Shiv Shankar Shukla

Data Analyst @Toluna

IN

Project handled: ANZ BankDomain: Credit riskRoles and Responsibilities: Bad Definition analysis: To find the appropriate bad definition for the tagging the account as bad. Vintage analysis: Find the optimal performance period. Roll rate analysis: Find the optimal delinquency bucket for bad.Started this Business objective using Logistic Regression Model but due to high non linearity is present,Decided to go for ensemble Tree models (Random Forest).Then check the performance byusing different accuracy metrics If customer defaults (90 days or more past due) during the performance window, borrower would be considered as a \'bad\' customer and labeled as \'event\' in dependent variable. Identify if accounts opened in a particular month or quarter are riskier than other Determine the optimal period of performance window in development of scorecard Monitor or track risk of a portfolio Estimate minimum months required after that we can cross-sell to new customers Forecast risk Can be used in stress testing Calculate the cumulative % of bad customers or cumulative charge-off rate (also known as cumulative loss rate) against months on books (MOB) which is the number of months that have completed since the loan origination Date. New Vintage means recently opened accounts. Validated multiple Non-Scoring models for major US banks by evaluating multiple business and statistical metrics across various out of time vintages. Responsible for performing quarterly validation of all non-scoring models and reviewing if the performance is in-line with the development sample. Doing Bad definition Analysis to capture the optimal number of bad accounts.Trend Analysis Average balance for customers opened in different quarters and see their trend in the subsequent months after account opening date.