The Reserve Bank of India (RBI) has laid down broad principles for management of model risks in credit for banks and other regulated entities to ensure prudence and robustness.

Regulated Entities (REs) generally employ various models as part of their credit management, including for credit appraisal, borrower scoring, pricing and risk management.

Inherently, model outputs are exposed to uncertainties as they are based on assumptions which may not manifest in the envisaged ways and may take different forms in a real-world scenario, the RBI said in a draft circular on Regulatory Principles for Management of Model Risks in Credit.

This potentially exposes REs to model risk, which has implications on prudential aspects of credit risk management, compliance and reputational risk.

"REs shall put in place a detailed Board approved policy with regard to model risk management framework for all models deployed, covering the entire model life cycle," the draft said.

The policy should cover details of governance and oversight aspects commensurate with model materiality; processes around model development or selection; and documentation for models deployed.

It should also cover independent vetting / ongoing validation or review processes; change control; and the monitoring and reporting framework including role of internal audit function.

Credit risk model refers to any quantitative method that applies statistical, economic, financial, or mathematical principles and assumptions to process data into an output to be used for credit decisions.

The models used by REs may either be developed internally or sourced from external third-party suppliers, including under collaborative lending arrangements, or can be a mix of both as per the provisions of the policy, the draft said.

"The objectives of the model, problem statements and solution sought from the model should be clearly defined and shall be essential pillars of model development," said the draft.

Further, the model should have the necessary scalability and flexibility to meet the needs of dynamic business conditions.

Each model should be validated before deployment as well as after any subsequent amendments owing to material events or as part of periodic reviews, which shall be at least on a yearly basis, the draft added.