Model Validation is the set of principles, practices and organizational arrangements supporting a rigorous (audited) model development and validation cycle.
Model Validation ensures that developed models offer good presentation of the variables, markets or risks under consideration, i.e. that models are "fit-for-purpose"
A defining characteristic of a mature model validation framework is that relevant models are critically reviewed by a qualified party with no direct dependency on the model sponsor. Such independent model validation is a key ingredient of broader Model Governance principles, which treat model development and model use as an essentially risky activity (Model Risk).
Model Validation Elements
Elements of the model validation process confirm that there is limited and controlled Model Risk through the examination of
- the Model Documentation,
- the Model Implementation,
- Model Output and Model Usage,
Quantitative Model Validation
Quantitative model validation is a subset of overall model validation activities that is particularly relevant for statistical models.
Distinct types of quantitative model validation include:
- Point-in-time validation involves checking that time zero characteristics or initial conditions of the model adequately match expectations
- In-sample validation involves taking historical data and determining the extent to which the stylized facts of the data are captured by the model
- Out-of-sample validation or Backtesting involves looking at the output of a model from a point in time T and seeing how it performs at future time periods T + t. Out-of-sample validation helps to monitor the ongoing appropriateness of the model in light of what actually happens 
- Economic Scenario Generators, A Practical Guide, Society of Actuaries