Agentic data quality control  for  regulatory capital models

Translate regulations into data requirements and a set of deterministic checks, and use agents to investigate what fails.

Scope

Starting point

Define scope by choosing model, portfolio, and history. This will determine data requirements.

Requirements

Agentic data requirements

Agentic extraction of relevant clauses in regulation and internal docs, consolidating against the bank's existing list, and resolving each requirement down to attributes, definitions and quality levels.

Reference data

Reference data model

Entity relationship diagram of a typical input data set for RegCap models.

Data quality

Deterministic check suite

Rules driven by the data dictionary across eight dimensions, each carrying a regulatory reference, complemented by statistical and ML analysis.

Data quality

Agentic investigation

Agentic investigation of each findings: hypothesize over potential causes and collect sufficient evidence to suggest remediation steps.