Default Definition and Model Validation


A prerequisite for the development of rating methodologies and the determination of risk parameters is a proper definition, identification and storage of the default event of a customer. We apply a default definition in accordance with the requirements of Section 125 SolvV as confirmed by the BaFin as part of the IRBA approval process.

As an important element of our risk management framework we regularly validate our rating methodologies and credit risk parameters. Whereas the rating methodology validation focuses on the discriminatory power of the models, the risk parameter validation for PD, LGD and EAD analyzes the predictive power of those parameters when compared against historical default experiences.

According to our standards, and in line with the SolvV-defined minimum requirements, the parameters PD, LGD and EAD are reviewed annually. The validation process for parameters as used by us excluding Postbank is coordinated and supervised by a validation working group composed of members from Finance, Risk Analytics and Instruments and Credit Risk Management. Risk parameter validations consist of quantitative analyses of internal historical data and are enriched by qualitative assessments in case data for validation is not sufficient for getting reliable results. A recalibration of specific parameter settings is triggered based on validation results if required. In addition to annual validations, ad hoc reviews are performed where appropriate as a reaction to quality deterioration at an early stage due to systematic changes of input factors (e.g. changes in payment behavior) or changes in the structure of the portfolio. The reviews conducted in 2012 for advanced IRBA rating systems triggered recalibrations as shown in the table below. 26 new risk parameters are applied due to newly approved rating systems or due to increased granularity in existing risk parameter settings. None of the recalibrations individually nor the impact of all recalibrations in the aggregate materially impacted our regulatory capital requirements.

Analogously at Postbank the allocation mechanism of the master scale to the probabilities of default as well as the results of the estimations of the input parameters PD, CCF and LGD are reviewed annually. Postbank’s model validation committee is responsible for supervising the annual validation process of all models. Via a cross committee membership Deutsche Bank senior managers join in Postbank committees and vice versa, to ensure a joint governance.

Validation results for risk parameters used in our advanced IRBA

 

 

Dec 31, 2012

 

PD

LGD

EAD

 

Count

EAD in %

Count

EAD in %

Count

EAD in %

Appropriate

104

91.4

100

89.8

40

79.5

Overly conservative

6

1.8

18

4.1

29

15.9

Progressive

16

6.8

11

6.1

5

4.6

Total

126

100.0

129

100.0

74

100.0

 

 

 

 

 

 

 

Thereof already recalibrated and introduced in 2012

 

 

 

 

 

 

 

 

 

 

 

 

 

Overly conservative

1

0.1

17

3.5

24

15.3

Progressive

1

0.1

7

2.0

5

4.6

Total

2

0.2

24

5.5

29

19.9

Above table summarizes the outcome of the model validations for risk parameters PD, LGD and EAD used in our advanced IRBA including Postbank. Individual risk parameter settings are classified as appropriate if no recalibration was triggered by the validation and thus the application of the current parameter setting is continued since still sufficiently conservative. A parameter classifies as overly conservative or progressive if the validation triggers a recalibration leading to a decrease or increase of the setting, respectively. The breakdown for PD, LGD and EAD is presented in counts as well as in the relative EAD attached to the respective parameter as of December 31, 2012.

The validations largely confirm our PD parameter settings. PDs classified as progressive can be subdivided into a group of 15 parameters with comparably low exposure and one parameter from Postbank assigned to approximately 4 % of our total EAD, for which a recalibration is awaiting final approval from BaFin. Similarly, LGD parameter validations show largely appropriate settings. Progressive LGDs can be subdivided analogously to PDs with one parameter from Postbank assigned to approximately 4 % of our total EAD, for which a recalibration is awaiting final approval from BaFin. For our EAD parameters, excluding Postbank, an improved and extended validation and recalibration approach has been implemented after approval from BaFin now taking into account the exposure changes for each of the twelve months prior to default and not only one year prior to default. Moreover, the extended time series leads to less variance and thus more stable parameter settings. These two effects lead to a reduction of a large number of EAD parameter settings.

Out of the 85 risk parameters where a change was suggested by the conducted validation, 55 were already introduced in 2012. The remaining 30 parameter changes are intended to be implemented in 2013. Some of these parameter changes require pending approval from BaFin prior to introduction. In addition, 5 recalibrated LGD parameters were introduced in 2012 based on validation results in 2011.

In addition to the above, the comparison of regulatory expected loss (“EL”) estimates with actual losses recorded also provides some insight into the predictive power of our parameter estimations and, therefore, EL calculations.

The EL used in this comparison is the forecast credit loss from counterparty defaults of our exposures over a one year period and is computed as the product of PD, LGD and EAD for performing exposures as of December 31 of the preceding year. The actual loss measure is defined by us as new provisions including recoveries on newly impaired exposures recorded in our financial statements through profit and loss during the respective reported years.

While we believe that this approach provides some insight, the comparison has limitations as the two measures are not directly comparable. In particular, the parameter LGD underlying the EL calculation represents the loss expectation until finalization of the workout period while the actual loss as defined above represents the accounting information recorded for one particular financial year. Furthermore, EL is a measure of expected credit losses for a snapshot of our credit exposure at a certain balance sheet date while the actual loss is recorded for a fluctuating credit portfolio over the course of a financial year, i.e., including losses in relation to new loans entered into during the year as well as offsetting releases of allowances for loan losses for loans considered impaired at time of EL determination.

According to the methodology described above, the following table provides a comparison of EL estimates for loans, commitments and contingent liabilities as of year-end 2011 till 2007, with actual losses recorded for the financial years 2012 till 2008, by regulatory exposure class for advanced IRBA exposures. Postbank is firstly reflected in the comparison of EL estimates as of year end 2010 with actual losses recorded for the financial year 2011.

Comparison of EL estimates for loans, commitments and contingent liabilities with actual losses recorded by regulatory exposure class

 

Dec 31, 2011

2012

Dec 31, 2010

2011

Dec 31, 2009

2010

Dec 31, 2008

2009

Dec 31, 2007

2008

in € m.

Expected loss

Actual
loss

Expected loss1

Actual
loss1

Expected loss

Actual
loss

Expected loss

Actual
loss

Expected loss

Actual
loss2

1

The 2010 Expected Loss and 2011 Actual Loss figures have been restated to limit disclosure to Postbank’s advanced IRBA exposure only.

2

Losses related to assets reclassified into loans under IAS 39 amendments were excluded from the actual loss for 2008 since, as of December 31, 2007, the related assets were not within the scope of the corresponding expected loss calculation for loans.

Central governments

1

2

2

2

2

Institutions

7

14

22

2

16

1

21

16

13

55

Corporates

445

393

449

363

471

358

591

1,665

320

251

Retail exposures secured by real estate property

294

337

222

359

118

101

120

140

127

125

Qualifying revolving retail exposures

23

17

2

30

2

5

2

7

2

4

Other retail exposures

418

348

390

301

301

282

311

315

226

223

Total expected loss and actual loss in the advanced IRBA

1,188

1,109

1,088

1,055

910

747

1,047

2,143

690

658

The actual loss in 2012 was 7 % lower than the expected loss and was primarily driven by the lower level of provisions in our Other retail portfolios.

The increase in expected loss as of December 31, 2011 and as of December 31, 2010 in comparison to December 31, 2009 as well as the higher actual losses in 2012 and 2011 is primarily related to the inclusion of Postbank.

In 2010 the actual loss was 18 % below the expected loss as the actual loss and was positively influenced by lower provisions taken for assets reclassified in accordance with IAS 39.

The decrease of the expected loss for 2010 compared to the expected loss for 2009 reflected the slightly improved economic environment after the financial crisis.

In 2009 actual losses exceeded the expected loss by 104 % driven mainly by material charges taken against a small number of exposures, primarily concentrated in Leveraged Finance, as well as the further deteriorating credit conditions not reflected in the expected losses for our corporate exposures at the beginning of the year.

The following table provides a year-to-year comparison of the actual loss by regulatory exposure class. Postbank is firstly included in the reporting period 2011.

Year-to-year comparison of the actual loss by IRBA exposure class

in € m.

2012

2011

2010

2009

2008

Central governments

73

Institutions

14

2

1

16

55

Corporates

393

363

358

1,665

295

Retail exposures secured by real estate property

337

359

101

140

125

Qualifying revolving retail exposures

17

30

5

7

4

Other retail exposures

348

301

282

315

223

Total actual loss by IRBA in the advanced IRBA

1,109

1,055

747

2,143

775

Our actual loss increased by € 54 million or 5 % in 2012 compared to previous year. The drivers of this increase were primarily higher actual losses in the IRBA exposure classes Other retail exposures as well as Corporates excluding Postbank partly being offset by reduction throughout Postbank’s advanced IRBA exposure classes.

New provisions established in 2011 were € 308 million higher compared to 2010 primarily due to the first time inclusion of Postbank in full year reporting.

New provisions established in 2010 were lower by € 1.4 billion compared to 2009, reflecting predominately significantly reduced provisions required for assets reclassified in accordance with IAS 39. Measures taken on portfolio and country level led to a reduction in the actual loss for our retail exposures in Spain and India, partially offset by increases in the consumer finance business in Poland. The observed decrease in actual loss were partially offset by provisions taken relating to the commercial banking activities acquired from ABN AMRO and Postbank.

The observed increase in actual loss of € 1.4 billion in 2009 compared to 2008 reflected the overall deterioration in credit conditions, predominantly on our exposure against corporates. Of this increase, 83 % was attributable to assets which had been reclassified in accordance with IAS 39, relating primarily to exposures in Leveraged Finance. Further provisions against corporate exposures were a result of deteriorating credit conditions, predominantly in Europe and the Americas. Increases recorded for our retail exposures reflected our strategy to invest in higher margin consumer finance business and were mainly a result of exacerbating economic crisis in Spain which adversely affected our mortgage loan and commercial finance portfolios there and by its consumer finance business in Poland and India.

Key figures comparison

Compare key figures of the past years. more