Introduction to Modeling Expected Losses

Steven Lackowski

Expected losses are a current estimate of all contractual cash flows not expected to be collected; a model is a set of ideas and numbers that describe a particular state. There are several approaches to expected loss modeling:

Cohort Approach:

  • Consider the loss accumulations period.
  • Group loans outstanding at the beginning of that period by relevant risk characteristics.
  • Measure losses accumulated on each “cohort” over the following loss accumulation period.
  • Average the quarterly results over a loss cycle period for an expected loss rate for each cohort.
  • At measurement date, adjust expected loss rate for current conditions and “reasonable and supportable” forecasts.

Vintage Approach:

  • Track homogeneous loans on the basis of calendar year or quarter of origination.
  • Measure losses accumulated on each “vintage.”
  • Apply the expected cumulative loss to the outstanding vintages.
  • At measurement date, adjust expected loss rate for current conditions and “reasonable and supportable” forecasts.


May be a statistical methodology using quantitative measures to determine PD/LGD or from an external source:

PD – the probability of default for a borrower over a given time horizon: for CECL, the life of the loan
LGD – the loss given default as a percentage
EAD – the exposure at default
EL – the expected loss which equals PD/LGD/EAD

In selecting adjustment factors, the bank should:

  • Use business sense
  • Determine factors that have historically published data.
  • Verify the reliability of future economic forecasts.
  • Determine statistically significant factors.
  • Explain the causation.

Transitioning to CECL will involve:

  • Preparing data: granularity allows for developing more supportable models, data from a full business cycle is more relevant, and policy changes may impact the relevance of certain types of data.
  • Governance: Assemble a committee, develop policies and procedures, discuss the plan with auditors and regulators and plan for staff.
  • Validation: Have an independent validation at least annually or as required by model updates, back-test adjustment factors and models before implementing, and document the theory behind the model.