r espectively, for all sub-classes. T he reference to the application of the default

E C B • T h e N e w Ba se l C a p i t a l A c c o r d • A u g u st 2 0 0 3 8 ascer tain the r elative significance of the exposur e. Fur ther mor e, the data used in or der to deter mine the banks’ own estimates Pr obability of D efault, Loss G iven D efault and Exposur e At D efault may be r elevant at the gr oup level but not at the local level. H ence, it may be useful to ensur e the r elevance of the char acter istics of the own estimates by imposing the condition that banks’ own estimates used locally should r eflect local char acter istics. T her e seems to be an imbalance between the exhaustive but necessar y list of minimum r equir ements that banks should comply with in or der to be eligible for the IRB appr oach par agr aphs 349 to 500 and the assessment and action to be taken in cases of non-compliance. W ith r egar d to the latter , only ver y gener al r efer ences to situations that might lead to super visor y action, using phr ases such as “not in complete compliance”, “timely r etur n to compliance plan” and “duration of non-compliance”, are found in a single paragraph paragraph 355. M or e guidance on these issues would pr omote consistent implementation. Alter natively, the AIG should addr ess these issues. T her e ar e some elements of the pr oposals that make sense fr om a functional point of view to ensur e that banks’ pr actices ar e pr oper ly taken into account when laying down the r ules on compliance with minimum requirements. H owever, these elements need enhanced monitoring to ensure a level playing field and a pr udent outcome when estimating capital char ges. Against this backgr ound, the permission for banks to use human judgement to cor r ect the r ating outcome of cr edit scor ing models par agr aph 379 or to over r ide the outputs of the r ating pr ocess par agr aph 390 is welcomed as a potential means to impr ove the over all pr ocess. In this context, a r efer ence would be welcomed to the need for a per iodic r eview by inter nal andor exter nal auditor s of the pr actices of banks wher e human judgement has been allowed to over r ide the r ating gener ated by the model. In addition, banks should be r eady to demonstr ate pr udent use of human judgement in their models’ r atings whenever requested to do so by supervisory authorities or in the context of per iodic r eviews. T he time hor izon used in pr obability of default PD estimations is set at one year par agr aph 376. A r efer ence to the fact that the BC BS will monitor developments in r isk modelling and banking pr actice with r espect to the assessment horizon would be desirable. A sentence could be added in the text stating that the BC BS may r evisit its r equir ements on the assessment hor izon. T his would also put the afor ementioned r efer ences to pr udent PD valuations into context. T he use of str ess tests par agr aphs 396 to 399 is welcomed as a step towar ds ensur ing capital adequacy under adver se economic, mar ket and liquidity conditions. T he EC B would suppor t a r efer ence to the fact that, in addition to the alr eady mentioned follow up at the national level par agr aph 399, the BC BS will monitor developments in the field of str ess tests and may come up with mor e concr ete guidance, if r equir ed. In the same vein, an explicit r efer ence to the r esults of str ess tests could be made in the context of the inter nal r ating r epor ting to senior management paragraphs 401 and 402, section on corporate governance and oversight. T he pr oposed tr eatment of specialised lending and its five sub-classes par agr aphs 187 to 196, 218 to 220 and 244 to 253, although detailed, seems to leave open issues such as the assessment of r estr uctur ing by banks in distr essed situations, which may in tur n have level playing field implications as the r isk weights of the “weak” and “default” categor ies differ significantly 350 ver sus

625, r espectively, for all sub-classes. T he reference to the application of the default

r ate for r etail exposur es at the level of the facility r ather than at the level of obligor , and to the fact that a default by one bor r ower on one obligation does not r equir e a bank to tr eat all other obligations to the banking group as defaulted paragraph 417, makes 9 E C B • T h e N e w Ba se l C a p i t a l A c c o r d • A u g u st 2 0 0 3 sense fr om a functional point of view. H owever , a default by a bor r ower on one obligation may in pr actice also signal defaults on other obligations acr oss the banking gr oup. T hus it is pr oposed that ther e be an explicit r efer ence to r ecognise the fact that banks will be allowed to make their own assessments and that super visor s will r eview them. T he tr eatment of pr eviously defaulted exposures as non-defaulted and their possible subsequent r eclassification as second defaulted exposures paragraph 419 could be complemented by a r efer ence to the need for pr udent inter nal pr ocedur es by banks, including r eviews by the inter nal or exter nal auditor s and their ability to demonstr ate pr udent tr eatment to their super visor s. In the same vein, the r e-ageing of facilities and gr anting of extensions, defer r als, r enewals and r ewr ites to existing accounts par agr aph 420 may r equir e guidance fr om the BC BS with a view to pr omoting consistency in banks’ practices. 12 W ith r egar d to the length of the under lying histor ical obser vation per iod i.e. 5 year s, it is stated that, if the available obser vation per iod spans a longer per iod, the latter must be used par agr aph 425. H owever , it is not clear that long ser ies of histor ical data would always be mor e appr opr iate because of changes in the por tfolio base, r ating methodologies or economic cir cumstances. A bank should not have to give equal impor tance to histor ical data if it is possible to demonstr ate that r ecent data ar e mor e useful for the estimation of r isk par ameter s. O n the mitigating effect of guar antees wher e own estimates of LG D ar e used two options ar e given, an adjustment of the PD estimate or an adjustment of the LG D estimate par agr aph 442. It is not clear why the adjustment should concer n the PD , or the bor r ower ’s specific r isk. T he adjustment should be confined to the LG D , which takes into account tr ansaction-specific r isks. T his would also make the adjustment consistent with paragraphs 359 and 393. T he r ule stating that a bank cannot assign an adjusted PD or LG D to the guar anteed exposur e if the adjusted r isk weight would be lower than that of a compar able dir ect exposure to the guarantor paragraph 444 may be over ly str ingent. A pr udent move towards a more risk-sensitive approach could be consider ed in this context, based on fur ther r esear ch on the r isk mitigation of “double default”. T he wide r ecognition of types of eligible guar antor s 13 which is unlimited for banks under the advanced IRB approach paragraph

445, including the r ecognition of conditional guar antees par agr aph 446, may war r ant