Credit Risk

Basel compliant regulatory and economic capital models that help lending institutions curtail risk and have more capital open for lending at lower risk or expected loss.

$2B

assets handled

30%

lower default rates

40%

lower LGD

69%

increase in accuracy

Problem Statement

The problem statement was to create a complete end to end credit rating or risk management system from decision model to performance models. Also we needed to create scorecards, expected loss models , simulate unexpected loss, build CCAR models using DGSE approach. Hence the above also included a LGD and EAD model building exercise.

68% decrease in NPA

Integrate data from application for loan, social data (under GDPR norms), and other ancilliary information to more accurately track credit-worthiness

Fed data real time to the applicant/cro/sales manager so as to streamline the decision making process

Created intuitive scorecards for risk assessment and offer generation

Created simulations of economy with different interventions and shocks based on a derivation of Smets-Wouters model which helped to do risk assessment of bank lending porfolio and helped model unexpected losses better.

Continuous learning over time through reinforcement techniques and feedbacks to improve the accuracy of the entire AI -Credit Risk deployment

The
Finarb
Formula

Algorithm

The models presented so far can be organized into three categories. The first category applies already existing models such as decision trees and rule induction, neural networks, fuzzy modeling, support vector machines (SVMs), k-nearest neighbors (k-NN), Bayesian networks (BNs), instance-based algorithms, learning classifier systems, ANN and SVM. The second category proposes a new hybrid model based on the existing models. The third category involves the use of more complicated deep learning models. Apart from above which are used Iin PD models we also use simulation based models, MCMC models in stress testing, etc.

Data Sources

Credit data includes complete application data, loan performance data, cibil/fico data, social data, loan performance data, economics data. The above data is essential. What was an addon to this was news information, customer support information and any other data on customer interaction. As we can see that the above data belongs to separate sources and types, robust data pipelines was set up on open source technology in private infrastructure to serve all AI and BI needs.

Platform Agnostic

Based ona wide variety of users the choice of backend and front end technology was made such as to avoid rework and promote reuse. The front end was based on react native while the backend involved a Python/R microservices stack

Deployment models

The platform was deployed as a licensed model wherein the user will have to state his/her loan preferences and based on that will be made an offer. Post that user can accept and reject the offer and then proceed with the necessary norms and regulations

Integrations

The entire solution was integrated to older legacy systems like IBM AS400 data and tape data whole was also compatible with the RDBMS/NOSQL data including graph DBs. The api s were integrated to web apps, mobile apps and tools used by CROs

Let's Get Started

Credit Risk

Algorithm

Data Sources

Platform Agnostic

Deployment models

Integrations

Finarb

Solutions

Industries

Follow Us