Internal rating systems and processes

The Group defines a rating system as all of the methods, processes, controls, data collection and IT systems that are used for the assessment of credit risk and for classification of exposures to a pool with a specified risk level, including the rules on the priority of rating models, if applicable, and the rules for overriding rating grades. Elements of the rating system include PD, LGD, CCF-EAD models (hereinafter: models) and methodologies for evaluating specialized lending.

Evaluation of the client’s credit risk in respect to its probability of default (PD) is based on a uniform rating scale, referred to as the Master Scale.

The Master Scale (MS) consists of 15 rating grades. where the given ratings are as follows:

1. Maximum security – only for sovereigns

2. Superior quality

3. Very high quality

4. High quality

5. Very good quality

6. Good quality

7. Average high quality

8. Average quality

9. Average low quality

10. Low quality

11. Very low quality

12. Restricted crediting

13. Soft signs of impairment

14. String signs of impairment

15. Default.

Ratings 13 – 15 are procedural ones, reserved to exposures with deteriorated quality.

All the clients with available lending, whether or not they actually use the approved credit limits and all other participants of credit transactions should have a previously awarded rating and should be assigned to an appropriate pool.

An adequate credit or rating policy should specify the model to be used for rating purposes or a homogenous pool for a given client segment.

Each PD model used must be calibrated to MS based on the observed or estimated probability of default.

The rating for governments, central banks, international organizations, multilateral development banks and Institutions may be assigned based on a rating awarded by recognized rating agencies, mapped to the Master Scale. Should the above-mentioned entities have more than one classification awarded by recognized rating agencies (split rating) the rating corresponding to the second best risk shall be taken into account at all times. The table showing relationships between internal and external risk grades is presented in chapter 7.3 of the Disclosures. The Bank recognizes the following external rating agencies for comparison purposes: Fitch, Moody’s, Standard & Poor’s.

Master Scale Rating


In case of retail customers, rating awarded through a behavioral model (behavioral rating) by default takes precedence over a rating awarded through an application model (application rating) if behavioral rating only is awarded. In case of corporate customers, awarded rating comes from a combination of a quantitative model based on an analysis of data from financial statements and on a qualitative model of customer including wages for each of components. Currently Bank is in a course of an implementation of an additional module in corporate models that is based on behavioral variables.

Procedural ratings (13. 14 and 15 according to the Master Scale) are awarded to clients showing signs of deteriorating borrowing capacity and creditworthiness or with overdue debt.

Procedural ratings by default take precedence over application ratings.

After the pre-conditions necessary to award any of the procedural ratings are no longer satisfied, ratings 13 and 14 expire immediately, while rating 15 either expires or is maintained for a “quarantine period”.

Description of the internal ratings process

1. Central governments and central banks

This exposure class is excluded permanently from the IRB approach.

2. Institutions

This exposure class is excluded permanently from the IRB approach.

3. Corporates. including SMEs. specialized lending and purchased corporate receivables

Exposure classes subject to the plan of gradual implementation of the IRB approach.

4. Retail exposures

PD models

The rating process in Bank Millennium is based on the following principles:

  • Awarding risk grades to all customers and credit exposures;
  • All credit decisions should be preceded by awarding a risk grade to the client;
  • In the retail segment, the rating process is based on PD scoring/rating models;
  • The rating process is separated and independent from the credit decision process.


The presented rating principles apply to all categories of retail exposures: retail exposures to individuals secured by residential real estate, qualifying revolving retail exposures and other retail exposures.

The class of retail exposures to individuals secured by residential real estate include exposures which are mortgage loans or home equity loans granted to retail clients (small businesses and private individuals) and secured by mortgage.

The class of qualifying revolving retail exposures includes exposures to natural persons which are unsecured, renewable, with total exposure not exceeding EUR 100.000 and which meet the requirement of low volatility of loss rates.

All the retail exposures that do not qualify to the above categories are treated as other retail exposures. They are covered by the IRB roll-out plan, and according to the last IRB permission, IRB Approval Pack regarding these exposures was delivered on the 30th June, 2015 (update of that Approval Pack will be submitted at the turn of 2017 and 2018).

In the rating process, the powers are allocated as follows:

  • Data input;
  • Verification of data;
  • Awarding of the final risk grade (automated decision).

Model-based risk grades and procedural ratings are awarded automatically and are not subject to expert adjustments.

In the rating process. the Bank uses data from various available sources:

  •  internal sources (Bank’s IT systems);
  •  external sources (Biuro Informacji Kredytowej S.A.);
  • data received from customers.

With respect to probability of default (PD) models for retail exposures, there is a rating system in place for microbusinesses and a rating system for private individuals. Both systems use behavioral scoring models and application scoring models designed for specific client and/or product groups. Procedural ratings are awarded to clients showing signs of deteriorating borrowing capacity and creditworthiness or with past due debt.

A procedural rating has the priority in use. If the client has no procedural rating then the behavioral rating should be used, provided that it has been awarded. Behavioral rating is awarded for the first time after three months of the client’s cooperation with the Bank and then monthly, provided that the client’s accounts meet the requirements of the behavioral model. If the client has no behavioral or procedural rating then the application rating should be used.

LGD models

Loss Given Default (LGD) models have been built for the following two portfolios:

  •  unsecured portfolio for retail clients.
  • portfolio secured by residential real estate for retail clients.

Pursuant to CRR. as amended, banks must estimate LGD parameters using data on defaulted exposures from all the available sources, taking into account all information that is significant for the estimation of economic loss levels.

Accordingly, the Bank has estimated LGD parameters using a database that contains all the defaults resulting from quantitative and qualitative premises included in default definitions.

According to the LGD calculation methodology, the main factors in the calculation include: probability of cure or completion of the client recovery process, value of recoveries, costs and discount rate.

The Bank has taken the following approach to building LGD models:

  • Estimate the probability of the path of cure from the default status. i.e. a probability tree;
  • Estimate loss parameters for each path of cure from default.

Loss given default is estimated at a transaction level.

Exposure at Default (EAD) models

An EAD model has been built for retail portfolio exposures. When estimating EAD, exposure at default was compared to the value of the limit and the book value of the exposure observed one year before the default event. Credit Conversion Factor (CCF) parameters have been calculated for product groups for which an off-balance sheet exposure could occur and where the Bank had a significant number of observations that enabled statistical conclusions to be drawn. i.e. for overdraft limits and for credit cards. In the case of guarantees, where the number of observations was too low to carry out statistical analyses, a conservative CCF value was used. At the same time, EAD model for RRE portfolio was not developed due to immaterial number of observations.

5. Equity exposures

In equity exposures. the Millennium Group classifies shares and equity instruments held by any of the Bank’s units. On the consolidated basis, however, the shares representing investments in subsidiaries are excluded, since those are classified as intragroup transactions. However, due to the fact that the total value of the Group’s equity portfolio is insignificant, it has been decided that these exposures should be excluded from the IRB approach permanently and the capital requirement for these exposures should be calculated based on the standardized approach.

6. Exposure values and adjustments

The below table presents the basic aggregates and parameters used in calculation of own funds requirements in IRB method. As for exposure classes under IRB method, exposure amounts, CCF’s, average PD’s, debtors amount, average LGD’s, risk-weighted assets, risk density, expected loss and specific credit risk adjustments, break downed by probability of default (PD) brackets are showed.

Exposures to credit risk by exposure classes and PD brackets
PD scale Original
gross exposures
exposures pre-CCF
Average CCF EAD
and post CCF
Average PD Number
of obligors
Average LGD RWAs RWA density EL Value
and provisions
QRRE 0.00 do <0.15 206 297 1 133 041 77.65% 1 086 081 0.08% 148 249 81.98% 47 145 4.34% 712 386
QRRE 0.15 do <0.25 112 053 262 370 82.57% 328 702 0.20% 58 807 77.48% 28 710 8.73% 509 234
QRRE 0.25 to <0.50 119 677 168 826 84.23% 261 874 0.39% 45 961 74.69% 37 965 14.50% 763 244
QRRE 0.50 to <0.75 162 764 138 751 84.10% 279 457 0.71% 46 869 75.49% 65 714 23.51% 1 498 367
QRRE 0.75 to <2.50 324 395 172 099 86.63% 473 478 1.67% 70 133 73.70% 205 915 43.49% 5 816 794
QRRE 2.50 to <10.00 215 105 64 265 85.36% 269 959 5.73% 38 415 75.31% 280 717 103.99% 11 735 944
QRRE 10.00 to <100.00 114 049 19 350 85.31% 130 556 27.84% 18 425 78.54% 278 716 213.48% 28 238 1 357
QRRE 100.00 (default) 103 936 3 866 0.00% 103 936 100.00% 17 095 97.98% 4 378 4.21% 101 840 48 117
QRRE Razem 1 358 277 1 962 569 80.29% 2 934 042 5.73% 443 954 78.67% 949 259 32.35% 151 112 52 442
Residential Retail 0.00 do <0.15 17 418 586 66 358 0.00% 17 418 586 0.08% 80 660 31.40% 1 159 233 6.66% 4 375 14 327
Residential Retail 0.15 do <0.25 2 807 693 13 864 0.14% 2 807 712 0.19% 11 769 32.14% 369 272 13.15% 1 725 2 901
Residential Retail 0.25 to <0.50 1 811 681 10 440 1.96% 1 811 886 0.39% 7 602 32.25% 404 818 22.34% 2 279 2 708
Residential Retail 0.50 to <0.75 1 365 964 6 827 0.00% 1 365 964 0.71% 5 384 32.52% 468 281 34.28% 3 154 2 641
Residential Retail 0.75 to <2.50 1 722 193 12 784 0.00% 1 722 193 1.66% 6 946 32.54% 1 022 123 59.35% 9 324 4 456
Residential Retail 2.50 to <10.00 1 045 814 5 343 0.02% 1 045 815 5.79% 4 079 32.40% 1 235 080 118.10% 19 576 5 340
Residential Retail 10.00 to <100.00 809 295 647 0.00% 809 295 19.97% 2 895 33.14% 1 501 917 185.58% 53 956 9 294
Residential Retail 100.00 (default) 854 927 0 854 927 100.00% 2 640 64.07% 126 447 14.79% 547 748 224 861
Residential Retail Total 27 836 153 116 262 0.19% 27 836 378 4.10% 121 975 32.75% 6 287 171 22.59% 642 138 266 528
Total (all classes)   29 194 430 2 078 831 75.81% 30 770 420 4.26% 516 058 37.12% 7 236 430 23.52% 793 250 318 969


7. Drivers that impacted on the loss experience, in conjunction with the actual results in a longer term

Comparison of actual and modeled PD +

The tables below present the calculation of actual default rates and estimated default rates for portfolios covered by the permission to use the IRB approach.

Actual and estimated default rates for the QRRE portfolio (in %)
Term Estimated
2012 3.12% 2.16%
2013 2.74% 2.16%
2014 2.68% 1.99%
2015 2.69% 1.82%
2016 2.47% 1.59%


Actual and estimated default rates for the portfolio of loans secured by residential property (in %)
Term Estimated
2012 1.12% 0.49%
2013 0.80% 0.49%
2014 0.78% 0.48%
2015 0.82% 0.55%
2016 0.88% 0.60%


In case of QRRE portfolio, actual default rates have been decreased in the last years and are lower than estimated. The observed decrease in actual default rates in recent years is consistent with the observed market trends and may be explained by the following:

  • Stable macroeconomic situation,
  • Change of the Bank’s credit policy introduced in 2010 in response to the global financial crisis.

In case of RRE portfolio, actual default rates are slightly higher in the last years. (however they remains on a visible lower level than estimated values); that increase regards the portfolio denominated in foreign currencies (mostly CHF) and stems from an increase of exchange rate and increase of an installments paid by customers.

The actual default rates were also lower than the average probability of default (PD) mainly because of consideration in estimation of long-term PD an additional conservative buffer, connected with estimation errors, that increases PD values.

Comparison of actual and modeled CCF +

The analysis of actual CCF has been conducted for QRRE portfolio cases defaulted during the calendar year 2016 (reporting period) and 2015 (comparative period) which, at the end of the preceding year (31 December 2015 and 31 December 2014, respectively) were not defaulted and had a positive off-balance sheet exposure. The analysis involved a comparison of the average actual conversion factors with average modeled levels (average weighed by the amount of off-balance sheet exposure was applied in both cases). The modeled factors include a number of conservative haircuts and should be higher than the actual figures. The results are presented in the table below.

Comparison of actual and modeled CCF (in %)
CCF 2016 2015
Modeled CCF 86.1% 86.1%
Actual CCF 61.0% 61.6%


In both the reporting period and the comparative period, actual CCF levels did not exceed the modeled levels. Since the Bank has not recorded higher than expected credit conversion factors, this credit risk element does not lead to the occurrence of higher than expected losses.


Comparison of actual and modeled LGD +

The analysis of actual LGD was carried out for cases from RRE and QRRE portfolios. Calculation of actual LGD figures requires a longer time horizon, because recoveries may occur only after the exposure achieves the default status. Accordingly. the calculation of actual LGDs was based on the cases, which defaulted by December 2014. The average LGD calculated on the basis of these cases (average weighed by the exposure size) was compared with the average LGD level obtained from the model used (average weighed by the exposure size was applied in both cases). The modeled values include a number of conservative haircuts and should be higher than the actual losses. The results are presented in the table below.

Comparison of actual and modeled LGD (in %)
LGD Portfolio
Modeled LGD 23.1% 49.4%
Actual LGD 31.9% 82.2%

For both analyzed portfolio, modeled loss amounts were much higher than actual figures. We can therefore state that there were no unexpected losses associated with LGD levels and the model used has proven to be sufficiently conservative.

The Group does not have companies conducting credit activity abroad. (CRR 452.j)

Previous page Approval to use the IRB approach
Next page Use of internal estimates