Transcription
Application of SAS !Enterprise Miner in Credit Risk Analytics Presented byMinakshi Srivastava, VP, Bank of America1
Table of Contents Credit Risk Analytics OverviewJourney from DATA to DECISIONSExploratory Data AnalysisDevelop Quantitative ToolsValidate and CompareBest Strategy EvaluationSAS Enterprise Miner – additional benefitsChallenges of Credit Risk IndustryQuestionsReferences2
Credit Risk Analysis OverviewCredit Risk Analysis (CRA) is integral to every step in the credit lifecycle process,from prospect and customer segmentation, through origination scorecards, to thedesign and execution of account management and collection strategies, whether formortgages, personal loans and lines of credit, credit cards, educational loans, autoloans, and other consumer finance vehicles.CRA is about identifying and mitigating riskassociated with financing credit product tocustomers. Risk team quantify risk, monitor andreport risk of prospect or customer bydevelopment of risk monitoring tool, scorecardand models.CRA guides us in taking decision on customer’sassociated risk for pricing as well creditexposure decision associated with it.3
Journey from DATA to DECISIONIn Credit Risk Analysis,Team explore credit bureau data tounderstand and gather informationabout customers [Exploratory DataAnalysis]Analyze raw data to synthesize theknowledge and develop quantitative risktools [Develop Quantitative Tools]Validate best tool by comparing resultsin different time and scenarios. [Validate& Compare ]Integrate tool into strategy for creditdecisions by evaluate best strategy[Best Strategy Evaluation]Credit Industry heavily relies on analyzing standard credit scoring and customerprovided information for credit decisions.4
Exploratory Data Analysis (EDA)SAS Enterprise Miner is theSAS solution for data mining.In EDA phase, risk team gathersinformation to get familiar withstructure of data and identifyinitial drivers of risk. SAS Enterprise Miner providesseveral data exploration nodes Graph Explore node : explore data graphically to uncover patterns and trends. Stat-Explore node: generates summary statistics and can examine variable distributionsand statistics. It contributes to the initial analysis before getting deep into causation analysis. Multi-Plot node: explore data graphically to observe data distributions and to examinerelationships among the variables [i.e. By Bar graph or scatter plot]. Variable Selection: identifies initial input variables useful for predicting the target.Note: These nodes eliminate the need to write many lines of code in PC SAS to accomplish the same results.5
Exploratory Data Analysis – contd.Sample Bureau Data*Let’s analyze a sample bureau data withpeople demographics, payment history,length of credit, type of delinquency etcwhere target** is STATUS ‘OK’ or ‘BadDebt’. We started exploring data bygenerating descriptive statistics, bar charts,scatter plots for variables as well as howtarget is related to other variables.* Data is for illustrative purposes only ** Target is term used for response variable.6
Exploratory Data Analysis – contd.In the exploration nodes ofEnterprise Miner , visualizationtools are useful in graphicallyrepresenting the distributionsof target vs. other variables.* Data is for illustrative purposes only.7
Develop Quantitative ToolsIn developing quantitative tools, SAS Enterprise Miner provides us numerous tools &techniques to identify top predictors . Selection of final variables can be done through different approaches [i.e. VariableSelection, CHAID Analysis or Regression selection procedure stepwise, forward or backward]. Variable Clustering is useful in identifying variables from groups of highly correlatedvariables. In the Risk Industry, quantitative tools are used in developing predictive models orscorecard or strategies.8
Develop Quantitative Tools contd.The traditional form of a credit scoring model is a scorecard.In Credit Risk Analysis, scorecard plays a key role in decision making.Team uses different types of credit information to calculate the FICO score for the generalpopulation.9
Develop Quantitative Tools contd.SAS Enterprise Miner development of scorecard takes following steps Careful selection of best attributes with high information values Binning of variables and then grouping bins variables [Interactive grouping node] Modeling of approved credit accounts (‘Accepts’) [Scorecard node] Building scorecard on accepts as well as inference performance of “rejects”(reject Inference node).10
Validate & CompareWith SAS Enterprise Miner , it ispossible to create, validate and comparea variety of model types such asregression, scorecards, decision trees orneural networks. When we evaluatewhich model type is best suited forachieving our goals, we consider criteriasuch as Parsimony (complexity) Integration efficiency AccuracyThe Score node functionality ifEnterprise Miner facilitates scoring. Italso generates SAS codes for outsidevalidation.11
Validate & Compare contd.Model Comparison output provides model statistics to compare and assist indecision making process.12
Best Strategy EvaluationIn the final step of Credit Risk Analysis: The Risk team compares and evaluates newly developed strategies/modelswith existing strategies. Validate strategies in different scenariosSAS Enterprise Miner offers number of benefits in best strategy evaluation13
SAS Enterprise Miner – additional featuresSAS Enterprise Miner is the SAS solution for data mining. Easy handling of huge amount of data, no sampling requiredSeveral nodes for customization and exploration of raw data for faster data analysisVariety of model types such as scorecards, regression, decision trees or neural networksTesting new ideas and experimenting with new modeling approachesSpecialize nodes to meet industry specific need and standard regulationProvides required documents and graphs for governance reviewEasy graphical representation of complex quantitative analysis for senior leadersScoring code in many programming languages for easy and fast technologyimplementation14
Challenge of Credit Risk IndustryIt’s a dynamic and continually evolving industry; it’s sensitive tomacroeconomic environment, government regulation and risk appetite ofcompanies. CRA plays a great role in monitoring and predicting future riskunder regulatory environment.15
Questions?Minakshi Srivastava, minakshi.srivastava@bankofamerica.com16
ReferenceGetting Started with SAS Enterprise Miner sj/64144/PDF/default/emgsj.pdfPredictive Modeling With SAS Enterprise Miner: Practical Solutions for BusinessApplications by Kattamuri S. SarmaData Mining Using SAS Enterprise Miner: A Case Study Approach, Second Edition by SASInstituteCustomer Segmentation and Clustering Using SAS Enterprise Miner, Second Edition byRandy S. Collica17
Predictive Modeling With SAS Enterprise Miner: Practical Solutions for Business Applications by KattamuriS. Sarma 17 Data Mining Using SAS Enterprise Miner: A Case Study Approach, Second Edition by SAS Institute Customer Segmentation and Clustering Using SAS Ente
