Risk Integrated

Transcription

1 3 July 2013 Enterprse Rsk Management and CRE Lendng Introducton Fve years after the worst of the fnancal crss, companes are movng from the hghly reactve patchng of ther rsk management nfrastructure to consderng the bgger pcture of enterprse rsk management. Enterprse rsk management (ERM) s the set of polces, controls and rsk measurement tools put n place so that an nsttuton can coherently and consstently manage ts rsks. Ths paper dscusses how ERM can be acheved whle stll gvng the commercal real estate lendng teams useful rsk tool that meet the specal requrements of commercal real estate (CRE) lendng. The mplementaton of ERM s dffcult because t means brngng together rsk management both vertcally up through the dfferent rsk measurement functons wthn one asset-class and horzontally across all asset-classes. Wthn one asset-class, vertcal ntegraton requres consstency for all the rsk measurement functons appled to that asset e.g., from structurng, prcng and approvals through to portfolo reportng, captal management and stress-testng. However, n most nsttutons the data and models are hghly fragmented, especally for CRE. For example, there may be fve dfferent rsk models n use: one each for underwrtng, prcng, gradng, economc captal and stress-testng. One consequence s that the dfferent models produce dfferent rsk estmates for the same asset. Another consequence s that the data entered nto the models s scattered across spreadsheets and dsparate systems, requrng multple rounds of data entry and causng nconsstent reportng. Horzontally, the challenge s to measure rsk consstently across the dfferent asset-classes. It s common for there to be a patchwork of models for gradng dfferent assets and typcally the models are not consstent. An example of nconsstency wthn CRE s that a loan to a property wth 49% offce ncome and 51% retal ncome may be rated very dfferently than a loan to a property wth 51% offce ncome and 49% retal ncome because the retal and offce models are separate. A more subtle set of horzontal problems s embedded n the detals of the underlyng rsk methodology for each model e.g., whether the ratng phlosophy s Pont-n-Tme vs. Throughthe-Cycle vs. Through-the-Lfe, or mult-year vs. sngle-year 1. These subtletes can have a 1 Models can be nconsstent as to whether they are attemptng to measure the rsk at a pont n tme (PIT), through the lfe (TTL) or through the cycle (TTC). They can also be nconsstent n the way they Rsk Integrated

2 sgnfcant effect on the rsk numbers reported and therefore a sgnfcant effect on prcng, provsons and captal. One of the greatest horzontal challenges s estmatng the loss correlatons across assetclasses, e.g., the correlaton between loss on a CRE deal and loss n retal lendng. These correlatons are mportant for estmatng the requred economc captal, for stress-testng, and, n some cases, for prcng. Often the estmaton of correlatons s done n yet another separate model, but as dscussed later t s also possble to ntegrate loss correlaton wth the rest of the rsk analyss. The Specal Challenges Posed for ERM by CRE Before we go-on to look at solutons to these challenges, let us frst consder why t s especally dffcult to mplement ERM for CRE lendng. Unlke lendng to retal or plan corporate customers, CRE lendng s long-term and there s a hgh degree of structure n the underlyng asset. It s also greatly affected by the overall market condtons and there s a strong lnkage between the cause of default and the degree of subsequent loss. From an ERM perspectve ths has two mportant consequences. One s that the long-term rsk profle s very unlke that of the shorter term, less structured asset-classes: a typcal CRE default profle, as shown below, s hghly structured and has spkes of rsk correspondng to events n the deal such as lease expratons, changes n the nterest rate from fxed to floatng and refnancng rsk. Each of these spkes has a dfferent correlaton wth the market and wth the other assets. The other consequence of CRE s long-term structured nature s that the scorecard approaches frequently used to assess the rsk for retal and C&I loans do not work well for the mult-year complexty of CRE deals and therefore an approach such as cashflow smulaton s requred to capture the nterdependences. The challenge then s to use cashflow smulaton n such a way that t s consstent wth the models used for other classes and does not volate the ntegrty of the overall ERM framework. look at mult-year rsks when brngng them back to equvalent one-year rsks, and they can smply be nconsstent n the way they map the grade to the probablty of default (PD), loss gven default (LGD) and/or expected loss (EL). Rsk Integrated

3 Example of the Structured Rsk Profle for a CRE Loan Probablty Of Default 7.0% 6.0% 5.0% 4.0% 3.0% 2.0% 1.0% 0.0% '13 '14 '15 '16 '17 '18 '19 '20 '21 '22 '23 '24 '25 '26 '27 '28 '29 '30 '31 '32 Tenant Default Rsk Interest Rate Rsk Lease Rsk Refnancng Rsk Year Achevng Vertcal Integraton For vertcal ntegraton the deal s that for a gven asset, the rsk results should be the same whether t s beng assessed for underwrtng, prcng, or as part of stress-testng 2. For vertcal ntegraton one of the man problems s that the models typcally used are not suffcently flexble for all the requred tasks, so that nstead of there beng one consstent model for all tasks, there s one model per task. Ths s especally the case for CRE and a typcal stuaton n CRE lendng s descrbed n Appendx 1. Rsk Integrated s soluton for vertcal ntegraton s to ensure that there s a sngle set of models and central data-base for all stages of analyss. There are three key elements to the approach: 1. Use a model whch s comprehensve and has suffcent flexblty and detal to provde all of the followng: o the cashflows desred by underwrters o the rsk grades for PD and LGD 2 the base case used for gradng and the base case used for stress reportng should be the same but of course the results condtonal on a stress should be dfferent from the base-case gradng results. Rsk Integrated

4 o o o o o the regulatory captal the loan-loss provsons the rsk-theoretcal prce the mult-year loss profle requred for stress-testng the loss correlatons used for economc captal Ths model s bult to allow users to enter dfferent levels of deal nformaton dependng on the data avalable and the purpose. Fully detaled nformaton of s entered for deal underwrtng, when the lendng offcer has full nformaton and there s the power to accept, reject, restructure or re-prce the deal. Lmted data s entered when assessng new portfolos or for ntal deal structurng. Ths flexblty allows a sngle model to be used for all functons and allows data to be added ncrementally as t becomes avalable. Although the use of the model s relatvely straghtforward for any sngle deal, behnd the scenes the underlyng model needs to be relatvely sophstcated so that t s suffcently flexble to cater for all deal types and to cater for all contngences such as tenant defaults. In order to gve the bank fath n the sophstcated underlyng model, t s fully exposed to ther analysts to understand and change f desred. 2. The second key s to have a centralzed set of data, accessed wth web-technology, and wth maxmum flexblty for addng to the data structure as requred 3. The sources must nclude data typed nto web-forms, lease nformaton uploaded from spreadsheets, and loan nformaton updated automatcally from the central bankng systems. Importantly, because the model s also used for generatng the cashflows durng underwrtng, the lenders have a strong ncentve to type-n accurate detaled data because the results wll be shown as part of the credt revew. The consequence s that there s a central set of detaled, accurate, and up-to-date nformaton for generatng portfolo reports for senor managers, regulators and nvestors. 3. The thrd key pece of functonalty needed for vertcal ntegraton s the ablty for a portfolo manager to run a batch of deals (e.g., all lve deals) and re-grade them usng the latest models, forecasts and data wthout requrng the lendng teams to manually re-grade ther assets. Ths functonalty means that the portfolo manager has use of the full granularty that s avalable for assessng ndvdual deals and same rch dataset. 3 For ths we use an XML data structure n an open SQL database wth blendng to collate multple datasources. XML also allows new nformaton to be captured quckly, e.g., for new markets or custom data felds, wthout the need to refactor a database. Rsk Integrated

5 These prncples underpn the dataflow embedded n Rsk Integrated s Specalzed Fnance System, as llustrated below. Dataflow, Model Management and Reportng n the Specalzed Fnance System Lve Data Busness Management Raw MIS Regulatory Reports Credt Analysts Deal Data Portfolo Manager Scenaros Deal Reports SFS SQL Database SFS Servers Portfolo Reports Scenaro CFM Result Market Stats Rsk Analysts Models Wthn the framework of the SFS there are four man user groups: the rsk analysts, credt analysts, portfolo managers and busness managers: The rsk analysts control the models, the rsk parameters and the forecasts. Once they approve and load a model nto the system t becomes avalable to the other users. The other users can be granted dfferent levels of access to the system. Typcally the portfolo analyst has access to all models and settngs but the credt analysts can use only one model as specfed by the rsk analysts. The credt analysts use the system to assess ndvdual deals. They type deal-data nto web screens, upload t from other systems and/or upload t from templates, e.g., spreadsheets. Ths data s stored n the SQL database and then used when the credt analyst wants to generate the cashflows and rsk statstcs. Once the credt analyst has tred dfferent fnancng combnatons and s satsfed wth the rsk/return profle, he or she can prnt out a credt applcaton form contanng all the fnancal and rsk projectons for the deal. Ths same nformaton s also stored n the database for use by the portfolo manager and busness managers. Rsk Integrated

6 The portfolo manager can select sets of deals to be run n a batch, e.g., all lve deals, or all deals flagged as beng n the ppelne, or all deals wth a partcular tenant. That batch can be run aganst any model and set of forecasts, ncludng stressed forecasts. Ths means that the portfolo manager can easly and quckly run stress-tests and whatf analyses wthout burdenng the credt teams. The busness managers can of course get the same loan-level and portfolo-level reports beng generated by the credt teams and portfolo managers, but they can also get detaled management nformaton (MI) reports usng the raw data that was captured durng the underwrtng process. For example they can see the total balance lent to each sector and geography, they can see the LTV of the deals orgnated by each team, and they can see the dstrbuton of ext yelds for the loans maturng n one, two or three years. Ths data can be presented wth mnmal statstcal analyss 4, but because t s detaled and relable, t can strongly nform the ntuton of the busnesses managers about the level of rsk n the portfolo. The prncples descrbed above can be used to acheve vertcal ntegraton for any asset-class, but they are especally mportant for CRE where there s so much nformaton that has a meanngful mpact on the understandng of the rsk profle. Achevng Horzontal Integraton For horzontal ntegraton across the dfferent asset-classes we need to address the followng ssues: No jump n grade when an asset s categorzed as one type vs. another Consstency n ratng phlosophy: e.g., PIT vs. TTC Consstency n mappng mult-year rsks Meanngful calculaton of correlatons Prcng Avodng Dscontnutes n Grades Jumps n grades can be avoded by desgnng the models to be contnuous and avodng strct categorzaton so far as possble. For example rather than classfyng the asset as beng n a partcular sector and geography, recognze the degree to whch t s related to each sector and geography and the extent to whch t s, for example, under-constructon, beng rented-up or stablzed. Ths contnuty s natural n a full cashflow smulaton model because each property 4 For example graphs of LTV requre smply dvdng the current balance by the current value, whereas estmatng ext yelds requres more complex cashflow projecton, and rsk estmates requre full statstcal analyss. Rsk Integrated

7 and lease responds to ts own market nfluences, e.g., all the retal unts respond to changes n the retal rental market and offce unts respond to the offce rental market. Smlarly, large leases naturally have a larger mpact than small leases so t s not necessary to have a category of anchor tenant, nstead we just use the amount of rent for each tenant and larger tenants naturally have a larger mpact on the results. Ensurng Consstent Ratng Phlosophes Ensurng consstency n ratng phlosophy between dfferent models can be dffcult, especally snce there s often nconsstency wthn any ndvdual phlosophy. Cashflow smulaton models naturally behave as mult-year Pont-n-Tme models. The results from cashflow smulaton can easly be converted to Through-the-Cycle grades, but because cashflow smulaton naturally takes nto account the whether the market s currently hgh or low, the results are much more stable than typcal Through-the-Cycle scorecards 5. Mappng Mult-year Rsk Profles to a Sngle Grade The mappng of mult-year rsks becomes an ssue when lookng at CRE deals because CRE deals have an rregular rsk profle over ther lves, whereas C&I loans are assumed to have a smooth ncrease n rsk over tme. For conventonal C&I loans, the rsk n year-one completely defnes the assumed rsk for all subsequent years, therefore t has become customary for purposes lke Basel captal to expect that a sngle number wll be used to represent the grade. To create a sngle grade for CRE there must be a way to average the fundamental mult-year rsk profle nto a sngle number. There are multple ways of dong ths averagng, and t prmarly depends on the purpose of the fnal grade, but the fnal sngle number grade should be drectly comparable wth the grade for vanlla C&I loans 6. However, f the models and data are vertcally ntegrated as dscussed earler, the sngle grade becomes less mportant because mult-year rsks such as stress losses can be estmated drectly rather than beng projected from the sngle one-year grade numbers for the probablty of default and loss gven default. Estmatng Correlatons The estmaton of loss correlatons typcally uses two approaches: top-down and bottom-up. Top-down calculaton looks at the hstory of losses on overall portfolos and estmates how the losses moved together. Ths has the great advantage of gvng drect emprcal evdence, but 5 For example most Through the Cycle (TTC) scorecards keep the models weghts fxed over tme, but the outputs vary wdely because the nputs vary wdely. Consder a CRE scorecard that assgned PD based smply on the Loan-to-Value (LTV). At the top of the market the scorecard would assgn a relatvely low PD to a loan wth a 65% LTV. At the bottom of the market the same loan mght now have a 95% LTV and now be assgned a very hgh PD. The result s that typcal TTC models produce very unstable results and assgn low rsk, captal and reserves at the top of the market and hgh captal after the fall. 6 For example a CRE loan graded as a 5 mght be defned as havng the same loss rate over ts lfe as a C&I loan graded 5. Rsk Integrated

8 has two major dsadvantages: loss data s rarely avalable for long tme perods, and past correlatons may not reflect future correlatons because the asset mx may have changed 7 or the loss drver may change 8. These lmtatons mean that top-down estmates of correlaton are almost always supplemented by bottom-up estmates based on the estmated volatlty of asset values and the Merton model 9. For CRE the same process can be accomplshed wth more granularty and subtlety by co-smulatng defaults drven by market and economc condtons. The planest example s to smulate two CRE deals together across thousands of alternatve market envronments and calculate the correlaton between them by lookng at how often they default together. A slghtly more subtle example s to consder the smulaton of a commercal real estate deal next to a C&I deal where the rsk of the C&I deal ncludes the state of the economy 10. In ths case the drvng economc factors are correlated and the correlaton between losses on the CRE deal and the C&I deal s estmated from the number of scenaros n whch they default jontly and separately. Gven ths mechansm for estmatng correlatons, there are two alternatve paths to mplementaton: one s to put all the assets nto the smulaton engne and drectly generate the loss dstrbuton. The other alternatve s to take the correlaton coeffcents estmated by the smulaton engne and put those coeffcents nto some other calculaton engne for economc captal. Both approaches gve results that allow the CRE results from cashflow smulaton to be used wth results for the other asset-classes and the detals are dscussed n Appendx 2. Integrated Loan Prcng Prcng s where all the aspects of rsk measurement come together and mpact the busness drectly. There are four prmary approaches to prcng usng quanttatve rsk measurement: Grade grd RARORC (Rsk Adjusted Return on Regulatory Captal) RAROEC (Rsk Adjusted Return on Economc Captal) CAPM (Captal Asset Prcng Model) 7 For example f the current portfolo has hgher LTVs than the hstorcal portfolo. 8 For example n the prevous credt-cycle, losses on two dfferent portfolos may have moved closely together because n that crss both portfolos were drven by a fall n the economy, whereas the next losses may be drven by a rse n nterest rates and f one portfolo s prmarly floatng and the other s long term fxed, there wll be lttle correlaton. 9 See Chapter 20 of The Fundamentals of Rsk Measurement, C. Marrson, McGraw Hll, Ether because the orgnal model was bult accountng for the state of the economy n ts regressors, or because senstvty to the economy was retro-ftted ( See Lnkng Stress Tests to the Real Economy, Rsk Integrated

9 The advantages and dsadvantages of each approach are dscussed n Appendx 3. Each approach can be used for commercal real estate, and cashflow smulaton can support each, but n our opnon an approach lke CAPM s the most drect. Overall, cashflow smulaton mproves horzontal ntegraton by provdng rsk metrcs that are consstent wth the metrcs beng provded for the other asset classes and by gvng a framework to evaluate the correlaton across assets. Conclusons To summarze, Enterprse Rsk Management needs to be supported by ntegrated rsk measurement, and that measurement needs to be ntegrated both vertcally across functons and horzontally across asset-classes. It s partcularly challengng for commercal real estate because of the nature of the assets and the nature of the rsk models that are best suted to assessng those assets. However, the cashflow smulaton framework gves the flexblty to have a sngle model for all of the vertcal functons wthn CRE and output results that are horzontally consstent wth the modelng frameworks used for other asset-classes. Dr. Chrs Marrson, Founder & CEO, Tel: +1 (845) , Chrs.Marrson@RskIntegrated.com Rsk Integrated

10 Appendx 1: The Typcal Profuson of Models used for Managng CRE Rsks Many banks have found t to be dffcult to acheve vertcal ntegraton for rsk measurement of the CRE assets and nstead have developed dfferent models for dfferent functons. Ths stuaton arses because t s assumed that detaled data cannot be made avalable at the portfolo level and also because t s assumed that no sngle model can fulfll all the requrements from deal structurng to economc captal. The typcal stuaton s descrbed below: For structurng and underwrtng a new loan, the lendng team makes a spreadsheet wth a cashflow model for net ncome and debt servcng over tme. The spreadsheet s coped and modfed from one deal to the next because t does not capture the unque features of each new deal. The result s that some of the rchest data s stranded n spreadsheets on desktops. Also, managers can never qute be sure of the consstency of the model beng used from one deal to the next, and of course there s a strong possblty of spreadsheet errors. For gradng the asset there s typcally some scorecard whch requres a new set of nformaton to be typed n to a system. The nputs are usually manpulated a lttle untl the grade that comes out matches expectatons. That grade s then added to a credt applcaton form, along wth the results of the spreadsheet to form the bass of the approval and prcng decson. A few nsttutons have the ablty to re-run the gradng analyss on batches of loans as loan balances, forecasts or model parameters change. But n most cases, ths level of ntegraton and automaton does not yet exst so the resultng PD, LGD or grade s manually typed nto another system. In ths case, f re-gradng s needed, t s done manually by the lendng team. For economc captal, the PDs are dumped from the gradng system or are retyped nto a system that wll feed the captal calculator. The captal calculator then takes these PDs and estmates some correlaton wth the other assets to come up wth the overall captal. By ths stage all the detal of the cashflow and rsk structure over tme has been lost so t s not possble to tell f the rsk of an ndvdual deal s correlated to nterest rates, rental rates, etc., or whch years are the most rsky. Ths loss of nformaton means that the economc captal results are ndcatve and may be drectonally correct, but are not really actonable. Smlarly for portfolo stress-testng, the nformaton on the underlyng property, lease and fnancng structure s typcally lost and the rsk s only represented by the basecase PD and LGD. Wth some assumptons those rsk numbers can be stressed wth changes n the economy and possbly wth assumed changes n the CRE market, but agan the results are so general that t s not possble to draw detaled conclusons to drect how new assets should be structured. Rsk Integrated

11 Appendx 2: Approaches to Estmatng the Correlated Loss Dstrbutons There are two prmary approaches to usng the Merton model wth cashflow smulaton to estmate correlated loss dstrbutons. One s to assess the losses together wthn the smulaton. The other s to use the smulaton to estmate the correlaton and export that correlaton to a dfferent loss calculator: In the frst case, n whch we generate the loss dstrbuton drectly n the smulaton, we smply record the sum of the losses from all assets n each scenaro as representng the portfolo loss for that scenaro. We then generate thousands of scenaros and thousands of scenaro losses to generate the loss dstrbuton and thereby read-off the requred economc captal. In ths approach we also record the loss contrbuton of each asset. In ths approach models can be used for each ndvdual asset, or the assets may be lumped together: for example nstead of assessng each ndvdual credt card holder we mght use a model for a sub-portfolo of credt cards wthn a gven FICO band. In the second case where there s an economc captal calculator whch s external to the smulaton engne there are broadly two optons: we can gve the standalone loss dstrbuton for the CRE portfolo, plus the correlaton between CRE and each of the other asset-classes, or at the other extreme we can gve the PD and LGD for each CRE loan, plus the correlaton between that loan and the rest of the portfolo. Rsk Integrated

12 Appendx 3: Rsk-based Prcng for CRE Loans There are four prmary approaches to prcng CRE loans usng quanttatve rsk models: Grade grd RARORC (Rsk Adjusted Return on Regulatory Captal) RAROEC (Rsk Adjusted Return on Economc Captal) CAPM (Captal Asset Prcng Model) The grade grd s the most common approach. In ths approach the rsk models assgn a grade, possbly wth some modfcaton from the lendng team, and then a grd s publshed by management gvng the target prce dependng on the grade and possbly the sze and assetclass. The flexblty n defnng the prces n the grd provdes management wth a pragmatc tool for controllng the lendng whlst takng nto consderaton the prevalng compettve condtons and the bank s appette for dfferent asset-classes. Ths type of prcng s farly easy to ntegrate across assets so long as the grades for each asset-class reflect the same amount of rsk, e.g., a corporate loan grade 5 has the same rsk as a CRE loan graded 5. Some banks adopt RARORC whereby they prce accordng to the expected loss and the cost of the regulatory captal to be held. Ths has the advantage that there s no need to calculate nter-asset correlatons and that there s an external benchmark as to the amount of captal requred, but t has the dsadvantage that regulatory captal s only expected to be reasonably accurate at the portfolo level and t may be a very poor estmate of the rsk for ndvdual assets. To use RARORC the bank needs to estmate the expected loss over the lfe of the loan, and t needs to have a calculator embedded n the system reflectng the latest captal rules. RAROEC requres an estmaton of the margnal contrbuton of the deal to the bank s economc captal. The frst problem n assessng RAROEC s n estmatng the correlaton wth other assets. Ths s especally dffcult for CRE where the correlaton for one deal may be very dfferent from another. Even wthn a deal the correlaton may change from year to year: n one year the drver of loss mght be changes n rental rates and n another year t mght be changes n nterest rates. The full estmaton of economc captal for a CRE deal therefore requres a mult-year smulaton of the asset alongsde a model for the shape of the bank s future portfolo. Ths s qute possble once a form has been decded for the future portfolo. However there are a couple of problems wth RAROEC that are exposed by CRE. One s that as other busness unts change ther asset mx over the lfe of the CRE deal, ts correlaton to that portfolo wll change and so ts measured captal wll also change. The other problem s that RAROEC does not take nto account the fact that when a new asset s added to the portfolo t not only changes the amount of requred economc captal, but also changes the cost of the captal because t changes the correlaton between the bank s captal and the rest of the market. In extreme cases t can cause a sgnfcant dstorton n prcng and behavors, and on Rsk Integrated

13 frequent occasons ths effect leads to arguments between busness groups who thought they had locked n ther cost of economc captal, but then found t was changed because of the actons of another group. In RAROEC the cost of captal for asset s defned as follows: Where h s the hurdle rate for the bank m s the captal multpler Captal Cost hm UL s the average correlaton between loss on asset and loss on the portfolo UL s the standard devaton of loss for asset In theory, h s set for the bank as a whole and depends on the expected return on equty for the bank. The expected return depends on the volatlty of credt losses, the market s prce for rsk, and the correlaton between the equty value and the general market. The RAROEC approach has several well recognzed operatonal ssues such as determnng how, or whether, the cost of captal the measured proftablty should change f the composton of the portfolo changes and therefore causes a change n. Ths partcular ssue leads busness unts to be n conflct f the actons of one unt affect the correlaton and prcng of another unt. For mult-year assets, another challenge s to determne what the future portfolo composton s lkely to be, especally f there are alternatve strategc plans wth dfferent mplcatons for. These challenges are well recognzed, but a more subtle ssue s that the addton of each new asset changes the hurdle rate. To see ths, consder the followng example. Consder a bank wth a portfolo whch has zero correlaton to the general market level. The equty of ths bank would be a good dversfyng asset for any portfolo manager and would be prced accordng to the expected cashflows dscounted at the rsk-free rate,.e., the hurdle rate would be the rsk-free rate. Now consder a busness unt who wants to buy a large block of the S&P 500. It would be uncorrelated wth the bank s portfolo and therefore would be zero and there would be no captal cost. However, by addng ths asset, the bank s whole portfolo s now correlated a lttle wth the market and the hurdle rate wll rse for all the busness unts. Rsk Integrated

14 In the RAROEC framework ths problem mght be tackled by a polcy rule, or by complcated math to relate and h. A more drect approach s to drectly focus on the correlaton between asset and the market. Ths eventually collapses nto a drect-market approach such as CAPM. Economc captal s an mportant metrc to manage the soundness of the nsttuton as a whole, but for prcng ndvdual transactons, t s easer and more correct to use CAPM. In CAPM the prce of the asset stll depends on ts EL and UL, but and the market s prce for rsk. compared wth RAROEC: hm s replaced wth beta The drect-market approaches have several advantages The prcng s consstent wth asset managers and traders (.e., a smlar bond and loan to the same company wll have the same prce) There s no nterference between one busness unt and another There are no condtons n whch assets are underprced relatve to ther full market value They are much easer to mplement than an hm approach CAPM has the advantage of not relyng prmarly on estmates of the correlaton between the bank s assets because the correlaton s assessed relatve to the broad market. dsadvantage of CAPM s that t does not, on ts own, dctate hgher prces for assets that concentrate the portfolo and lead to a hgher probablty of the bank s default. The typcal soluton to the correlaton problem s that the portfolo manager orders lmts to the orgnaton of any asset-class whch s causng excessve concentraton. The However, the theoretcally pure answer s that a small cost should be added for the asset s contrbuton to the probablty of the bank defaultng, multpled by the cost of a bank default. Ths small correcton does requre the correlaton to be known wth the rest of the portfolo, but f ths term s neglected, t s possble to assess the prce of the asset on a standalone bass. Rsk Integrated