Credit scoring is a set of statistical tools to objectively assess credit risk of consumers. Application credit scoring can help determine the probability of default of a consumer based on the application form parameters when he/she applies for credit. A credit scoring model is optimized to maximize profits from lending operations. Credit scoring helps a bank (or any credit granting firm) reduce its defaults, improve process efficiencies & approval times and move to an objective and consistent method of making credit decisions.
A sample size of about 3000 customers (1500 good customers and 1500 bad customers) can give a very robust credit scoring model. Data on rejected applicants is also vital for building the scoring model. Usually, a base of 1500 bad customers requires a large base of overall credit population. Good scoring models (with restricted scope) can be built even with as many as 250 bad customers.
If an organization doesn’t have sufficient past data on credit performance and no credit bureau information is available, scoring models of different kind can be built. These are expert judgment models which extract the business rules and the intelligence implicit the lending practices of the bank’s analysts. These models are less accurate but can still attain objectivity and consistency in lending practices.
Scoring models need to be optimized for the business at hand incorporating the opportunity loss on a rejected consumer who might have been good and the credit loss on accepting a consumer who may be bad. Once a scoring model is implemented, approval/acceptance rates of consumer applications can be predicted with fair degree of confidence. A scoring model can be optimized to maximize profitability of the business or meet a given acceptance rate requirement.
Few niche analytics service providers such as Fractal Analytics provide modeling solutions that enable businesses to develop credit scoring models. They are known to use cutting edge mathematics and technology, and back it with exhaustive data models built off bureau and transaction data.
A credit scorecard should be tracked closely for its stability and its ability to distinguish between the less risky and more risky consumers. A risk analysis and reporting system can perform model tracking and give warning signals when the underlying population or business practices are undergoing a change. A model should be recalibrated at this stage. Analysis and reporting system can also help to assess the portfolio concentration and performance and estimate portfolio profitability.
A credit scoring solution is built on a modeling population. This model when ‘tested’ with external data (“out of sample”) and simulate the savings that can accrue to the firm on using the scoring model instead of judgment based credit decision. This is the acid test for model performance. After the model implementation (either as a “challenger” or a “champion”) – actual benefits can start to accrue to the firm over a period of time.
Tags: Analytics Models, Credit Risk, Credit Risk Analytics, Credit Risk Management, Credit Score, Credit Scorecards, Credit Scoring, Data Analytics, Fractal Analytics, Risk Analysis, Risk Analytics, Risk Modeling, Scoring Models
