Alternative Credit Scores for Small and Medium Enterprises

This study presents insights from an alternative credit scoring system, developed to address the shortage of data for informing credit decisions for the small and medium enterprise segment.

Background
In developing countries where credit bureau information is often not available or unreliable, transaction-based lending models allow ‘good’ applicants to demonstrate their quality. The study aims to tackle the persisting shortage of data faced by fraternity of researchers and policymakers in the SME sector. In order to do so, this study developed a credit scoring system that is based on combination of customer transactions and characteristics-based screening. It has been observed that such a combination of personal and business activities provides more relevant data for credit scoring. Behavioural factors are a rich source of information and when computed and used optimally with the history of daily transactions of the accounts, a better measure of creditworthiness can be obtained. Keeping these views in mind, it was hypothesised that certain credit behaviour, driven by underlying business operation and financial health, are relatively stable over time and may serve as a better predictive tool for default risk.
Approach
The dataset for the study consisted of 27,386 small business accounts that received limits between January 2005 and March 2008. Basic transaction-level data was used to construct a behavioural profile of each account that is inclusive of transaction, loan utilization and repayment characteristics.
The product used for the purpose of study was a revolving credit product from a major retail bank in India, which allowed businesses to keep negative balance on their existing current account. The transaction level data contained the following extractable information: 1) date the transaction was made, 2) amount of transaction, 3) withdraw or deposit, 4) type of transaction (cash, cheque, transfer), and 5) bank-induced or customer-induced. To supplement the transaction level data, bounced cheque history was also taken into account. Payment history was set in the transaction history and delinquent status was assessed every 30 days. Once interest charges reached 90 days past-due, the account was considered as a NPA.
Behavioural factors that were examined in the study included: transaction level behaviour, account balance parameters and delinquent behaviour. Transaction level behaviour refers to the size and types of transactions an account usually made encompassing total number of transactions, average transaction amount, share of transactions from cash and number of bounced cheques. For each parameter, two values were calculated, one for withdrawal and one for deposit.

Key Findings

From the results, it can be concluded that utilizing the behavioural model to make renewal decisions leads to a significant improvement in portfolio performance regardless of what filtering criteria the account is sanctioned under.

Implications

Such a model can serve as an alternative to close the gaps in client credit scoring models. Similarly, the findings may improve the application of credit scoring on SME lending, especially in developing countries, where consumer credit reports and business financial statements are not easily available and allow more effective distribution of capital to creditworthy customers.

Thematic area

MSME & Entrepreneurship

Project Leads

Antoinette Schoar

Location

Pan India

Partners

ICICI Bank

Status

Completed