CREDIT SCORING FOR SME

on

SMALL AND MEDIUM ENTERPRISES (SME)

  • However, about half of formal SMEs do not have access to formal credit. Access to finance is the key constraint on SME growth compared to large firms.  
  • To address this, the World Bank has taken the initiative to support innovative finance and improved credit infrastructure.
  • This has resulted in generating increased level of fund for SMEs in the last decade.


SME LENDING

  • In the context of growth in SME financing, there is a greater need on the part of lending institutions to mitigate risk.
  • SME credit analysis produces a score indicating the probability of default of the borrowing entity.
  • Having a credit rating not only helps SMEs go beyond obtaining a loan but also provides leverage to negotiate with suppliers for procuring durables, equipment, and raw materials. 


SCORECARD DEVELOPMENT METHODS   

  • Expert scorecard : Human subject matter expert
  • Classical methods : Generalized Linear Model (GLM)
  • ML models : Random Forest, Boosting
  • Deep learning : Neural network for classification

CASE STUDY 1

  • A Bank in Indonesia had a newly acquired SME portfolio.
  • This was a key component in scaling up operations in accordance with Indonesian government directives.
Objective
  • Application scorecard to facilitate SME loan origination decisions
Solution
  • The history of default was not well established with sparse data since the SME portfolio was new. 
  • The bootstrap method was used to overcome the limitation of a small sample.
  • Business risk, financial risk, and moral risk were considered in the model.
  • Reject-inference was successfully employed.
  • High model performance achieved 
Benefits
  • Predictive modelling : replaced gut feel
  • Efficient process : from instant scoring
  • Consistent decision : from adopting a mathematical model

CASE STUDY 2

  • A lender in the UK finances a lease of office equipment to SME
  • Rapid depreciation of leased assets does not make them a good choice for collateral.
  • The lender used to cherry-pick customers who seldom went bad.
  • The finance company wanted to expand its customer base while mitigating risk.
Objective
  • A scorecard to replace rule driven underwriting for better screening
Solution
  • Various default criteria were tested.
  • Use of bureau data representing the status of SME (liquidation, insolvency, dissolution) to construct the default.
  • Separate scorecards for full-account companies and micro entities
  • Checked the quality and coverage of two different bureau data to assess the indeterminates (Not high or low score so no clear indication of Good or Bad)
Benefits
  • Data-driven model : replaced rule-based model
  • Reduce underwrite workload : by restricting scrutiny to fewer applications which get mid-level scores and not extreme scores
  • Separate scorecard for micro entities : ensured these entities are not penalized for lack of accounts data

CASE STUDY 3

  • A multi-finance company in Indonesia provided loans to SMEs.
  • The company did not have a fully developed data warehouse and loan origination system for SMEs.
  • The bank had a policy of strict data security, whereby no data was to leave their server for analytics work.
Objectives
  • Multiple scorecards for various portfolios
  • Link the scorecard to the core banking system to produce an instant score at the time of application.
Solution
  • The data processing and modelling exercises were accomplished in-house with proprietary software for credit scoring. 
  • Separate scorecards were developed for four portfolios
  • In the absence of a core banking system, webservice was created to input data and receive instant score. 
  • A database was created to store newly entered application data.
Benefits
  • Instant scoring of new applications : improves efficiency in decision-making.
  • Process automation : ensured consistency, decreased manual work, and increased accuracy
  • Monitoring of score performance : from interactive reports generated from customized software

Comments

Post a Comment