Under the Internal Capital Adequacy and Assessment Process (ICAAP) the bank will make use of internal models to assess, quantify and stress test risk. There are a number of principles regarding the model building process that the bank should keep in mind when developing these internal models. This post is based on the paper “Model Risk” by Emanuel Derman (Quantitative Strategies Research Notes – April 1996).

# Process and Risks of Model Building

The internal capital adequacy and assessment process employs the use of internal models of a bank (either in-built or outsourced) for quantifying risk and consequently assessing capital charge. These models include models for assessing and stress testing market risk, credit risk, interest rate mismatch risk, etc. Technical construction and use of these models of course varies greatly based on the subject being modelled but there are some basic principles to and risks of modelling that the modeller and subsequent users of the model should be aware of. These principles and model risks along with ways to avoid or reduce the risks are discussed in the following sub-sections.

## The Model Building Process

- The first step in a modelling process is to have a thorough and intimate understanding of the subject been modelled. If we are modelling the value of a particular security then we should be knowledgeable about that security, the markets in which the security trades, how market participants believe that the security is valued and what the risk factors associated with the security are.
- Once an understanding has been developed the modeller should be able to identify which variables are considered important by market participants for discerning value and risk.
- The modeller should then assess which of these important variables can be mathematically modelled.
- Some of these selected variables will be considered as independent variables whereas others will be considered as dependent where their values are based on the values taken by the independent variables. In the case of the latter, there would need to be a specification of how the dependent variable would be affected by the independent variables. Some of these variables would be able to be measured directly from observable data; others would have to be based on judgment and inferences and therefore based on some indirect mode of measurement.
- Some variables can be treated as deterministic, that is their estimated values will not contain any level or degree of uncertainty. However there are other variables whose values must incorporate a degree of uncertainty and thus must be assessed using stochastic processes.
- The source and availability of market data for variables that can be measured from observable data needs to be researched. For variables that are indirectly measured the mechanism for deducing their values should be defined.
- For stochastic variables, we would need to assess which stochastic process best describes the observed process.
- There should be a suitable trade-off between level of content to be included and the simplicity of the process for obtaining a solution from the model.
- The next stage is to develop the methodology for arriving at an analytical or numerical solution
- Once the methodology has been defined, a computer program for the model should be developed.
- After the program has been developed the model has to be tested

Incorporate the model into existing computer and human systems.

We have looked at some important aspects in the model building process. Using a model however entails taking on some risks. In the next post we will consider some of the model risks that the bank may be exposed to when using its internal models.

This is such a great resource that you are podviring and you give it away for free. I enjoy seeing websites that understand the value of podviring a prime resource for free. I truly loved reading your post. Thanks!