The Margin of Conservatism (MoC) Framework

Since the publication of Basel II in 2004, and its first application in 2007-2008, banks have been considering the so-called margin of conservatism (MoC) in the estimation of the risk parameters: Probability of Default (PD), Loss Given Default (LGD), and Credit Conversion Factor (CCF). However, there have been no concrete technical guidance about the technical implementation until the European Banking Authority (EBA) published guidelines (GLs) on PD estimation, LGD estimation and the treatment of defaulted exposures in November 2017.

MoC related chapter of these GLs aims at addressing deficiencies and any other uncertainties related to the estimation of the risk parameters. This is done by identifying any uncertainties and quantifying them in order to determine the appropriate MoC add-on.

Banks tend to struggle with finding adequate trade-off between  efforts to be made  and complexity of implementation of their MoC frameworks. Considering the likely significant impact of the MoC on the RWA, simplified approaches, though could be used as a preliminary step, are not recommended, because high MoC might be assigned to the deficiencies because of the lack of numerical evidence in support of the quantification methods.

The following graph summarizes the development process of the MoC framework (click to enlarge).

 

 

Sources of the Deficiencies

The first step for the identification of the possible deficiencies is to go through all relevant internal and external modelling documentation. Examples of sources for potential deficiencies could be as follows:

  • Model validation findings
  • Model development documentation (model weakness, issues etc.)
  • Internal audit reports and findings
  • ECB findings
  • Deficiencies found in reference data set (RDS)
  • Data representativeness analyses
  • Documents describing changes in the credit risk management processes

In order to avoid redundant work during subsequent prioritization and quantification of the factors leading to the deficiencies, one needs to make sure that only relevant deficiencies are taken into consideration (click to enlarge).

 

Next important step after filtering relevant deficiencies is prioritization their deficiencies as, in practice, there might exist hundreds of deficiencies for a single IRB model. Examples of simple but practical prioritization criteria might be based on the percentage of observations impacted by deficiencies and the significance of deficiencies identified by validation or internal audit reports.

MoC Categorization

As required by the GLs, deficiencies that should be driving the MoC have been divided into the three categories as shown in the figure below (click to enlarge):

As an example of methodological deficiencies, one can mention the biased allocation methodology of the indirect costs, which are usually oversimplified for the purposes of IRB model development (e.g. total indirect costs per year divided by the number of processed customers/loans). The simplification might be here the real driving element of the deficiency and clearly the relevant add on of margin of error can only be measured once we now the outcome of the advanced model’s prediction level. unless there is the more advanced approach in place you may not measure the level of the inaccuracy.

It is important to note that deficiencies from category B are the most difficult ones to quantify as they reflect changes in the processes, the policies and the risk appetite of the bank. Sometimes it is difficult to follow these changes based on historical data or available documentation as few banks had kept record of their improvement work across the processes. An important sub-category to be emphasized here is the sub-category related to the forward-looking expectations (predictions). This means that banks are expected to anticipate future changes in their internal policies and processes, even if they have not yet come into force.

Appropriate Adjustments and MoC

In order to overcome biases in risk parameter estimates stemming from the identified deficiencies, institutions should apply adequate methodologies to correct the identified deficiencies as far as possible. The impact of these methodologies on the risk parameter (‘appropriate adjustment’) should lead to a more accurate estimate of the risk parameter in question (‘the best estimate’). It represents either an increase or a decrease in the value of the risk parameter, and institutions should ensure and provide evidence that the application of an appropriate adjustment results in a better estimate.

Once all relevant deficiencies have been identified, it is necessary to understand which of these deficiencies require appropriate adjustment. Reasoning behind selection of the appropriate adjustment should be documented as part of IRB modelling documentation. As an example, when selecting between mean and median values of the risk driver for replacement of missing values one must specify arguments for the selected approach.

MoC Quantification

At this point, it is important to state that there have been no widely accepted best practices for the MoC quantification yet. The general MoC framework as it is proposed by the GLs consists of the deficiencies of different nature and subcategories. Hence, it is hardly possible to develop a single top-down methodology for its quantification. Depending on the individual deficiency, the estimation of the MoC add-on can be technically very challenging.

The following high-level framework could be proposed as an optimal trade-off between providing the much needed flexibility for the estimation of the individual deficiencies, and at the same time establishing an umbrella logic to automate the calculations and aggregating the add-on MoCs from the different deficiencies.

The first step is to group observations corresponding to the deficiency into a separate „Deficiency“ group and define „Control“ group representative of the “Deficiency” group. Whenever possible, the “Control” group should be defined using quantitative methods (e.g. representativeness tests). If it is not possible to conduct the representativeness analysis, at least a logical expert judgement based on the economic arguments should be provided for the selection of the “Control” group.

The second step, is to generate a distribution of the risk parameter errors (difference between the average predicted and average realized values of the risk parameter) using bootstrapping techniques both for the “Control” and the “Deficiency” groups.

As can be seen on the picture below, comparison of the “Deficiency” and the “Control” error distributions could lead to one of the following general cases (click to enlarge):

  • Additional uncertainty in the “Deficiency” group error,
  • Additional bias in the “Deficiency” group error,
  • Both of the above

 

In the third step, the MoC add-on is quantified as a difference between the percentiles (i.e. 95th) of the error distributions generated in step 2.

Finally, the risk parameter is adjusted using the quantified MoC.

Conclusion

Guidelines on PD estimation, LGD estimation and the treatment of defaulted exposures have set up standards for the design and estimation of the MoC add-on, which are significantly different from what most banks have been using in the past. In most cases, bank’s approaches were too simplistic, not relying on the actual data level discrepancies and had not been systematically covering the whole deck of the prospective deficiencies. The margin of error has been treated as an one-off add-on for all the prospective deficiencies, which might or might not be sufficient or in times even can turn out to be too conservative. Proper implementation of this methodology is crucially important for banks as it directly affects RWA levels and at the same time it requires highly skilled team due to its complex nature and lack of still not fully clear instructions on the quantification methodologies.

The EBA guidelines are very helpful to give the right direction to the analytical experts how to build up the necessary archive of methods, and how to manage the level of comprehensiveness. However, they seem not to discuss the issues around data needs for the estimation of the MoC add-ons for the various deficiencies, or what would be the cost of the simplification in terms of the increased capital needs.  Consequently, it is important to plan MoC development projects in advance, involve experience team members and avoid oversimplified methodological approaches, as they will likely lead to higher levels of RWA. It would be a serious planning mistake to believe that the MoC framework can be set up without significant data collection exercise or without commitment of the expert knowledge.

Please contact us for more information:

 

 

Kaan Aksel

Telefon: +49 69 9585 5874

kaan.aksel@pwc.com

 

 

Petr Geraskin

Telefon: +49 69 9585 6006

geraskin.petr@pwc.com

 

Hinterlassen Sie einen Kommentar

Ihre E-Mail-Adresse wird nicht veröffentlicht. Erforderliche Felder sind mit * markiert.

/* */