credit scoring and credit control xi, edinburgh 2009€¦ · © 2009 experian limited. all rights...

© 2009 Experian Limited. All rights reserved. Experian and the marks used herein are service marks or registered trademarks of Experian Limited. Other product and company names mentioned herein may be the trademarks of their respective owners. No part of this copyrighted work may be reproduced, modified, or distributed in any form or manner without the prior written permission of Experian Limited.

[email protected]

Credit Scoring and Credit Control XI, Edinburgh 2009

Scoring Models for Collections & Debt Recovery

© 2009 Experian Limited. All rights reserved. 2

Introduction

� Why do we need collections / recovery scores?

► Due to difficult financial conditions more people are going into arrears

► Optimising the use of collections resources is therefore key

► The use of collections scores facilitates the automation of a number of collections activities

► Account level scores achieve this reasonably well but a more holistic approach offers a marked improvement

� Targeting the best prospects for debt recovery can also be automated


Agenda

� High Level Overview

� Collections

� Debt Recovery (Charge-Off’s)

� Summary


High Level Overview

� Investigated 2 main areas

► Tools to improve the management of accounts in collections across a range of financial products

► Tools to highlight the best prospects for debt recovery action

� Show how each solution has evolved

� Illustrate the power of consumer level data in collections / debt recovery


Agenda


� Collections


� Summary


Up to Date Accounts Accounts in Arrears

Collections Identifying development samples

X

Credit Portfolio

Outstanding DebtSettled Good’sX


Collections Objectives

� Determine the probability an account in arrears will make further payments in order to determine the most appropriate collections path

� Objective function is therefore:

► Goods: At least one (minimum) payment received in the 3 months following observation and account does not default during outcome

► Bads: Not Good

� For Communications accounts the good / bad definition is generally harsher to reflect the industry view of accounts in arrears


CollectionsSegmentation

� For the active element of the portfolio which is in arrears there will be a range of profiles that make applying a single solution difficult

� These range from slow payers who normally are up to date to accounts in serious financial difficulty

� These profiles will behave differently across credit products

� Further segmentation is therefore required to build an effective collections score

► You don’t want to put a good payer who has simply forgotten to pay the bill through an intensive collections path


Collections Segmentation

� Early collections includes “target accounts” 30 days past due

� Mid collections would be 60-90 days past due (60 dpd for Communications)

� Late Collections are 120 days past due (90 dpd for Communications)

Accounts in Arrears

Early Collections Mid Collections Late Collections


CollectionsSample profiles

0.120.411.04Communications

0.371.025.08B&F (Unsecured)

Good / Bad Odds

1.714.8515.40B&F (Secured)

0.982.4012.24Retail

LateMidEarly


CollectionsCombining client / consumer data

Target Account Data

Consumer Data (Other Lenders)

All Accounts in Arrears



CollectionsKey predictors – Financial services

Financial Services - Key Predictors

0.00

5.00

10.00

15.00

20.00

25.00

30.00

0 1-10 11-20 21-30 31-40 41-50 51-60 61-70 71-80 80+

Consumer Indebtedness Index

GB

Odd

s

Early

Mid

Late


CollectionsBuilding a solution

� Results from Bureau only contribution

Performance of Collections Scorecards

70

50

68

59

0

10

20

30

40

50

60

70

80

90

100

B&F Unsecured B&F Secured Retail Communications

Sector

Gin

i Coe

ffici

ent

Bureau Data


CollectionsCombining client / consumer data

Target Account Data

Consumer Data (Other Lenders)

Target Account Data + Key Bureau

Variables

All Accounts in Arrears


Vertical Market Niches




CollectionsKey predictors

Observation Balance

0.00

2.00

4.00

6.00

8.00

10.00

12.00

Q1 Q2 Q3 Q4 Q5

Quintile

GB

Odd

s B&F

Mortgages

Retail

Telcos


CollectionsPerformance

� Results from Combined Bureau + Target Account contribution


70

50

68

59

76

54

75

65

0

10

20

30

40

50

60

70

80

90

100

B&F Unsecured B&F Secured Retail Communications

Sector

Gin

i Coe

ffici

ent

Bureau Data

Combined T&B


• The large uplift over the generic behavioural score illustrates the importance of utilising a score based on collections specific predictors & performance


76 75

54

65

5054

19

30

0

10

20

30

40

50

60

70

80

Banking & Finance Retail Mortgage Communications

Sector

Gin

i Coe

ffici

ent

Collections

Behavioural

CollectionsComparison to generic, all purpose behavioural scor e


CollectionsLimitations of generic collections scores

� Clearly collections specific scores are much more effective than using a standard behavioural score

� Bureau data is very powerful when combined with account level data for use in collections

� However, generic collections scores cannot take into account

► Client specific account performance

► The effectiveness of each clients own collections departments


CollectionsBespoke collections scoring

� Here the overall objective is to build a bespoke consumer collections score derived on client data

� Combine the power of specific external data with a lenders internal data

� Derived using similar Early, Mid & Late collections segmentation

� In addition to the usual bureau variables a few new ones were generated looking solely at the recent performance of other accounts in collections

► Number of accounts in collections

► Number which have received a recent payment

► Number which have received a significant recent payment


CollectionsBespoke collections scoring - Results

52.945.5All

50.641.5Early

GINI

41.531.7Late

42.833.9Mid

Bespoke Collections Score

Generic Collections Score

• The existing scores performed considerably worse than either the generic or bespoke collections solutions above


Agenda


� Collections


� Summary


Debt Recovery (Charge-Off’s) Objectives

� These are typically non-paying accounts with Debt Collection Agencies (DCA’s) that can also be passed back to the original lender for further action

� Focus on non-paying / dormant debt where the objective is to identify which individuals are likely to repay some or all of the balance owed

� Objective Function is therefore

► Goods: Any payment received in the next month

► Bads: No payment received within the next month

► Exclusions: Forwarding address located


Debt Recovery (Charge-Off’s) Segmentation / sample composition

� Sample of credit defaults from multiple DCA’s

� No segmentation applied for initial investigations

� Sample G/B odds equate to 0.14


Debt Recovery (Charge-Off’s)Combining client / consumer data

Stage 1

Stage 2

Target account information &Default level consumer Variables

Full credit information


Debt Recovery (Charge-Off’s) Key predictors

Number of Paid Up Defaults L12M

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

None 1 2 3+

Number

GB

Odd

s


Debt Recovery (Charge-Off’s) Key predictors

Original Default BalanceLow - High

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th

Decile

GB

Odd

s


Debt Recovery (Charge-Off’s) Score Distribution

12.00%Overall

55.48%10th

34.57%9th

19.52%8th

9.51%7th

4.33%6th

3.32%5th

2.19%4th

1.35%3rd

1.07%2nd

0.57%1st

Propensity to

Pay

Score Decile

(Low to High)

• Initial results produced GINI figures in excess of 75 for a score using combined client and consumer level default data


Debt Recovery (Charge-Off’s) Using a DR Score

• The resulting score can be used to target the top X% of the population for additional debt recovery action

• Such a score can be delivered relatively easily in a batch environment that monitors an individuals credit profile to detect signs of improvement

• Experian’s batch scoring system also indicates where an individual in debt recovery has changed address


Agenda


� Collections


� Summary


Summary

� Account level information is limited in predicting collections (and debt recovery) success

� A more holistic solution provides a significant uplift

� This is even greater when a client specific model is developed

� A bureau based collections / debt recovery approach can greatly simplify the overall solution

credit scoring and credit control xi, edinburgh 2009€¦ · © 2009 experian limited. all rights...

Documents