Assess whether a model update needs a fair-lending revalidation (73e4d8)

Situation

After an exception rate twice as high for one group after credit controls, adverse-action notice principal-reason sample is what adverse-action notice operations lead can touch in a lender expanding into majority-minority census tracts. Fair Lending will live with Remove access or reverse the item versus Temporary compensating control on this Redlining and HMDA Data file.

Decision

Adverse-action notice operations lead in a lender expanding into majority-minority census tracts must choose Remove access or reverse the item / Temporary compensating control / Approve a documented exception / Hold using adverse-action notice principal-reason sample after an exception rate twice as high for one group after credit controls.

Hypotheses to test

• The population in adverse-action notice principal-reason sample is the one an exception rate twice as high for one group after credit controls named, so Remove access or reverse the item follows for this Redlining and HMDA Data file.
• The population in adverse-action notice principal-reason sample is adjacent only to an exception rate twice as high for one group after credit controls; Temporary compensating control is the honest Fair Lending call.
• A lender expanding into majority-minority census tracts already contained an exception rate twice as high for one group after credit controls before adverse-action notice principal-reason sample arrived; no new Redlining and HMDA Data path.
• Provenance on adverse-action notice principal-reason sample after an exception rate twice as high for one group after credit controls is broken; do not pick Remove access or reverse the item or Temporary compensating control yet.

Analysis required

• Compare adverse-action notice principal-reason sample to similarly situated files, second-review notes, and reason codes after an exception rate twice as high for one group after credit controls.
• Flag any disparate-impact table adverse-action notice operations lead cannot explain from adverse-action notice principal-reason sample.
• Test a documented exception versus a pattern a lender expanding into majority-minority census tracts must defend.
• For this Fair Lending Redlining and HMDA Data file, read adverse-action notice principal-reason sample against an exception rate twice as high for one group after credit controls and write the one fact that would move a model update needs for adverse-action notice operations lead.