Model Assumptions
From Open Risk Manual
Contents
Definition
Model Assumptions denotes the large collection of explicitly stated (or implicit premised), conventions, choices and other specifications on which any Risk Model is based. The suitability of those assumptions is a major factor behind the Model Risk associated with a given model.
Context
In the context of modelling economic, financial or other complex systems, model assumptions are necessary to simplify Model Development, or even make a model feasible / tractable.
Examples
Assumptions can be conceptual, mathematical or numerical in nature
Conceptual Assumptions
This class concerns idealizations and simplifications of the underlying entity or system that is being modeled
- Idealizing and simplifying the behavior of complex systems such as markets (e.g. via assumptions on liquidity)
- Simplifying the model of counterparties to legal contracts
- Simplifying the model of an economy by assuming away heterogeneity
- Injecting rational behavior assumptions
Mathematical Assumptions
This class concerns assumptions around the mathematical representation of the modelled system
- Overall approach: e.g. Bayesian or Frequentist for statistical models
- Choice of univariate distributions among competing choices
- Implicity or explicit choices about multi-variate distributions and dependency
- Choice of parameter fitting / calibration approaches among competing choices
Numerical Assumptions
- Explicit selection of numerical values
Mitigation
- Proper Model Documentation is a primary mechanism for recognizing (identifying) and aiming to control the possibly adverse influence of assumptions.
- Model Validation, an independent review of modelling frameworks offers a second opinion.
Issues and Challenges
- Undocumented / unrecognised model assumptions
