Anirudh Ramaswamy

Portfolio Monitoring for GPL

• Creation of a weekly acquisition chart view to understand the acquisition volumes,approved volumes, approval rates, score quality of newly acquired customers on a weekly basis
• Monitoring using a portfolio month and an intra-month view to look at the Entry, B1-B2 roll rates along with other portfolio level delinquency metrics like 30+,60+ and CO on a report month basis
• Creation of a daily roll rate view across buckets to understand any divergence on a roll rate level so that an investigation can be done to understand the root cause
• Monitoring a GPL Vintage curve view which would help understand the vintage level delinquencies at a month on book level across different vintages which would help provide an understanding on the effectiveness of an implemented strategy
  

GPL-PM: Loss Analysis

• Understanding the effectiveness of internal scores vs bureau scores as a combination of short term scores were being used for strategy along with FICO
• Understanding the seasonality of payments for GPL-PM as a significant volatility was observed in terms of the payments by customers from roll rate charts
• Identify customer segments - pockets of deterioration post strategy implementation in order to help them with new segments for their loss forecasting segmentation models.
• Comparison of shadow limits of customers using GPL-PM against the quality of such customers.
• Analysis of FICO degradation of PM customers over a period of time in order to understand if covid has any correlation with the rise in delinquencies
• Understanding of the common subset of customers in GPL-PM and other US - revolve products to get insight into customer behavior across products

High Risk Segmentation

• Deep dive into the application data and bureau data to find variables which would be good early delinquency indicators
• Identifying proxy variables which are not used in strategy in order to obtain a different perspective into the GPL portfolio
• Identifying wallet level variables which could be used as good indicators of performance of customers
• Creation of High Risk segmentation with the help of Decision Trees in order to make sure the portfolio could be tracked across portfolio months
• Monitoring the univariate, bi-variates and overall High risk segments on a portfolio monthly basis
  

CCAR - Regulatory Risk modelling

• Worked on building qualitative and quantitative models for forecasting loans, deposits and non interest revenue items
• Identified and analyzed data provided by Moody's and built basic OLS regression models
• Worked on preparing an exhaustive documentation for all the qualitative models

Fraud Analytics - Application Scorecard

• Analyzed and understood various types of application frauds with the help of modus of operandi used by fraudsters
• Built machine learning based fraud risk models(Using catboost and XGBoost) for application decisioning and devised strategies to enhance capture rate of frauds for credit cards
• Performed UAT/Bulk testing and deployed scorecards in the front end BRE environment
• Implementation of the scorecard resulted in greater efficiency where validating top 20% of risky applications resulted in capturing of around 50% of total frauds

Fraud Alert Tool: Rule Based Approach

• Solution was built to tag customers who tend to have differing personal identifiers in comparison with the details of inhouse customers of the bank
• Rules were created basis fuzzy logic to identify customers with same government identifiers and slightly different names/addresses using fuzzywuzzy package in python
• Served as an additional layer to capture fraudsters who were trying to perform personal identity thefts

Post Issuance Analytics - Credit Cards

• Identified possible set of customers who could be given offers by analyzing the demographic, bureau and liability data
• Prepared a base of customers with possible kind of credit card owners which could be given basis the types of spends made on different spend categories using non CC products like UPI,DC,NEFT etc.

Proof of Concepts - Vendors

• Performed POCs with bureaus like CIBIL,Equifax and Experian in order formulate strategies for enhancing the STPs and digital journey of the bank

Syndicate Fraud - Statistical Algorithm

• Analyzed the modus of operandi of syndicate fraudsters who tended to perform collusive frauds within a specified time period of a month
• Prepared a statistical approach (using python)which was run on an onprem GPU to analyze the deviations in volumes of customers applying for a loan across combination of different demographic variables
• Statistical algorithm was run on a fortnightly basis across metro locations which had historically high fraud rates to alert the stakeholder team to identify possible syndicate fraudsters
• Helped in identifying top 5% risky customers basis volume based deviations

• Responsible for driving Devops for reduction in the work􀃫ow time of development process
• Worked on building the backend API and frontend dashboard for the network analytics dashboard
• Developed widgets for an open source DevOps dashboard in Verizon's security product portfolio
• Involved in new DevOps initiatives and solutions as a part of DevOps council in VES portfolio