Credit Scorecard Taxonomy
Contents
Credit Scorecard Taxonomy
There are many Credit Scorecard types. The following categories break them down to the main classes by: usage, type of input, type of output, manner of development and algorithm used (if any)
By usage
Credit scorecards may be used for a number of distinct business activities across the Credit Life Cycle
- Prospect management (prospect scoring), that is, selecting from an existing client base for new product marketing campaigns
- Approval of new credit (Credit Decisioning using an Acceptance Scorecard or application scoring) at the Credit Origination stage
- Assessment of ongoing Credit Risk profile ( behavioural scorecards, internal credit rating systems)
- Collection Scoring, to help select the most appropriate collection strategy in the case of Delinquency and a Non-Performing Loan
- Input for the calculation of Regulatory Capital (Basel III scorecards) for regulated financial institutions
- Input for the calculation of IFRS 9 / CECL Expected Credit Loss and Loss Allowance for institutions reporting under these standards
By type of Data Input
- Most standard scorecards will tap existing Credit Data sources for the required information
- Increasingly, various Alternative Credit Data sources are being used
By type of Output
A scorecard might produce as output:
- a Credit Score, a numerical variable ranging continuously over some range (e.g. 0-1000)
- an actual estimated Probability of Default, e.g 3.5%, which may obtained directly from scoring model or via a Default Probability Table
- a Credit Rating, a Categorical Variable with less granularity that a continuous score, e.g. BB
By nature of development
Broadly speaking, credit scorecards can be divided into two development type categories
- Quantitative Scorecards (Credit Scoring Models) that use exclusively or primarily quantitative inputs and algorithmic processing (Machine Learning) to achieve the risk classification
- Expert Scorecards that are developed by human subject matter experts
In practice quantitative scorecards will admit also expert based adjustments (Overrides) that aim to incorporate additional information not captured by the scorecard but potentially available to the decision maker
By algorithm
Quantitative scorecards employ one of several possible algorithmic classes (For complete list see Credit Scoring Models), e.g. linear discriminant analysis, logistic regression, decision trees etc.