8. Türkye Ekonometr ve İstatstk Kongres 24-25 Mayıs 2007 İnönü Ünverstes Malatya Compettve Condtons n the Turksh Non-Lfe Insurance Industry Adnan Kasman 1 * Evrm Turgutlu 2 Abstract: Ths paper nvestgates the evoluton of market structure n the Turksh nsurance ndustry over the perod 1996-2004, usng the Panzar and Rosse (1987) methodology. The sample perod has been dvded nto three sub-perods (1996-1998, 1999-2001, and 2002-2004). The results suggest that n the frst and second sub-perods, the nsurance frms operatng n the Turksh nsurance ndustry earned revenues under the monopoly or conjectural varatons short-run olgopoly. In the thrd perod, however, the results ndcate that nsurance market was nether monopolstc nor perfectly compettve. Frm revenues were earned as f operatng under monopolstc competton. Overall, the results show that market concentraton s not sgnfcantly related wth compettve conduct. JEL classfcaton: G22; L13 Keywords: Competton; Market Structure; Turksh Non-lfe Insurance Industry. 1 Assocate Prof. Dr., Fax:+90 (0)232 453 5062, e-mal: adnan.kasman@deu.edu.tr 2 Dr., Department of Economcs, Faculty of Busness, Dokuz Eylül Unversty, 35160, Buca, Izmr, Turkey, Tel: +90 (0) 232 412 8209, Department of Economcs, Faculty of Busness, Dokuz Eylül Unversty, 35160, Buca, Izmr, Turkey, Tel: +90 (0) 232 412 8211, Fax:+90 (0)232 453 5062, e- mal: evrm.gursoy@deu.edu.tr
1.Introducton Turksh economy and the fnancal system have experenced major changes over the past decades. In partcular, polces towards fnancal lberalzaton and restructurng wth the goal of establshng a compettve envronment and enhancng effcency have been mplemented. Deregulaton process started predomnantly n the bankng system. However, transformaton phase n the nsurance ndustry followed wth 5 to 10 years lag. Untl md-1980s Turksh nsurance ndustry was heavly regulated. New entres were restraned and the prces were set under regulatory condtons. By the end of 1980s the lberalzaton process started and nsurance ndustry moved gradually from strct regulaton towards lberalsm. Ths process has consderable mplcatons for the structure of the ndustry and conduct of the frms. Hence, Turksh nsurance ndustry offers an attractve feld for researchers to examne the evoluton of ts compettve structure. On the contrary, the lterature lacks studes addressng ths ssue. Merely, a recent study by Celk and Kaplan (2007) focuses on the compettve structure of the Turksh nsurance ndustry over the perod 2002-2004. To the authors best knowledge there exsts no study examnng whether or not the market structure of the nsurance ndustry has changed due to the developments n last two decades. In ths context, we am to assess the compettve structure and contestablty of the Turksh non-lfe nsurance ndustry over the perod 1996-2004, through three-year sub-perods. We estmate the reduced-form revenue equatons for the sample of non-lfe nsurance frms and follow the Panzar and Rosse (1987) approach to test for competton. The accurate measures of the varables (nput prces and total revenues) have extreme mportance n ths methodology. Hence, we am to contrbute the relevant lterature by defnng an nsurance-ndustry specfc revenue functon. The rest of the study s organzed as follows: Secton 2 gves a bref overvew on the Turksh nsurance ndustry. Secton 3 presents the methodology. Data set, varables and emprcal results are reported n Secton 4. The last secton s devoted to conclusons. 2. An Overvew on Turksh Insurance Industry The hstory of the nsurance operatons n Turkey dates back to the md-19 th century. However, a natonal nsurance system s relatvely young snce foregners domnated the fnancal actvtes durng the Ottoman Empre perod. Untl the establshment of the Republc of Turkey n 1923, foregn domnaton, nsuffcent captal accumulaton and lack of legal and regulatory adjustments were the man features of the Turksh nsurance ndustry, whch caused unfar competton and created conflct of nterest between the frms and the customers. The exstng regulatons could not provde sound solutons to settle these conflcts. Hence, some attempts to orgnate a regulatory framework for the nsurance ndustry were made followng the establshment of the Republc of Turkey. The prmary goal of these attempts was to buld up a natonal nsurance ndustry. However, development of the nsurance ndustry lagged behnd the other sectors of the economy. Per capta premum ncreased only about $5 from 1923 to md-1980s. Protectonsm had been the man characterstc of the Turksh nsurance ndustry untl md 1980s. The market entry was restrcted and product prces were set by the state. The major step towards lberalzaton was the achevement of free entry and ext rght for frms n 1987. Moreover, by 1990, nsurance frms have started to set prce of the voluntary non-lfe nsurance polces under market condtons, freely. Another major step, whch enhanced competton, came about n 1994 by the legal adjustment enforcng nsurance frms to specalze n ether lfe or non-lfe branches. These changes resulted n a dramatc ncrease n the number of frms and total premums. However, the relatvely small portfolo wth respect to the number of frms created destructve competton. The larger frms 1
ntensfed the competton by cuttng prces and allowng longer payment perods for ther customers. Hence, n 1994 and 1999 new regulatory acts were effectuated to restructure the compettve condtons n the market. The Marmara earthquake n 1999 and the fnancal crss n 2001 had mportant mpacts on the Turksh nsurance ndustry. The earthquake lad a burden of $10 bllon on the Turksh economy. Hence, a new nsurance lne, compulsory earthquake nsurance was created. Ths precautonary nsurance formed a new channel for competton among the ncumbent frms. The effect of the fnancal crss n 2001 was observed manly n the bankng ndustry. The nsurance frms, however, were affected relatvely to a smaller extent. The operatng lcenses of some frms were cancelled. Total number of frms, premums and total assets declned n ths perod. A recent development n the Turksh nsurance ndustry s the mplementaton of the Prvate Penson Savngs and Investment System n October, 2003. Some of the exstng lfe nsurance companes started to operate both n lfe and penson lnes. The system created a new channel to attract potental nvestors and ams to ncrease the natonal propensty to save. Turksh nsurance frms are oblged to operate ether n lfe or non-lfe branches. In ths study, we focus on the dynamcs of the non-lfe nsurance ndustry. Table 1 summarzes the man structural characterstcs of the Turksh non-lfe nsurance ndustry over the perod 1996-2004 and reports total number of frms, total assets, total premums, the four-frm concentraton rato (C4) and the Herfndahl-Hrschman Index (HHI) as derved from total assets 3. Both the C4 and HHI are the ndcators of underlyng market structure. The number of frms fluctuates over the sample perod. After a slght ncrease n the begnnng of the perod, t declnes and reaches 32 n 2004. The concentraton ndcators C4 and HHI follow a cyclcal pattern smlar to the number of frms along the nne years. The concentraton ratos are hgh and seem to ncrease n recent years n the Turksh non-lfe nsurance ndustry. By examnng these concentraton ratos, one can nfer that the revenues n ths ndustry were generated from an olgopolstc-type market structure durng the sample perod. Table 1. Turksh non-lfe nsurance ndustry (1996-2004) Year Number of Frms Total Assets Total Premums (Mllon $) (Mllon $) C4 HHI 1996 38 1236.38 972.54 0.449 0.071 1997 42 1308.95 1102.91 0.422 0.065 1998 41 1499.86 1330.50 0.414 0.067 1999 40 1945.50 1315.52 0.488 0.081 2000 39 2271.39 1910.70 0.448 0.076 2001 36 1484.03 1279.98 0.461 0.078 2002 36 1611.32 1621.04 0.428 0.075 2003 39 2632.55 2540.12 0.445 0.077 2004 32 3634.54 3799.79 0.458 0.080 Source: Annual reports of the Insurance Supervsory Board of Republc of Turkey, Prme Mnstry Undersecretarat of Treasury. 3. Methodology Although the dscusson n the prevous secton provdes some nsghts on the compettve structure n the Turksh non-lfe nsurance ndustry, t reveals lttle nformaton about the compettve behavor of the nsurance frms n the ndustry. The prevous studes nvestgatng the compettve condtons of the fnancal nsttutons used structural and non-structural approaches. Structural approach nfers the 3 The HHI s the sum of the squared market shares of the ndvdual nsurance frm. 2
nature of competton n an ndustry from ts structural characterstcs, whch are concentraton, market share, number of frms (see Ban, 1951). The two man structural approaches are the structure-conduct-performance (SCP) hypothess and the effcentstructure (ES) hypothess. The SCP hypothess asserts that frms n a concentrated market can extract monopolstc rent as a result of mperfect competton. Hence, the ndustry s performance depends on the conduct of sellers and buyers, whch depends on the structure of the ndustry. The ES hypothess, however, states that the explanaton for the postve relatonshp between market concentraton and proftablty s effcency. Effcent frms ncrease n sze and market share due to ther ablty to generate hgher profts. Accordng to the ES hypothess, the concentraton-performance relaton s not due to colluson but s rather a result of frm effcency (see Demsetz, 1973 and Smrlock, 1985). As Bkker and Haaf (2002) state, the compettveness of a bankng ndustry cannot be clearly explaned usng structural approaches because the structural approaches gnore the relatonshp between market contestablty and revenue behavor at the frm level. In reacton to the theoretcal and emprcal defcences of the structural models, non-structural models of compettve behavor have been developed. Baumol (1982) states that olgopoles and monopoles sometmes behave very much lke perfectly compettve frms. Therefore, frms behave dfferently dependng on the market structure n whch they operate. The man non-structural models are Iwata model (Iwata, 1974), the Bresnahan model (Bresnahan, 1982) and the Panzar and Rosse (P-R) model (Rosse and Panzar, 1977; Panzar and Rosse, 1987). These non-structural models measure competton and emphasze the analyss of the compettve conduct of frms wthout usng explct nformaton on the structure of the market. In ths study, we use P-R approach to assess the degree of competton n the Turksh non-lfe nsurance ndustry. The Panzar and Rosse (P-R) model A test developed by Rosse and Panzar (1977) and Panzar and Rosse (1987) examnes the relatonshp between nput prces and equlbrum gross revenue derved from the theory of the frm under alternatve assumptons abut compettve condtons. Ths test s based on propertes of a reduced-form revenue equaton at the frm level. Panzar and Rosse defne a measure of competton, the H-statstc, whch s the sum of the elastctes of the reduced-form revenue functon wth respect to nput prces. Ths statstc represents the percentage varaton of the equlbrum revenue derved from a unt percent ncrease n the prce of all nputs used by the nsurance frm. The P-R test s derved from a general market model, whch determnes equlbrum output level and number of frms, by maxmzng profts at both the frm level and ndustry level. Ths mples that frm maxmzes ts proft where margnal revenue (MR) equals margnal cost (MC): MR ( q, n, z ) MC ( q, w, s ) = 0 (1) where q s the output of the frm, n s the number of frms, w s a vector of nput prces of frm, z and s are the vector of exogenous varables shftng frm s revenue and cost functons, respectvely. At the ndustry level, however, when frms are n equlbrum, the zero proft constrant holds: * * * * * MR ( q, n, z ) MC ( q, w, s ) = 0 (2) 3
where varables marked wth an astersk (*) denote equlbrum values. As mentoned above, Panzar and Rosse (1987) argue that the market power of a frm can be measured by the extent to whch a change n nput prces ( dw ) are reflected n the equlbrum * revenues ( dr ) earned by frm. By ndcatng H as the sum of elastcty of total revenues wth respect to nput prces 4 : k H = K k= 1 R w * k w R k * (3) Panzar and Rosse (1987) show that H-statstc s equal to 1, when frms operate under perfect competton 5. Therefore, an ncrease n nput prces ncreases both margnal costs and average costs wthout changng the equlbrum output level of the average frm n the ndustry. Ext of some frms ncreases the demand faced by each of the remanng frms, leadng to an ncrease n prces and revenues equvalent to the rse n costs. A negatve value of H ndcates that the structure of a market s a monopoly, a perfectly colludng olgopoly, or a conjectural varatons short-run olgopoly 6. If H s negatve, an ncrease n factor prces ncreases margnal costs and reduces equlbrum output. Snce a proftmaxmzng monopoly never pushes ts sales nto the range where the demand curve s nelastc, an ncrease n factor prces leads to a reducton n total frm revenue. H-statstc s postve and less than 1 n the case of monopolstc competton wth freedom of entry. Ths s based on the premse that under monopolstc competton, ndvdual frms face an nelastc demand curve and hence revenues ncrease less than proportonately to changes n nput prces. A crucal feature of the H-statstc s that t must be consdered as observatons, whch are n long-run equlbrum. Hence, equlbrum test should be carred out to verfy that nput prces are not correlated wth ndustry returns. To test f observatons are n longrun equlbrum, one can assume that compettve markets equalze the return rates across frms, so that n equlbrum these rates should not be correlated wth nput prces. Emprcally, the equlbrum test can be carred out by usng an ndcator of frm return, namely return on assets (ROA) as dependent varable wth the same ndependent varables used n the orgnal model n the estmaton of H. In ths context, H = 0 and H < 0 mply that the data represent ndustry equlbrum and dsequlbrum, respectvely (see Shaffer, 1982). As mentoned n Panzar and Rosse (1987), ths hypothess s mportant for the cases of perfect competton ( H = 1) and monopolstc competton ( H > 0), whle t does not consttute an mportant prerequste n the case of monopoly snce H 0 s a long-run condton for monopoly. Hence, f the data s not n the long-run equlbrum, H 0 no longer establshes monopolstc market condtons, but t remans true that H > 0 dsproves monopoly or short-run conjectural varaton olgopoly. 4 Several assumptons need to be made to use ths model n ths study. The assumptons are: a) nsurance frms are sngle product frms; b) hgher nput prces are not assocated wth hgher qualty servces that generate hgher revenues, snce such a correlaton may bas the computed H-statstc; c) nsurance frms are n long-run equlbrum. Further assumptons nclude proft maxmzaton and normally shaped revenue and cost functons. 5 Shaffer (1982) shows that the same results mght be obtaned for a natural monopolst operatng n a perfectly contestable market. 6 Vesala (1995) proves that the same results hold for monopolstc competton wthout the threat of entry,.e. wth a fxed number of nsurance frms. Moreover, Shaffer shows that H s negatve also for any conjectural varatons olgopoly. 4
To estmate the H-statstc, the followng specfcaton of the reduced-form revenue equaton for a panel data set s used: ln TRt = α + β1 ln PLt + β 2 ln PBSt + β 3 ln PFK t + β 4 ln TAt (4) + β ln ETA + β ln LTA + ε 5 t 6 t t Where t s the subscrpt ndcatng nsurance frm at tme t, TR s the total revenue. Insurance frms engage n two branches of actvtes; frst, they serve as fnancal ntermedares and receve a fnancal ncome through ths channel. Second, they provde rsk poolng and rsk bearng servces, from whch they generate techncal ncome. Hence, the total revenue of an ndvdual nsurance frm n the ndustry s defned as the sum of fnancal and techncal ncome. The PL, PBS, and PFK are the unt prce of labor, unt prce of busness servces, and unt prce of fnancal captal, respectvely. TA, ETA, and LTA are the control varables and represent total assets, the rato of equty captal to total assets, and rato of losses pad to total assets, respectvely 7. In ths study, we assume that nsurance frms generate ther revenues by employng three major nputs: labor, busness servces and fnancal captal n lne wth the prevous studes (see Cummns and Wess, 1993; Cummns and Z, 1998; Cummns et al., 2004; Cummns et al., 2006). Snce P-R methodology rests on testng the sum of the nput prce elastctes of the frm s revenue, we should defne the nput prces. Personnel expenses over the number of employees are used as a proxy for the unt prce of labor, PL. Insurance frms use varous non-labor nputs such as buldngs, computers, offce materals, etc. to produce nsurance servces. Non-labor nputs can be commonly named as busness servces. The prce of busness servces, PBS, s measured as the rato of nonlabor expenses to total assets. Fnancal captal, PFK, s the thrd nput of the nsurance frms whch serves as a prmary nput for rsk bearng and rsk poolng functons. Insurance frms hold captal to back the losses they promsed to pay whch are larger than expected and to satsfy the regularty requrements (Cummns and Z, 1998). Followng Cummns and Z (1998), we measure the prce of fnancal captal by takng the three year movng average of the rato of net ncome to equty captal. Our reduced-form revenue functon ncludes three frm-specfc explanatory varables to account for the sze, captal structure and rsk-compensaton of the frms. We use total assets (TA) to control for the scale of the nsurance frms. We expect a postve relatonshp between the sze of the nsurance frms and the revenue generated. The second control varable s the rato of equty captal to total assets (ETA). Ths rato represents the captal structure of the nsurance frms. The expected sgn of ths varable s uncertan. A hgher rato mples lower nsolvency probabltes and hence may nduce hgher revenue. However, a hgher rato may also prevent frms from producng ther own outputs whch n turn may curtal ther revenue. The thrd varable, rato of losses pad to total assets (LTA) represent the ablty of nsurance frms to compensate rsk. Amount of losses that an nsurance company undertakes s uncertan. A hgher rato means the company needs better rsk management polces to guard aganst future possble loss payments. We expect a postve sgn for the coeffcent of ths varable snce hgher loss payments may be an ndcator of hgher possble ncome. 7 For testng equlbrum assumpton of the model, followng equaton s estmated: ln ROAt = α + β1 ln PLt + β2 lnpbst + β3 lnpfkt + β4 lntat + β5 lnetat + where ROA represents net ncome β6 lnltat + εt before tax to total assets. Other varables remaned unchanged as defned n Eq. (4) 5
In the notaton of Equaton (4), the H-statstc s gven by β 1 +β2 + β3. In contrast to most of the lterature, we do not rely on a smple cross-sectonal estmaton, but carry out a panel estmaton wth fxed effects. The fxed effects model has several advantages. Frst, by ncludng nsurance frm fxed effects, unobserved heterogenety can be controlled 8. All frm-specfc, non tme-varyng determnants of revenues not explctly addressed n the regresson specfcaton are captured by the fxed effects. Second, panel estmaton allows us to obtan more relable estmates by observng the behavor of frms over tme and testng for changes n the coeffcents. Snce the Turksh nsurance ndustry has experenced sgnfcant changes durng the sample perod (1996-2004) due to the lberalzaton and deregulaton, t s worthwhle to check whether compettve structure has changed over tme. To verfy whether the compettve structure has changed over tme, the sample perod s dvded nto three sub-perods and H-statstc for each sub-perod s computed by applyng model (4) to each sub-perods. 4. Data and Emprcal Results Data Our data set conssts of a sample of 38 non-lfe nsurance frms representng about 80 percent of the ndustry assets over the perod 1996-2004. We use a frm-level data. The annual balance sheets, techncal and fnancal ncome statements of the frms are obtaned from the Insurance Supervsory Board of Republc of Turkey Prme Mnstry Undersecretarat of Treasury. The nsurance frms, whch have non-postve general expenses, total premums, total assets, equty captal, personnel expenses and fxed assets are excluded from the sample. Hence we obtan an unbalanced sample. The number of frms n each year of our sample perod s reported n Table 2. Table 2. Number of frms n the sample Year Number of Frms 1996 19 1997 24 1998 23 1999 24 2000 28 2001 25 2002 19 2003 15 2004 13 Emprcal Results The estmates of the reduced-form revenue equaton and ndustry equlbrum test results as derved from a panel data set analyss are reported n Table 3 and 4 9. The regresson model n (4) s estmated usng the fxed effects model for the three sub-perods n order to control for unobserved heterogenety 10. The choce of the fxed effects over the 8 Ths s mportant because OLS regresson s based f a varable s omtted that s related to the dependent varable. 9 To correct standard errors, the Whte s (1980) heteroscedastcty consstent t-statstcs were used. 10 As dscussed n the prevous secton, panel data models allow us to obtan more relable estmates by observng the behavor of frms over tme and testng for changes n the coeffcents. The test s mplemented by dvdng the sample perod 1996 2004 nto three sub-perods (1996-1998, 1999-2001, 2002-2004) and nteractng the nput prce varables wth a dummy varable that takes the value of one n the second subperod when we compare frst two sub-perods and dummy varable takes the value of one n the thrd subperod when we compare last two sub-perods. If the nteracton term yelds sgnfcant estmates, they ndcate a structural break n the statstcal relatonshp between nput prces and revenues, whch means that 6
random effects estmators s based on the result of the Hausman test 11. As seen n Table 3 2 and 4, the R s take very hgh values, whch s a sgn of better ft. The Wald test that follows an F dstrbuton s used n the competton models to test whether the estmated H- statstcs are statstcally dfferent from zero and unty. For the perod 1996-1998, the H value of 0.034 s not statstcally dfferent from zero but dfferent from unty. Hence, we are unable to reject the monopoly or conjectural varatons short-run olgopoly hypotheses for the Turksh nsurance frms. All three cost elements, nput prces, are statstcally sgnfcant. The major contrbuton to the H-statstc comes from the labor costs. Among the frm-specfc control varables, the total assets, whch controls the sze has a postve sgn and statstcally sgnfcant. For the second sub-perod, the estmated value of the H- statstc (0.087) s postve and statstcally dfferent from unty but not zero. Therefore, as n the frst sub-perod, we are unable to reject the monopoly or conjectural varatons shortrun olgopoly hypotheses for the Turksh nsurance frms. All the coeffcents of nput prces are postvely related to frm revenue and however, only the coeffcent of prce of busness servces s statstcally sgnfcant. The frm-specfc control varables are statstcally sgnfcant. The frm sze and rato of losses pad to total assets are postvely related to total revenue, and the rato of equty captal to total assets s, however, negatvely related to total revenue. As for the last sub-perod, the results suggest that n the Turksh nsurance ndustry, market structure can be characterzed by monopolstc competton. The H value of 0.798 s statstcally dfferent from both zero and unty at the one-percent level. Ths results rejects the monopoly hypothess and perfect competton hypothess. Turksh nsurance frms revenues behave as f they were earned under monopolstc competton for the perod 2002-2004. Two coeffcents of the costs components are statstcally sgnfcant and are postvely related wth the total revenues. In case of the frm-specfc varables, only the coeffcent of total assets s statstcally sgnfcant. The sgn of TA s postve, ndcatng that sze-nduced dfferences among nsurance frms lead to hgher total revenue. Hence, the regresson results ndcate that total assets appears to capture the sze effect n the model. Celk and Kaplan (2007) also focus on the 2002-2004 perod and nvestgate the competton n the non-lfe Turksh nsurance ndustry usng two dfferent revenue varables; net premum ncome and total ncome (ncludng total premum ncome plus nterest and other ncome). They assume three nput prces namely, unt labor costs, unt fxed asset costs and commsson expendture per asset, and use captal-asset rato and total assets as control varables. They do not control for the heteroscedastcty n the regressons and do not perform the equlbrum test. Ther results suggest that frms n the Turksh nonlfe nsurance ndustry over the perod 2002-2004 generated ther revenues under monopoly condtons 12. Snce our varable defntons and conduct of the study s entrely dfferent, our fndngs are not comparable wth thers. the drecton of the sum of the elastctes (.e., H) changed. The test results ndcate that the nteracton terms are sgnfcant at the conventonal levels for the frst and second, and second and thrd sub-perods. Although the results are not reported they are avalable from the authors upon request. 11 Hausman and Taylor (1981) developed a test on the correlaton between ndvdual effects and regressors. The null hypothess s no correlaton between ndvdual effects and regressors. Random effects model assumes that ndvdual effects and regressors are uncorrelated. Hence, random effects model s more effcent than fxed effects model, whch s neffcent but consstent. 12 Snce they do not perform the equlbrum tests, ther results on the market structure of the Turksh non-lfe nsurance ndustry could be msleadng. 7
Table 3. Compettve Structure Test Results Varable 1996-1998 1999-2001 2002-2004 Dependent varable: Coeffcent Coeffcent ln TR Coeffcent ln PL 0.125* 0.024 0.035 0.820 0.744* 0.134 ln PBS -0.050*** 0.027 0.034*** 0.083 0.074* 0.013 ln PFK -0.041* 0.012 0.018 0.648-0.020 0.057 ln TA 0.712* 0.137 0.959* 0.099 0.496** 0.228 ln ETA -0.136 0.122-0.045** 0.0218 0.189 0.436 ln LTA 0.178 0.108 0.434* 0.047 0.169 0.176 Constant 1.655* 0.465 1.646*** 0.841 6.001* 0.675 2 R 0.972 0.981 0.983 N 66 77 47 H 0.034 0.087 0.798 F0 0.097 0.204 4.376* F1 97.960* 37.476* 11.396* Note: TR, PL, PBS, PFK, TA, ETA, and LTA represent total revenue, the unt prce of labor, unt prce of busness servces, and unt prce of fnancal captal, total assets, the rato of equty captal to total assets, and rato of losses pad to total assets, respectvely. *,**,*** denote sgnfcance level at the 1%, 5%, and 10%, respectvely. H = the sum of elastctes of reduced form revenue functon wth respect to nput prces. F0 = F-statstc for testng the hypothess H = 0. F1 = F-statstc for testng the hypothess H = 1. The estmates of the H-statstcs for the long-run equlbrum, whch use the return on assets (ROA) as the dependent varable, are reported n Table 4. The estmated values of H for the long-run equlbrum test are not statstcally dfferent from zero. Hence, the long-run equlbrum condton appears to be establshed n each of the sub-perods and therefore, the nterpretaton of H-statstcs above s meanngful. Table 4. Equlbrum Test Results Varable 1996-1998 1999-2001 2002-2004 Dependent Coeffcent Coeffcent Coeffcent varable: ln ROA ln PL -0.065 0.052 0.236*** 0.137-0.346* 0.073 ln PBS -0.054** 0.024-0.099* 0.011 0.009 0.046 ln PFK 0.143* 0.038 0.095* 0.029 0.220* 0.038 ln TA -0.665* 0.135-0.941* 0.140-0.488 0.299 ln ETA -0.448* 0.156-0.008 0.016-0.094 1.050 ln LTA -0.063 0.052-0.081** 0.034 0.087 0.152 Constant 0.194 0.707 2.604* 0.956-0.530* 0.058 2 R 0.967 0.959 0.918 N 66 77 47 H 0.024 0.231-0.116 F0 0.044 1.43 0.039 Note: TR, PL, PBS, PFK, TA, ETA, and LTA represent total revenue, the unt prce of labor, unt prce of busness servces, and unt prce of fnancal captal, total assets, the rato of equty captal to total assets, and rato of losses pad to total assets, respectvely. *,**,*** denote sgnfcance level at the 1%, 5%, and 10%, respectvely. H = the sum of elastctes of reduced form revenue functon wth respect to nput prces. F0 = F-statstc for testng the hypothess H = 0. One could argue that the Turksh nsurance frms could exercse market power due to the hgh concentraton and declne of the number of frms, mparng compettveness n the nsurance market n recent years. The emprcal fndngs of ths paper suggest that even though the Turksh nsurance ndustry s domnated by large frms, revenues were earned 8
as f the ndustry was monopolstcally compettve n recent years. Overall results show that H-statstcs have changed over the sample perod. 5. Concluson The Turksh nsurance ndustry has undergone sgnfcant changes due to the reregulaton and deregulaton over the past decades. Hence, ths process has consderable mplcatons for the structure of the ndustry and conduct of the frms. In ths paper the compettve condtons of the Turksh non-lfe nsurance ndustry have been examned for the perod 1996-2004, usng non-structural method developed by Rosse and Panzar (1977) and Panzar and Rosse (1987). Ths approach, prevously, has been appled manly to the bankng ndustres of the developed countres. However, the lterature focusng on the compettve structure of the nsurance ndustry s relatvely poor. Murat et al. (2002) assess the competton n the Australan general nsurance ndustry usng the Panzar and Rosse approach. Celk and Kaplan (2007) also follow the same approach and focus on the compettve structure of the Turksh nsurance ndustry for the perod 2002-2004. However, these studes do not analyze the compettve structure n an evolutonary perspectve. To the authors best knowledge, ths s the frst paper applyng the Panzar and Rosse approach to nvestgate market structure of the nsurance ndustry n an evolutonary perspectve. Moreover, exstng studes do not provde consstent varable defntons for the revenue equaton. Hence, we contrbute to the exstng lterature by defnng an nsurance ndustry specfc revenue functon. Our regresson results ndcate that, n the frst (1996-1998) and second (1999-2001) sub-perods, the nsurance frms operatng n the Turksh nsurance ndustry earned revenues under the monopoly or conjectural varatons short-run olgopoly. As for the thrd perod (2002-2004), the results ndcate that nsurance market was nether monopolstc nor perfectly compettve. Frm revenues behave as f they were earned under monopolstc competton. Overall, results suggest that market conduct n the Turksh nsurance market has changed durng the sample perod. Therefore, the recent ncrease n concentraton appears not to have had a sgnfcant mpact on the conduct of the frms operatng n the Turksh nsurance ndustry, snce the data reject the hypotheses of both monopolstc and olgopolstc behavor n the thrd sub-perod. We hope that our fndngs may reveal some nferences about the market structure and the conduct of the frms of smlar developng countres nsurance ndustres. References Ban, J. S. (1951). Relaton of proft rate to ndustry concentraton: Amercan manufacturng, 1936-1940. Quarterly Journal of Economcs 65: 293-324. Bkker, J. A. and K. Haaf (2002). Competton, concentraton and ther relatonshp: an emprcal analyss of the bankng ndustry. Journal of Bankng and Fnance 26: 2191-2214. Baumol, W.J. (1982). Contestable markets: an uprsng n the theory of ndustry structure. Amercan Economc Revew 72: 1-15. Bresnahan, T.F. (1982). The olgopoly soluton concept s dentfed. Economc Letters 10: 87-92. 9
Celk, T. and M. Kaplan (2007). Türk sgortacılık sektöründe rekabet: 2002-2004. Iktsat, Isletme ve Fnans 251: 50-62. Cummns, J. D., G. Donne, R. Gagre and A. Noura (2006). Effcency of nsurance frms wth endogenous rsk management and fnancal ntermedaton actvtes. Canada Research Char n Rsk Management, Workng Paper 06-06, Montreal: Canada. Cummns, J. D., M. Rubo-Msas and H. Z (2004). The effect of organzatonal structure on effcency: evdence from the Spansh nsurance ndustry. Journal of Bankng and Fnance 28: 3113-3150. Cummns, J. D. and M.A.Wess (1993). Measurng cost-effcency n property-lablty nsurance ndustry. Journal of Bankng and Fnance 17: 463-481. Cummns, J.D. and H. Z (1998). Comparson of fronter effcency methods: an applcaton to the U.S. lfe nsurance ndustry. Journal of Productvty Analyss 10: 131-152. Demsetz, H. (1973). Industry structure, market rvalry and publc polcy. Journal of Law and Economcs 16: 1-9. Hausman, J. A. and W.E. Taylor (1981). Panel data and unobservable ndvdual effects. Journal of Econometrcs 16: 155-155. Iwata, G. (1974). The measurement of conjectural varatons n olgopoly. Econometrca 42: 947-966. Murat, G., R.S. Tonkn and D.J. Jüttner (2002). Competton n the general nsurance ndustry. Zetschrft fur de Gesamte Verscherungswssenschaft 91: 453-481. Panzar, J. C. and J.N. Rosse (1987). Testng for monopoly equlbrum. Journal of Industral Economcs 35: 443-456. Rosse, J.N. and J.C. Panzar (1977). Chamberln versus Robnson: an emprcal test for monopoly rents, Studes n Industry Economcs, Research Paper no. 77, Stanford Unversty, Stanford: Calforna. Shaffer, S. (1982). A non-structural test for competton n fnancal markets. Bank Structure and Competton, Conference Proceedngs, Federal Reserve Bank of Chcago: 225-243. Smrlock, M. (1985). Evdence on the (non) relatonshp between concentraton and proftablty n bankng. Journal of Money, Credt and Bankng 17: 69-83. Vesala, J. (1995). Testng for competton n bankng: Behavoral evdence from Fnland, Bank of Fnland Studes, E1, Bank of Fnland. Whte, H. (1980). A heteroscedastcty-consstent covarance matrx estmator and a drect test for heteroscedastcty. Econometrca 48: 817-838. 10