides he auhoriies concerned wih suggesions for monioring non-life insurers more effecively. REFERENCE Beaver, W.H., M.F. McNichols and K. Nelson, Managemen of he loss reserve accrual and he disribuion of earnings in he propery-casualy insurance indusry. Journal of Accouning and Economics, 35: Bikker, J.A. and P.A.J. Mezemakers, Bank provisioning behavior and procyclicaliy. Journal of Inernaional Financial Markes, Insiuions & Money, 15: Burgsahler, D. and I. Dichev, Earnings managemen o avoid earnings decreases and losses. Journal of Accouning and Economics, Dechow, P.M., S.A. Richardson and I. Tuna, Why are earnings kinky? An examinaion of he earnings managemen explanaion. Review of Accouning Sudies, 8: DeFond, M.L. and C.W. Park, Smoohing income in anicipaion of fuure earnings. Journal of Accouning and Economics, 23:

18 Dohery, N.A. and J.R. Garven, Insurance cycle:ineres raes and he capaciy consrain model. Journal of Business, 68: Fields, J.A. and E.C. Venezian, Ineres raes and profi cycles:a disaggregaed approach. The Journal of Risk and Insurance, 56: Fung, H.G., G.C. Lai, G.A. Paerson and R.C. Wi, Underwriing cycles in propery and liabiliy insurance:an empirical analysis of he indusry and by line daa. The Journal of Risk and Insurance, 65: Gaver, J.J. and J.S. Paerson, Do insurers manipulae loss reserves o mask solvency problems? Journal of Accouning and Economics, 37. Grace, M.F. and J.L. Hochkiss, Exernal impacs on he propery-liabiliy insurance cycle. The Journal of Risk and Insurance, 62: Hayn, C., The informaion conen of losses. Journal of Accouning and Economics, 20: Kanagarenam, K., G.J. Lobo and R. Mahieu, Managerial incenives for income smoohing hrough bank loan loss provisions. Review of Quaniaive Finance and Accouning, 20: Lamm, Tennan, J. and M.A. Weiss, Inernaional insurance cycle raional expecaions/insiuional inervenion. The Journal of Risk and Insurance, 645: Leng, C.C. and U.B. Meier, Analysis of muli-naional underwriing cycles in propery-liabiliy insurance. The Journal of Risk Finance, 7: Mayers, D. and C.W. Smih, Incenives for managing accouning informaion: Propery-liabiliy insurer sock-charer conversions. The Journal of Risk and Insurance, 71: Pain, D., The provisioning experience of he major uk banks: A small panel invesigaion. Bank of England Working Paper (177). Peroni, K.R., Opimisic reporing in he propery-casualy insurance indusry. Journal of Accouning and Economics, 15: Rivard, R.J., E. Bland and G.B.H. Morris, Income smoohing behavior of u.s banks under revised inernaional capial requiremens. Inernaional Advances in Economic Research, 9: Swiss Reinsurance Company., 2003 World insurance in 2002: High premium growh in he non-life insurance, sigma, Wang, W.H., Reinsurance regulaion: A conemporary and comparaive sudy. Kluwer law inernaional firs Edn. Weiss, M., A mulivariae analysis of loss reserving esimaes in propery-liabiliy insurers. The Journal of Risk and Insurance, 52: