QIS4 background document: Calibration of SCR and MCR. Calibration of the SCR: Calibration of the underwriting risk and market risk

Transcription

1 CEIOPS-DOC-02/ January 2008 QIS4 background document: Calbraton of SCR and MCR. Purpose of the document 1. The paper has been prepared by CEIOPS as a background document to the QIS4 techncal specfcatons that are currently under consultaton by the European Commsson. 2. The frst part gves an overvew of the calbraton of the SCR rsk modules n QIS4 compared to QIS3. The second part of the paper explans the MCR calbraton of the lnear approach that wll be tested n QIS4. Calbraton of the SCR: Calbraton of the underwrtng rsk and market rsk 3. In the short tme span between the publcaton of the QIS3 results and the draftng of the QIS4 specfcatons, attenton was pad to the ndustry's QIS3 feed-back regardng the calbraton of the SCR formula, but beng globally pleased wth the QIS3 calbraton, CEIOPS decded not to substantally challenge the QIS3 calbraton. As a consequence, the Aprl 2007 QIS calbraton paper 1 should stll be used as a reference calbraton paper for QIS4, but for transparency, we would lke to hghlght the areas where, often n answer to QIS3 feed-back, we have changed the calbraton. 4. The paper wll dscuss the dfferences n calbraton between QIS3 and QIS4 of followng rsk modules and sub-modules: Lfe underwrtng rsk: Lapse rsk, Lfe catastrophe rsk. Non-Lfe underwrtng rsk: Number of hstorc years, Lne of busness standard devatons, Geographcal dversfcaton, NL Cat Layer 1. Health underwrtng rsk: Accdent & health Short Term, Health Workers compensaton. Market Rsk: Concentraton rsk. Counterparty default: Counterparty default. 1 See CEIOPS webste: QIS3 Calbraton of the underwrtng rsk, market rsk and MCR, Aprl 2007 at and QIS3 Calbraton of the credt rsk at

2 Lfe underwrtng rsk Lapse rsk Reference n QIS3: I to I Reference n QIS 4: TS.XI.E 5. In QIS3, lapse rsk was covered n two sub-modules. In the lfe lapse sub-module, the rsk of msestmatng lapse rates or of a permanent change n these rates was analysed n two scenaros: (1) a 50% ncrease n rates and (2) a combned scenaro of an ncrease n absolute terms of 3% for polces where ths s an adverse event and a 50% reducton n rates for the remanng polces. Addtonally, n the lfe CAT module a mass lapse scenaro affectng 75% of those lnked polces where a lapse would cause a loss for the undertakng was assessed. 6. Some aspects of ths approach were crtcsed by QIS3 partcpants: The 75% shock n the CAT module was consdered to be too hgh. And partcpants dentfed a potental for double countng as the rsk of ncrease n lapse rates s covered by two sub-modules. Moreover, a smplfcaton of the approach was asked for. 7. In response to ths feedback, the followng changes were made: The 75% shock of the mass lapse scenaro was reduced to 30%. The new calbraton s an expert estmate based on past mass lapse events n the German lfe nsurance market. The scope of applcaton of the shock was extended from lnked polces to all polces. In order to avod double countng, only the more adverse of the mass lapse shock and the scenaro of permanent 50% ncrease n lapse rates s used to determne the captal charge. 2 The scenaro component of an ncrease n absolute terms of 3% was removed for reasons of smplfcaton. Lfe catastrophe rsk Reference n QIS3: I to I Reference n QIS 4: TS.XI.H 8. The QIS4 calbraton of the mortalty and dsablty catastrophe rsk s unchanged compared to QIS3. The captal charge s calculated as 1.5 of the captal at rsk. The calbraton s supported by a recent study of Swss 2 The ratonale of the calbraton of the 50% rsk factor can be found n the paper Calbraton of the Enhanced Captal Requrement for wth-proft lfe nsurers of the UK Fnancal Servces Authorty (

3 Rensurance Company. 3 Based on an epdemologcal model, for a pandemc wth a level of severty expected once every 200 years, the excess mortalty wthn an nsurance portfolo s estmated to be between 1 and 1.5 deaths per 1000 lves n most developed countres. Non-Lfe underwrtng rsk Number of hstorc years Reference n QIS3: I Reference n QIS 4: TS.XIII.B Thanks to the work done by workng groups on Best Estmate at natonal level and by the Groupe Consultatf, a recognton of a dfferentaton regardng the number of hstorc years between the varous non-lfe lnes of busness could be ntroduced n the QIS4 specfcaton. The table n TS.XIII.B.13 has been desgned usng the recommendatons from the "Work Group report on the Best Estmate n Non Lfe nsurance" (Publshed by "ACAM", November ), pages 15 to 29. Lne of busness standard devatons Reference n QIS3: I to I Reference n QIS 4: TS.XIII.B.23 and TS.XIII.B Followng feedback from QIS3, the factors used wthn the SCR non-lfe underwrtng rsk module were adjusted to better reflect the relatve and overall rskness of dfferent lnes of busness. Based on the QIS3 calbraton of premum rsk n the German market 4, the recalbraton reflects nformaton collected through QIS3 on nternal models, the results from current regulatory regmes and other market nformaton from several Members States (UK, PT, NL). Results from over (46) frms were used to recalbrate the factors. Geographcal dversfcaton New n QIS 4 Reference n QIS 4: TS.XIII.B.28 to TS.XIII.B.30 3 Cf GRL_Pandemc%20nfluenza.pdf. 4 Cf. QIS3 calbraton papers (Calbraton of the underwrtng rsk, market rsk and MCR), p

4 11.Durng the QIS 3 exercse, CEIOPS has receved comments suggestng that the proposed geographcal dversfcaton for groups, whch was based on the locaton of the enttes headquarters, was not enough rsk senstve. It has also been hghlghted that an entty operatng n dfferent countres wth branches or under the Freedom to Provde Servces regme should also beneft from geographcal dversfcaton. 12.Consequently, CEIOPS has revsed the structure of geographcal dversfcaton and extended t to solo enttes. In QIS 4, geographcal dversfcaton s calculated wth a dstrbuton ndex (Herfndahl) based on the locaton of rsks (for premums and reserves) for each lne of busness (except for credt and suretyshp and mscellaneous). 13.The dversfcaton beneft has been capped to 25% for the concerned lnes of busness. That cap seems to be reasonable for the standard formula consderng the lmted number of data gathered by CEIOPS on well dversfed groups durng the QIS 3 process. 14.Nevertheless, regardng geographcal dversfcaton, CEIOPS thnkng s stll under dscusson and at an early stage. Therefore CEIOPS nvtes comments on the sutablty and practcablty of the proposed approach. NL Cat Layer 1 New n QIS 4 Reference n QIS 4: TS.XIII.C.3 to TS.XIII.C.6 15.Followng QIS3 feedback, many frms expressed the vew that the methodology for calculatng the NL CAT module produced results that were nconsstent wth ther own assessment of rsk. The layer one formulae were confgured on the bass of QIS3 returns and benchmarked aganst market practce for a range of more than 20 nsurers under the UK FSA supervson to ensure a reasonable calbraton. Health underwrtng rsk 16.Insurers n varous countres wrtng "Accdent & health" and "Workers compensaton" types of busness found ther actvtes dffcult to ft wth the QIS3 splt of the non-lfe underwrtng module. For ths reason, CEIOPS has restructured the Health module for QIS4. Due to the restructurng of the Health module, we have ntroduced a 0.25 correlaton factor between the SCR NL and the SCR Health n TS.VIII.C.4. Accdent & health Short Term New n QIS 4 Reference n QIS 4: TS.XII.F - 4 -

5 17.Calbraton for the new "Accdent & health Short Term" sub-module s dentcal to the calbraton used for Lne of busness 2 and Lne of busness 3 of the QIS3 non-lfe underwrtng rsk specfcaton (I to I.3.251). Health Workers compensaton New n QIS 4 Reference n QIS 4: TS.XII.G 18.Market-wde factor for premum rsk The calbraton was based on the analyss of hstorcal data from Portugal. It s based on a smlar approach to that used for the premum rsk n other Non-lfe LoB's,.e. t s based on the observaton of the volatlty of hstorc loss ratos. These hstorc loss ratos reflect the volatlty of clams fallng nto the standard non-lfe type of labltes category, as well as that of annutes and lfe assstance labltes. 19.Market-wde factor for reserve rsk The calbraton was based on the analyss of hstorcal data from Portugal. The factor was derved from the observaton of the mpact of applyng a stress test to the development pattern of the run-off trangle (clams pad), correspondng to a VaR 99,5% 1-year scenaro. The data used comprses only the standard non-lfe type of labltes. 20.Longevty rsk No specfc analyss was made. The same shock was assumed as for the lfe u/w rsk module. 21.Revson rsk The calbraton procedure s detaled on the QIS3 calbraton paper. The only addton was the use of a more granular approach, to derve separate factors for annutes and lfe assstance labltes. 22.Expense rsk No specfc analyss was made. The same shock was assumed as for the lfe u/w rsk module. Market Rsk Concentraton rsk Reference n QIS3: I to I Reference n QIS 4: TS.IX.G - 5 -

6 23.The quadratc formula used n QIS3 s replaced by a lnear formula n QIS4. Although the man prncples of calbraton reman the same, the updated calbraton paper s ncluded as an annex to the present paper. Counterparty default Reference n QIS3: I to I Reference n QIS4: TS.X 24.The counterparty default rsk module follows a factor based approach. In QIS3, the volume measure for ths approach was termed replacement cost. The techncal specfcatons gave no detaled defnton of ths term. Ths was consdered to be a shortcomng by many QIS3 partcpants. Therefore, the QIS4 Techncal Specfcatons try to provde an explct nstructon on how to estmate the volume measure. In order to algn wth common termnology, the name of the measure was changed to loss gven default (LGD). 25.Usually, the current exposure to the counterparty s not an adequate measure of the loss gven default, because the exposure may change over tme and a default s much more lkely f the exposure s at peak level. In order to assess the potental loss, the SCR standard formula calculatons for the underwrtng and market modules are used (for rensurance and dervatves respectvely). These calculatons try to answer the followng queston: what s the exposure f the counterparty fals n a one n 200 year event that underles the SCR calculaton? Ths exposure s determned as the sum of the current exposure and the rsk mtgatng effect of the rensurance or dervatve that s allowed for n the underwrtng or market rsk module. Ths rsk mtgatng effect s the dfference between the gross SCR (whch s calculated under the assumpton that the rensurance or dervatve s not n place) and the current SCR of the module whch allows for the rsk mtgatng effect. 26.In case of default, typcally a part of the exposure can stll be collected. In order to allow for ths, the above descrbed exposure s multpled by a factor of 50% to arrve at the loss gven default. The factor s an estmate based on the analyss of loss gven default n the followng studes: Standard & Poor s, Annual 2005 Global Corporate Default Study And Ratng Transtons, January 2006, and Ftch Ratng, Prsm: Favorable Market Feedback and Clarfyng Responses Part 1, September

7 Annex: Calbraton of the Concentraton Rsk Module Descrpton 27.Market rsk concentratons present an addtonal rsk to an nsurer because of: addtonal volatlty that exsts n concentrated asset portfolos; and the addtonal rsk of partal or total permanent losses of value due to the default of an ssuer. Input 28.For the sake of smplcty and consstency, the defnton of market rsk concentratons s restrcted to the rsk regardng the accumulaton of exposures wth the same counterparty. It does not nclude other types of concentratons (e.g. geographcal area, ndustry sector etc.). 29.Rsk exposures n assets need to be grouped accordng to the counterpartes nvolved. E = Net exposure at default to counterparty Assets xl = Amount of total assets excludng those where the polcyholder bears the nvestment rsk ratng = external ratng of the counterparty Output 30.The module delvers the followng output: Mkt conc = Captal charge for market concentraton rsk Calculaton 31.The calculaton s performed n three steps: (a) excess exposure, (b) rsk concentraton charge per name, (c) aggregaton. (a) The excess exposure s calculated as: XS E = max 0 CT, Assetsxl ; where the concentraton threshold CT, dependng on the ratng of counterparty, s set as follows: - 7 -

8 ratng CT AA-AAA 5% A 5% BBB 3% BB or lower 3% (b) The rsk concentraton charge per name s calculated as: Conc = Assets xl XS g where XS s expressed wth reference to the unt (.e. an excess of exposure above the threshold of 8%, delvers XS = 0.08) and the parameter g, dependng on the credt ratng of the counterparty, s determned as follows: ratng Credt Qualty Step G AAA AA A BBB BB or lower, unrated 4 6, (c) The total captal requrement for market rsk concentratons s determned assumng ndependence between the requrements for each counterparty : Mkt conc = Conc 2 32.The general goal of ths exercse s to provde a workable evdence of the mpact that a concentraton n a sngle counterparty may have n the rsk profle of a well-dversfed portfolo of assets. 33.The methodology appled for ths purpose was crculated n advance wthn CEIOPS Fnancal Stablty Commttee (prevously responsble for QIS3), and the comments receved were consdered to mprove the method of calbraton. Ths method may be descrbed as follows: - 8 -

9 1 st. step.- The startng pont s the desgn of a well-dversfed portfolo of nvestments n ndvdual names wth the followng characterstcs: a) The portfolo has a mx, representatve of EU average nsurers portfolos of nvestments n bonds and equtes. The mx proposed s 70% - 30% correspondng bonds equtes respectvely (see fgure 11A, page 12, Fnancal Stablty Report Conglomerates ). b) Wthn each of these two groups, a sector-dstrbuton of nvestments s bult, also accordng to an EU expected average, as follows: a. Investment n bonds: We have assumed that 25 % of bondsportfolo s nvested n rsk-free bonds, and the rest (75 %) s nvested n dfferent sectors and ratngs as descrbed below. b. Investment n equtes: To the extent that ths exercse assumes as startng pont a well-dversfed portfolo, consequently t should replcate some equty ndex suffcently representatve and well-known. The selected ndex s Eurostoxx 50, and the perod used to record data on prces of each of ts element, ranges from 1993-january-11 st untl november-30 th. The length of ths perod guarantees suffcent hstorcal data to derve VaR 99.5% wth a hgh degree of relablty. Some elements of the selected ndex have been removed, snce ther records of data prces are only avalable for a sgnfcantly shorter perod than that above mentoned 5. Descrpton of bonds-portfolo 34.In order to avod the effect of the change n Macaulay Duraton (as the lfe of the bond expres) and the renewal of the nvestment 6, and what s more mportant, to reflect the whole rsk belongng to each sector/ratng t was decded: 1) Bonds used n the computaton are notonal bonds, all of them ssued at 5% rate and pendng 5 years to maturty. At any moment of the smulaton each bond mantans these features (whch could be accepted as representatve average features of the bonds exstng n nsurance portfolos) 5 As part of the ntal steps of calbraton exercse of concentraton rsk, a complete set of tentatve checkngtests was carred out to optmze the desgn of the method. The outputs of these prelmnary calculatons may be summarzed as follows: - Dealng wth concentraton rsk requres obvously the use, as startng pont, of a suffcently hgh number of exposures, - Nevertheless, as mportant as the number of dfferent exposures s to guarantee that the selected names reflect a varety of behavours suffcently dsperse, n such a way that almost all exstng and possble equtes/bonds fall n the range of behavours consdered - Under the above assumpton, ncreasng the number of names dd not have a sgnfcant added value (the outputs were rather smlar), whle the computatonal burden ncreased and the analyss of a hgher number of names became less transparent. 6 Ths could be seen as beng qute arbtrary, because we should have to select agan another smlar bond to substtute the prevous one

10 2) To capture and summarze market nformaton about each sector/ratng, notonal bonds descrbed n pont 1) are valued wth Bloomberg corporate yeld curves, accordng the correspondng sector/ratng. The followng table lsts these yeld curves: INTEREST RATES DATA 1 F888 EUR BANK AAA 2 F462 INDS AA+ 3 F890 BANK AA 4 F580 UTIL AA 5 F892 BANK A 6 F583 UTIL A 7 F465 INDUS A 8 F898 BANK BBB 9 F625 TELEF A 10 F468 INDUS BBB 11 F469 INDUS BBB- 12 F682 TELEF BBB+ 13 F470 INDUS BB Descrpton of equtes portfolo 35.To obtan a well-dversfed portfolo, after selectng the components of Eurostoxx 50 mentoned above, other addtonal names have been added to complete all the buckets of the cross-table resultng from, on one dmenson ratng categores consdered, and on the other dmenson economc sectors ncluded n ths exercse. 36.Weghts for each name n ths ntal portfolo depend on the followng features: 1) When calculatng BB concentraton rsk polynomal: we use names ranged from B to AAA; 2) When calculatng BBB, A and AA-AAA concentraton rsk polynomals: we use the names ranged from BBB to AAA wth the relevant adjustment n ther ntal weght; 3) Besdes, the level or qualty of dversfcaton of these two startng portfolos has beng checked by calculatng ther Herfndal ndex and comparng wth ther mnmum possble value: Herfndal _ Index = n w =

11 37.Ths table shows n all cases a Herfndal ndex for each portfolo only hgher than the mnmum possble value, whch confrms that the selected portfolos are actually well-dversfed. 38.Fnally, the calbraton exercse has calculated the hstorc 1-year VaR 99.5% of a mxed portfolo (30 % nvested n the equtes portfolo, and 70 % nvested n the bonds portfolo). Ths measure s calculated twce: Frstly, takng nto account all the names and ts correspondng yeld curves as lsted above: VaR (99.5 %) = % Secondly, excludng BB names and ts correspondng yeld curve, as lsted above. VaR (99.5 %) = % In both cases, rsk-free bonds are prced wth the German soveregn curve. 39.As one can apprecate, there s suffcent ratonale to calbrate frstly BB polynomal usng the whole portfolo and afterwards, n a second step, to calbrate BBB, A and AA-AAA polynomal wth a less volatle portfolo. 7 7 One has to bear n mnd that due ther hgh volatlty, consderng BB curve and BB-B equtes ncreases (n relatve terms) the goodness of the rest of names/ratngs

12 2 nd step- Concentratng exposures n the ntal portfolo: 40.Frst of all, we have establshed a bjectve correspondence between each equty name and one of the nterest rates curves above lsted, takng nto account ts sector / ratng. Ths means that when we concentrate the whole portfolo we concentrate at the same tme the nvestment n the selected equty and ts correspondent notonal bond. (1) The exercse begns by selectng a concrete name wth a certan ratng, (.e. a bank rated AAA) and ts correspondent notonal bond (Banks AAA). Then, we ncrease n steps of 1 per cent ts total weght n respect of the whole portfolo, obvously reducng smultaneously the partcpaton of the rest of counterpartes (to solate purely the effect of concentraton on the selected name). 8 (2) Increases of concentraton levels range from the startng weght up to the startng weght plus 50%, (as above mentoned, usng 1% steps). For each level of concentraton, we calculate the dfference between the hstorc 1-year VaR 99.5% of the startng portfolo and hstorc 1-year VaR 99.5% of the resultng concentrated portfolo, and ths dfference s consdered a raw proxy of an eventual concentraton charge (t s called Varaton VaR.) (3) Ponts of raw-concentratons charges obtaned n the successve ncreases of concentraton for each name are drawn, nterpolatng a straght lne, and then dervng the parameter g. (4) Thus, for each level of ratng we wll have: Conc = Assets xl XS g 3 rd step.- The same procedure s repeated for names rated AA, A, BBB and BB or worse and dfferent sectors. 41.Note that the ntal nvestment n rsk-free bonds remans unchanged. Therefore concentraton exercse refers to the whole equty portfolo and 75% of the bonds-portfolo. 42.Tables below compare 1-year hstorcal VaR 99 5% for the startng portfolo versus the extreme 1-year hstorcal VaR 99 5% (portfolos wth a concentraton ncrease of 71 % above the ntal weght). See Table 2 for 8 Note that equtes and bonds are smply added, wthout applyng the weghts contaned n par of CEIOPS Consultaton Paper 20. Ths mnor and techncal change s proposed for varous reasons, presented for approval durng CEIOPS Pllar I WG meetng held n January 10, and havng obtaned the necessary agreement

13 calculatons ncludng BB & B exposures and Table 3 for calculatons excludng such exposures. Table 2. Calculatons ncludng BB & B exposures Those who Those who mprove decrease AAA+AA A BBB WORSE Mean 2,3784% -10,7817% -0,70% -1,44% -6,15% -26,00% Standard Dev 1,1290% 12,6649% 6,1265% 5,9730% 9,1125% 20,1174% N Varaton Coef 47,4674% -117,4670% 870,2377% 414,3150% 148,2736% 77,3748% Table 3. Calculatons excludng BB & B exposures Those who Those who mprove decrease AAA+AA A BBB Mean 1,6138% -6,8795% -2,01% -3,25% -7,75% Standard Dev 1,4912% 7,4379% 6,3729% 5,8990% 9,1288% N Varaton Coef 92,3996% -108,1159% 316,6407% 181,2757% 117,7904% 43.Once ths pont has been reached and the graphs obtaned have been analysed, the nterpolaton of a straght lne s carred out takng nto account the worst-behaved names are. Ths crteron s necessary to guarantee the consstency of the calbraton exercse wth the ratonale groundng the standard SCR formula, whch focus on stressed scenaros. 9 9 Due to ts own characterstcs, the mean VaR for each group of ratng (BB, BBB, A and AA-AAA), tends to smooth the rsk of concentraton, thus understatng the correspondng captal charge

14 44.See n fgure 1 the selected lnes and the nterpolated one for AA-AAA ratng (the last one). Each lne means the Varaton VaR when the portfolo ncreases the concentraton n each equty and ts correspondng AA bond data1 data2 data3 data Fgure 1 45.Fgure 2 plots the selected lnes and the nterpolated one for A ratng, wth smlar meanng and methodology as the prevous graph data1 data2 data Fgure

15 46.Fgure 3 depcts the selected lnes and the nterpolated one for BBB ratng, followng the same ratonale and presentaton as above data1 data2 data Fgure 3 47.Fgure 4 contans the lnes and the nterpolated one for BB or worse ratng Fgure 4 - Fnally, g parameters for each ratng are estmated usng a conventonal mnmum squares method. where Fnal result 48.Concentraton rsk model for each group of ratng : Conc = Assets * XS * g XS = Excess exposure at each group of ratng

16 XS Exposure _ group _ ratng = max 0; Concentraton _ Threshold _ group _ ratng Assetsxl Assets * XS = excess of exposure above the threshold, expressed n unts nstead of percentage g * XS = the captal charge obtaned as result of the calbraton exercse 49.As one can see, the formula has been calbrated for dfferent thresholds dependng on each group of ratng. These thresholds are lsted n the followng table: Group of ratng Threshold AA-AAA 0.05 A 0.05 BBB 0.03 BB-worse-unrated The exstence of dfferent thresholds grounds on the fact that captal charges obtaned accordng the calbrated parameters for buckets AAA-AA and A are not materal for concentratons between 3-5%. 51.The fnal coeffcents for each group of ratng are the followng ones: ratng Credt Qualty Step G AAA AA A BBB BB or lower, unrated 4 6,

17 Calbraton of the MCR Introducton 1. In January 2008, CEIOPS undertook an effort to test the calbraton of the MCR n the QIS4 Techncal Specfcatons on several countres QIS3 data. Ths secton gves a bref descrpton of the calbraton ratonale and the results of ths testng exercse. 2. Followng Artcle 126(1) of the framework Drectve proposal, the MCR should be calbrated to a confdence level n the range of 80% to 90% Value-at-Rsk over a one-year perod. In developng and testng the calbraton for QIS4, CEIOPS used a percentage of the SCR as a proxy calbraton target. It s recognsed that there s no lnear relatonshp between 80% or 90% VaR, and 99.5% VaR through all dstrbutons. In CEIOPS calbraton exercse, followng the lognormal assumptons underlyng the SCR standard formula, the 25% SCR to 45% SCR nterval was taken as a rough equvalent of the 80% to 90% VaR range, and the mdpont of ths nterval.e. 35% of the SCR was used as a proxy calbraton target. Non-lfe busness Analyss: back-testng the calbraton proposed n the draft QIS4 specfcatons 3. The non-lfe MCR premum and techncal provsons factors n TS.XV.C.4 have been derved from the same underlyng lognormal assumptons as n the SCR standard formula for non-lfe premum and reserve rsk, respectvely. Factors correspondng to 90% VaR over a one-year tme horzon (.e. the hgh end of the 80% 90% target nterval of the framework Drectve proposal) were chosen to mplctly compensate for the fact that ths calbraton approach does not take nto account rsks other than premum and reserve rsk. 4. The α h factor n TS.XV.D.5 for long-term health nsurance provsons was calbrated to 35% of the observed QIS3 SCR to techncal provsons rato on one local market. The α a factor was calbrated usng non-lfe annuty data on a local market, reflectng the mddle pont between the 80% VaR and the 90% VaR calbraton (yeldng a and a factor respectvely on that local market). 5. In January 2008, CEIOPS back-tested ths calbraton on QIS3 data. The testng took nto account the QIS4 changes n SCR premum and reservng rsk factors. For composte frms, a proxy non-lfe SCR was calculated to allow a separate non-lfe MCR to SCR comparson. 6. Testng for 460 nsurers n 19 countres resulted n the followng non-lfe MCR to SCR ratos:

18 MCR to SCR rato (non-lfe) number of frms lower than 10% 20 10% to 20% 60 20% to 30% % to 40% % to 50% 88 50% to 60% 20 60% to 70% 9 70% to 80% 2 80% to 90% 8 90% to 100% 0 hgher than 100% 2 7. One sgnfcant outler group wth hgh MCR to SCR ratos that has been dentfed n the testng are health nsurers on a local market where a market-wde mandatory equalsaton system s n place. All but one other non-lfe MCR to SCR ratos observed n the testng were lower than 70%, wth three-quarters of the results fallng between 20% and 50%. 8. Gven these results, CEIOPS observes that the factors proposed for non-lfe busness generally provde a satsfactory nterplay between the SCR and the MCR. Lfe busness Analyss: back-testng the calbraton proposed n the draft QIS4 specfcatons 9. The calbraton presented n the draft QIS4 specfcatons n December 2007 was derved va least squares lnear fttng for 35% of the SCR on the QIS3 data of one local market, takng nto account the followng adjustments: the counterparty default rsk was removed from the SCR, as ths rsk component was concentrated n a small number of frms, and was dffcult to reproduce by a lnear formula; the lapse catastrophe component was removed from the SCR, gven the change of methodology n QIS4; the SCR was adjusted to exclude free assets, so that the calbraton of the MCR reflect the fnancal poston of a company wth lttle to no free assets above the techncal provsons and the SCR. The ratonale for ths adjustment s that the MCR beng tested s unaffected by assets. 10. Where QIS3 data were nsuffcent to yeld a reasonable factor, expert adjustments were appled to the fttng results to obtan a calbraton. These ncluded the techncal provsons charge for the wth-proft death, dsablty

19 and survvorshp; unt-lnked death, dsablty and survvorshp; non-proft death, dsablty and savngs classes; and the captal-at-rsk charge for remanng contract term of less than 5 years. 11. In January 2008, CEIOPS back-tested ths calbraton on QIS3 data. The testng took nto account the QIS4 changes regardng lapse catastrophe rsk. For composte frms, a proxy lfe SCR was calculated to allow a separate lfe MCR to SCR comparson (t s noted however that, on some markets, compostes have dfferent lfe rsk profles than lfe-only frms, so the splttng of the SCR for compostes dd not always lead to comparable results). 12. A major ssue that emerged from the testng related to the dfferent rsk absorpton characterstcs of future proft sharng on dfferent markets. On those markets where future dscretonary benefts have a hgh rsk absorbency, there s a strong negatve correlaton between dscretonary bonus provsons and rsks, justfyng a negatve factor. On some of these markets, the ntal calbraton (wth a zero factor for dscretonary bonus provsons and a 2.5%-3.5% factor on provsons for guaranteed benefts) resulted n hgh MCR to SCR ratos n the testng. For one specfc market, a 26% factor for dscretonary bonus provsons has been suggested nstead, whle a 6.8% factor would apply to provsons for guaranteed benefts (the latter factor reflectng the rsks of a frm that has no dscetonary bonus provsons to absorb losses). 13. On other markets however the relatonshp between future dscretonary bonuses and rsk mtgaton s less straghtforward. It has been rased that future dscretonary bonuses may actually have a hgher rsk profle (e.g. through rsker nvestments) on some markets. On such markets the factors suggested above could lead to negatve MCR results (stopped only by the absolute floor). Refnement of the ntal approach and second round of backtestng 14. CEIOPS therefore tred to refne the ntal approach n the followng way (see paragraph TS.XV.E.3 4): MCR Lfe where = max { α TP + α TP ; γ TP } WPg WPg WPb WPb WPg WPg + { non WP} α TP + j β CAR TP WPg = techncal provsons (net best estmate) for guaranteed benefts relatng to wth-profts contracts TP WPb = techncal provsons (net best estmate) for dscretonary bonuses relatng to wth-profts contracts j j

20 and where the captal charge on techncal provsons other than wth-profts and on captal at rsk s unchanged from the QIS4 Techncal Specfcatons, and where the new alpha and gamma factors are the followng: 1st level segment sub-segment factor α factor γ (wth-proft floor) WP g guaranteed benefts wth-proft WP b dscretonary bonuses Ths refned approach for the wth-profts segment was suggested as a mddle ground between the two types of market dentfed above. It recognses future proft sharng as a rsk mtgatng factor, however t also ncludes a floor equal to 1.5% of techncal provsons for guaranteed benefts to avod extremely low results. Thus the captal charge should reman n a band between 1.5% and 3.5% of the guaranteed part of provsons. 16. Then a second round of back-testng (ncludng QIS3 data of 286 frms n 18 countres, focusng on the refned approach for wth-profts contracts, led to the followng lfe MCR to SCR ratos: MCR to SCR rato (lfe) number of frms lower than 10% 33 10% to 20% 63 20% to 30% 62 30% to 40% 51 40% to 50% 26 50% to 60% 21 60% to 70% 12 70% to 80% 5 80% to 90% 6 90% to 100% 2 hgher than 100% 5 Please note that the above data do not nclude the lfe results of compostes n one market. The results of these undertakngs are heavly affected by ther accdent and health busnesses, and ncludng them n the summary table would ntroduce heterogenety and would dstort the readng of the testng results. 17. The two rounds of back-testng aganst the QIS3 results n varous Member States tend to show that the approach ntally proposed n draft QIS4 specfcatons for lfe busness could be further refned n order to take nto account the specfctes of wth-profts contracts