22
1 Profitability Based Real-Time Decisioning Credit Risk Modeling and Decisioning May 28-30, 2002 Vijay Desai, Director of Modeling Terrance Barker, Staff Scientist HNC Software

desai_wharton2002

Embed Size (px)

Citation preview

Page 1: desai_wharton2002

1

Profitability Based Real-Time Decisioning

Credit Risk Modeling and DecisioningMay 28-30, 2002

Vijay Desai, Director of ModelingTerrance Barker, Staff Scientist

HNC Software

Page 2: desai_wharton2002

2

Agenda

• HNC and HNC Profitability Predictor Introduction

• Profitability Prediction Algorithm• Going Beyond the Credit Risk Score • Monthly versus Transaction Based Scoring• Questions and Answers

Page 3: desai_wharton2002

3

HNC and HNC Profitability Predictor Introduction

Page 4: desai_wharton2002

4

• Founded 1986• U.S. offices

– San Diego HQ– Chicago– Irvine– Los Alamos– Philadelphia– New York

• International offices– London – Singapore– Tokyo

• 1,000 + employees • Initial Public Offering: June

1995(NASDAQ symbol: HNCS)

• 2000 Revenues $254.9 million

San Diego Headquarters

Who is HNC Software?

Page 5: desai_wharton2002

5

HNC Profitability Predictor Uses HNC Technology• HNC’s enhanced neural network technology

provides core predictive model capability• Transaction data provide additional

information typically lost in summarization• Models use account profiles

– Updated with every transaction – Reflect entire account relationship

• Models and software are designed for real-time decisions using the most recent available information

Page 6: desai_wharton2002

6

Revenue model Expected revenue if account does not attrite and does not fall into collection

Credit risk model Expected loss due to failure to payCombines probability of loss with amount

Attrition risk modelExpected loss of revenue due to a sharp and lasting reduction in balance and activity

Profit Calculator ties results together

Cost computationExpected operation & funding costs, using expense parameters & predicted transact, revolve, & delinquency behaviors

HNC Profitability Predictor Profitability Components

Page 7: desai_wharton2002

7

Profitability Overview

Risk Adjusted RevenueRisk Adjusted Revenue

(Cost of Funds + Operation

costs)

(Cost of Funds + Operation

costs)

less

equals

Profit ForecastProfit Forecast

(Forecasted Revolving Revenue + Fees + Forecasted

Interchange Revenue)

(Forecasted Revolving Revenue + Fees + Forecasted

Interchange Revenue)

(Credit Risk Adjustment + Attrition Risk Adjustment)

(Credit Risk Adjustment + Attrition Risk Adjustment)

less

equals

Risk Adjusted RevenueRisk Adjusted Revenue

Page 8: desai_wharton2002

8

Going Beyond the Credit Risk Score

•Credit risk score not the best measure of future profitability•Multi-dimensional view of customer vital•NPV and ROI based approaches

Page 9: desai_wharton2002

9

��������������������������������������������������������������������������������������������������

��������������������������������������������������������������������������������������������������

�����������������������

�����������������������

������

�����������

�����������

���������������

���������������

�������������������

�������������������

���������������������

���������������������

����������������������

����������������������

�����������������������

�����������������������

�����������������������

0 100 200 300 400 500 600

Avg. ProfitPortfolio organized by Credit Risk score

-50 0 50 100 150 200 250 300 350

Low

High

Cre

dit R

isk

Scor

e

% Bad

Managing Profitability:Using Credit Risk Predictions

Page 10: desai_wharton2002

10

Portfolio organized by Profit Score prediction

-50 0 50 100 150 200 250 300 350

��������

������

������

������

������

��������

���������

��������

���������

�����

50 70 90 110 130 150

Low

High

Prof

itabi

lity

Pre

dict

ion

Avg. Profit % Bad

Managing Profitability:Using Credit Risk Predictions

Page 11: desai_wharton2002

11

Most Common Uses

• Profit Score– Retention Queue– Segmentation Variable

• Credit Risk Score:– Minimal Acceptable level of risk– “Knock-out” criteria for Marketing Programs

• Revenue Score:– Needs based approach to credit line increases– Revenue reason codes used for targeting

• Attrition Score– Proactive Retention offers– Profit@Risk metric

Page 12: desai_wharton2002

12

Additional HNC Profitablity Predictor Benefits Will Also Come From...• Revenue Score

– Tracking and Reporting - trending over time– Testing Validation - incremental change over control

position• Profitability Score

– Prioritizing calls in VRU (best customers wait less)– Making it more difficult for unprofitable accounts to talk to a

live representative– Reactive retention queue has on-line access to profit number;

can tailor save offer – Best customers routed to best offer

Page 13: desai_wharton2002

13

HNC Profitability Predictor and Basic SegmentationRisk Revenue Attrition Generic Strategy

Low Low Low DevelopLow High Low GrowLow Low High DevelopLow High High DefendHigh Low Low ExitHigh High Low MaintainHigh Low High ExitHigh High High Selectively Defend

Page 14: desai_wharton2002

14

Organized by Profit Score NPV Organized by HNC Profit Score

Low

High

A Comparison of Profit Score v/s NPV

-1000 -500 0 500 1000 1500

1

2

3

4

5

6

7

8

9

10

Revlrs

Trans.

Other

-600 -400 -200 0 200 400 600 800 1000 1200

1

2

3

4

5

6

7

8

9

10

Revlrs

Trans.

Other

Page 15: desai_wharton2002

15

True ROI Organized by Predicted ROI Organized by HNC Profit Score

Low

High

A Comparison of Profit Score v/s ROI

-1 -0.5 0 0.5 1 1.5 2 2.5 3

1

2

3

4

5

6

7

8

9

10

P_ROI T_ROI

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3

1

2

3

4

5

6

7

8

9

10

Revlrs

Trans.

Other

Page 16: desai_wharton2002

16

Monthly v/s Transaction Based Scoring

Page 17: desai_wharton2002

17

Transaction Data Improves Predictive Power

0

10

20

30

40

50

60

70

80

0 2 4 6 8 10

% Goods

% B

ads

ProfitMax Monthly

Page 18: desai_wharton2002

18

Transaction Based Neural Net Score is Stable Over Time

Month 1 Month 2 Month 3 Month 6 Month 9%units Avg. Scr.%units Avg. Scr.%units Avg. Scr.%units Avg. Scr.%units Avg. Scr.

1st Pro 100 545 79.5 543 74.2 543 68.8 544 66.1 537QuartileMonthly 100 553 60.1 552 58.2 551 63.1 556 60.7 5512nd Pro 100 614 65.5 614 60.0 616 50.2 618 46.4 622QuartileMonthly 100 610 47.5 609 46.5 609 45.0 618 43.8 6203rd Pro 100 654 62.7 657 55.7 659 44.0 663 40.7 667QuartileMonthly 100 657 48.2 645 47.0 645 35.2 654 31.8 6544th Pro 100 707 84.2 709 78.4 712 71.9 715 66.8 717QuartileMonthly 100 707 77.1 688 76.8 688 69.3 691 67.7 681

Page 19: desai_wharton2002

19

Worst 20% of Accounts

Benefit Over Monthly Score Increases with Time

* 31 basis points higher than Monthly score

Dollar Based Performance Spread over Monthly Score10 Months12 Months14 Months19 Months

C/O 31* 65 97 191BKO 44 45 43 96Ever 95+ 30 88 114 185Ever 65+ -14 22 39 100

Page 20: desai_wharton2002

20

Best 30% of Accounts

Benefit Over Monthly Score Increases with Time (contd.)

* 6 basis points lower than Monthly score

Dollar Based Performance Spread over Monthly Score10 Months12 Months14 Months19 Months

C/O 6* 11 15 32BKO 8 9 10 20Ever 95+ 13 27 35 51Ever 65+ 18 30 37 55

Page 21: desai_wharton2002

21

Benefit by Scoring and Decision Points Comparison of Dollar Savings

$0.00

$1.00

$2.00

$3.00

$4.00

$5.00

$6.00

0.50% 1.00% 1.50% 2.00% 2.50% 3.00% 3.50% 4.00%

Percentage of Accts W orked per Month

Ann

ual D

olla

r Ben

efit

per A

ccou

nt

Cycle-Cut Real-Time Daily Batch

Page 22: desai_wharton2002

22

Questions &Answers