Analytics to Assess Credit Risk, with Small Samples

Auditorium

Organisational and business context (3 min):

  • New lending product in a “risky” segment
  • Approach is to provide access, but price-for- risk

Analytical Problem (2 mins):

  • Large amounts of customer data, from traditional and emerging sources
  • Very young lending program. Only 1000 disbursals, not “bads” as yet
  • How to estimate risk even as the data is thickening?

Data Structure (5 mins):

  • Specific data from the lending program
    • Nearly complete set of customer features at the time of booking: application form, credit bureau, “big-data” footprint through mobile phone, employment details, bank statements etc.
    • Partial outcome data: missed payments but no defaults.
  • Surrogate data from the credit bureau
    • Nearly complete outcome data: actual defaults on all organized sector loans.
    • Restrictions on data access, must be processed at the bureau.
    • Limited data on customer features: “big-data” footprint, bank statement insights, employment details etc. are not available at the bureau.
  • Adding new data every month
    • Missed payment/ default outcomes, and qualitative diagnostics provided by domain experts.

 

Analytic Approach #1 (6 mins)

  • Build model on surrogate outcomes by merging our partner’s data with credit bureau data, to the
    extent data sharing was possible.

 

Analytic Approach #2 (6 mins)

  • Build model on our own partial outcomes, using data sources from app form, credit bureau, “big- data” from mobile phone, summarized bank statements etc.

 

Social media & sharing icons powered by UltimatelySocial