© 2005 by Microfinance Risk Management, L.L.C., p. 1 Scoring for SMEs ‘SME Financing: Issues and...

Scoring for SMEs

‘SME Financing: Issues and Strategies’State Bank of PakistanLahore, May 10, 2005

Mark SchreinerMicrofinance Risk Managementhttp://www.microfinance.com

Agenda

1. What is scoring?

2. Scorecards

3. Use of scoring

4. Benefits and costs

5. Information requirements

6. Steps in a scoring project

1. What is scoring?• Scoring forecasts risk (% chance, like a weather

forecast) that client will behave ‘badly’ based on historical links between behavior and indicators

• Example risks:– Arrears >30 days– A ‘good’ borrower will not return for ‘repeat’ loan

• Example indicators:– Borrower (Education, Size of family)– Business (Debt/equity ratio, Type of business)– Loan history (Max. arrears previous loan, #

loans)– Bureau history (Arrears elsewhere, Debt/income)– Saving history (Presence of account, Balance)– Loan proposed (Amount, #/Frequency

installments)– Lender (Branch, Loan officer)

What is scoring? (cont.)

• Assumption: Cases today will behave like past cases with similar indicators

• Example: Historically, 18% of dairy farmers had arrears >30 days. Risk forecast for a loan approved today to a dairy farmer is assumed to be 18%

Scoring links risk with indicators

2. Scorecards

Indicator Indicator values Points

A. Sector business Trade Ag., services, manufacturing

0 3

B. Previous loans ‘Repeat’ ‘New’

0 2

C. Most days late, previous loan

None or ‘New’ ‘Repeat’ w/arrears

0 1/day, up to 7

D. Savings account Yes No

0 4

Total score:

Farmer w/savings acct. and no past loans (‘new’)

2. Example scorecard (I)



0 3 3


0 2 2



0 1/day, up to 7 0


0 4 0

Total score: 5

2. Example scorecard (II)



0 3 0


0 2 0



0 1/day, up to 7 4


0 4 4

Total score: 8

Trader, 1 previous loan with 4 days arrears, no savings acct.

2. Associating Scores with RiskScore is converted to risk (probability of

going ‘bad’)

Score Risk (%) Score Risk (%)

0 2 9 30

1 3 10 35

2 5 11 37

3 7 12 41

4 10 13 47

5 13 14 52

6 17 15 57

7 22 16 60

8 24 17 64

2. Scorecard notes

• In real life, 30-50 indicators (not 4)• Scoring for SMEs less powerful than

scoring for credit cards, homes, or cars– Self-employed, not salaried– Subject to greater and more varied risks– Less (or no) collateral– Less complete info. from credit bureaux

• Does not replace ‘traditional’ evaluation by loan officers

• Detects high-risk cases that otherwise slip through

2. Scorecards: Source of weights

• Scope– Generic: Made for ‘general’ lender– Tailored: Made for specific lender

• Source of information– Expert: Experience, common sense– Data: Statistics on past loans

• All have pros and cons

2. Scorecards (cont.)

Source of information

Expert Data

Scope

Generic

Tailored

Acceptance

AccuracyCost

2. Scorecards (Colombia example)

Age of borrower and risk >30 days late

-10

-9

-8

-7

-6

-5

-4

-3

-2

-1

0

Ch

ange i

n r

isk (

% p

oin

ts)

18 22 33 44 50 65Age of borrower

2. Scorecards (Bolivia example)

Longest spell of arrears, last three loans

0

1

2

3

4

5

6

7

8

9

10

Change i

n r

isk (

% p

oin

ts)

0 1 3 5 7 9 11 13 15 17 19 21Days in longest spell of arrears

Next-to-last and second-to-last loans

Last loan

2. Scorecards (Dominican Republic)

Type of business Weight (% pts.)• Taxi –3.6• Corner store –2.1• Fried street food –1.2

• Others 0

• Beauty salon +0.5• Clothesmaking +1.4• Farming +1.7• Construction +2.3• Carpentry +4.0

3. Use of scoring

• Two common views of scoring:– Cannot help at all– Can do all the work

• Reality: Scoring can help some– SME lending is too complex to automate

completely– Cannot replace loan officers’ evaluation of cash

flow, character (cannot rescue lender with big problems)

– Does not save time up-front, instead saves time in collections

– Scoring can reject, but not approve

3. Use of pre-disbursement scoring

After provisional approval, the credit committee applies scoring policy according to the risk forecast:

Reject

Reject

Pass data

Applicant submits application

Loan officer makes traditionalvisit and evaluation

Key-in data andcompute risk forecast

Credit committee makestraditional evaluation

Pass predicted risk

Rejection

Very safeApprove and reward

RegularApprove as-is

BorderlineExtra review

Very riskyReject

Provisionally approve

Lender checks applicationagainst initial screen

0% risk 100% risk

Verysafe Regular Borderline

Pass

Pass

Reject

Veryrisky

3. Use: Effects of contract changesClient: Hasan, M. Case: A 12345Loan officer: M. Schreiner Committee: 01/05/05

Amt. to disburse

Term (months)

Collateral coverage

Forecast risk (%)

Requested terms 1000 10 100 40

Amt. to disburse 900 10 100 38800 33700 29

Term (months) 1000 9 100 378 327 27

Collateral coverage 1000 10 125 39 (% of amt. disbursed) 150 37

200 36

3. Use: Reasons for high/low risk

Client: Hasan, M.Loan officer: M. Schreiner

Indicator Actual value

Ave. historical value

Effect (% pts.)

1. Longest spell of arrears, previous loan9 1.7 +5.8

2. # installments late, previous loan 6 4 +4.23. Loan-officer experience (# loans disbursed)

77 535 +3.7

4. Type of business Carpentry N/A +1.55. Inventory rotation (%) 250 326 +0.36. Installment / Net free cash flow (%) 20 18 +0.1

. . . . . . . . . . . .36. Collateral coverage (%) 350 300 –0.437. Number of employees 3 5 –1.938. Experiencia del Cliente (# meses) 36 14 –2.339. Age of borrower 55 43 –4.4

Forecast risk (%) 23.2 9.3 +13.9

Case: A 12345Date: 6/05/05

3. Use: Loan-officer portfolio summary

79922 03-May 2,106 77 Bad 90 Very risky50973 29-May 1,860 36 Bad 81 Very risky71596 29-May 1,323 25 Good 80 Very risky80816 29-May 1,032 42 Bad 80 Very risky62037 02-May 5,683 28 Good 72 Very risky

. . . . . . . . . . . . . . . . . . . . .

8154 16-May 102 0 Good 4 Very safe94799 07-May 289 0 Good 2 Very safe88687 05-May 35 0 Good 1 Very safe38388 01-May 73 1 Good 1 Very safe

243 764 6.4 9.7% 8.8%

Days late now

Max. arrears Policy band

Average: 32 3.5

Betancourt, J . 26 0Valencia, J . 36 0Fernán, Z. 38 0Sánchez, H. 36 0

. . . . . . . . .

Nuñez, D. 316 0Beltrán, D. 48 0Arboleda, N. 132 2Posada, M. 71 0

¿Good or Bad?

J avela, M. 83 23Next Amt. ($)Case Client

InstallmentsBal. ($)

Observed risk (arrears) Scoring risk (%)

Date: 01/05/05 Def. “bad”: > 30 days lateBranch: FirstOfficer: M. Schreiner

3. Use: Prioritizing collection efforts

Visit later

Collections risk(pre-disbursement score)

Value-at-risk

High

HighLow

Low

Visit now but be gentle

Visit now and be firm

3. Use of scoring: Back-testing

• Does scoring predict accurately?• Back-testing: Before use, check if

scorecard predicted historical ‘bads’• Bolovian example:

– “Bad” = In arrears on last day of month– Scorecard data: 1993 to 2001– Test period: January to August 2002– 36 percent of 8,549 cases were “bad”

• No need to trust that scoring works; with back-testing you can know

3. Use of scoring: Bolivian back-test

Criteria Formula 0 20 40 60 80 100

'Bads' avoided A 3,117 2,967 2,591 2,020 1,259 0'Goods' lost B 5,432 3,904 2,223 1,056 295 0

'Bads' avoided per 'good' lost A/ B 0.6 0.8 1.2 1.9 4.3 #N/ A

'Goods' approved C 0 1,528 3,209 4,376 5,137 5,432'Bads' approved D 0 150 526 1,097 1,858 3,117

% 'bads' avoided 100*A/ (A+D) 100 95 83 65 40 0% 'goods' approved 100*C/ (C+B) 0 28 59 81 95 100% cases rejected 100*(A+B)/ (C+A+D+B) 100 80 56 36 18 0Note: 8549 cases, with 5432 (64%) observed 'goods' and 3117 (36%) observed 'bads'.

'Very risky' threshold

3. Use: Bolivian back-test (cont.)

• Policy: Reject if risk > threshold• If ‘very risky’ threshold = 80% risk:

– Successfully reject 1,259 ‘bads’– Mistakenly reject 295 ‘goods’– Avoid 4.3 ‘bads’ per ‘good’ lost– Avoid 40% of ‘bads’– Approve 95% of ‘goods’

• Managers set policy based on effects of different thresholds

3. Use: Bolivian back-test (PAR)Effect on portfolio-at-risk

2%

3%

4%

5%

6%

7%

8%

9%

10%

11%

J an. Feb. March April May J une J uly Aug.

P ortfolio-at-risk without scoring: A verage 6.7%

P ortfolio-at-risk with scoring: A verage 4.9%Scoring reduces portfolio-at-risk by 27 percent.

3. Use: Effect on net profit

• Rejecting ‘very risky’ cases means:– Avoiding some ‘Bads’ (a benefit)

• Lower write-offs and lower loan-loss provisions• Less time spent in collections

– Losing some ‘Goods’ (a cost)

• Suppose:– Benefit of avoiding 1 ‘bad’ = $150– Cost of losing 1 ‘good’ = $150– In reality, cost of ‘bad’ > benefit of ‘good’

• Effect on net profit in Bolivia, 80% threshold: = ‘Bads’ avoided x benefit – ‘Goods’ lost x

cost = 1,259 x $150 – 295 x $150 = +$144,600 (in 8 months)

3. Use: Effect on portfolio size

• In Bolivia (and elsewhere), loan officers spend each week about:– 2 days on evaluation/disbursement– 1 day on meetings and office work– 2 days in collecting from delinquents

• Scoring reduces ‘bads’ by 40%, freeing up 6.4 hours/officer/week

• If extra time used to attract new borrowers, lending grows by 22%

• Thus, scoring decreases arrears and increases portfolio size (and depth)

3. Use of scoring: Test, test, test

• Make sure scoring works (and that users see that it works)– Test before use (back-testing):

• Regardless of scorecard type• Use to set policy

– Test during use:• Detect branches that ignore scoring policy

and users who ‘cook’ data• Control overrides• Know when to refresh scorecard

4. Benefits of scoring

• Finance is risk management; scoring improves risk management– Quantifies risk as probability something

‘bad’ will happen– Makes risk evaluation explicit, consistent

(not just loan officers’ ‘gut feeling’)– Helps train new loan officers and reduces

the need for highly qualified officers

• Better risk management means larger portfolios, more clients (and more poor clients), and greater profits

4. Benefits of scoring (cont.)

• Focus evaluation where it counts most:– Reward ‘very safe’, do nothing for ‘average’,

adjust ‘risky’, reject ‘very risky’• Manage risk after disbursement:

– ‘Preventive’ courtesy visits to ‘risky’ clients– Prioritize collections

• Facilitates portfolio management:– Precisely loosen/tighten credit policy– Strengthen centralized control– Foresee effects of new policies via back-testing – Detect shifts in risk profile before crisis hits– Set prices based on risk

• Biggest benefit: Strengthen culture of explicit, conscious risk management based on quantitative analysis of data

4. Costs of scoring

• Sharp organizational change:– Change is never easy– Power shifts from Credit to IT and from

loan officers to scorecards– Not a project but a (long-term) process– Must manage like any large change

• Users must believe scoring works. To believe, they must understand. To understand, they need demonstrations, training, tests, and follow-up

• Training and follow-up is the key!

4. Costs of scoring (cont.)

• Management skill is crucial. Scoring:– Does not approve nor reject– Counts characteristics, ignores character– Predicts risk, but does not manage it– Supposes the future will be like the past,

but everything changes (economy, competition, credit policy itself)

– Requires careful data collection• Change in habits of front-line workers• Store data received from credit bureaux

– Requires tracking overrides– Requires a local ‘Scoring Manager’

4. Costs of scoring (cont.)

• Integrate in MIS (or use on paper)• Hire external skill if needed• Public-relations risk:

– Seems ‘inhuman’– Makes rejected ‘goods’ explicit– Avoid illegal indicators (e.g. marital status)

• Acceptance by front-line users– No ‘extra’ work due to scoring– Demonstrate predictive accuracy– Persistent training and follow-up

5. Data requirements

• General guidelines:– Do not throw data away– Key-punchers are inexpensive– Record data on rejects also– Train loan officers and key-

punchers– Quantify loan officers’ judgments– Audit data quality– Need data on many past loans– Start to improve data now!– In interim, use ‘expert’ scorecard

5. Data: Credit bureau

• Arrears in current and paid-off loans• Number of inquiries in past year• Amounts owed to current creditors• For current and paid-off loans:

– Identity creditor– Date disbursed (and dates paid-off)– Amount disbursed– Monthly installments– Maximum line of credit

• Both positive and negative info.

5. Data: Loan officer subjective judgments

• 5-point scale (far below-average, below-average, average, above-average, or far above-average):– Overall credit risk– Honesty and transparency of responses– Quality of references– Entrepreneurial skill and ‘oomph’– Prospects for line of business– Variability and seasonality of cash flows– Recent investment in home and business– Grasp of rules in loan contract– Quality of family relationships and informal

support networks

5. Data: Applicant demographics

• Year of birth• Gender• Marital status (and year changed)• Last grade completed in school• # household members (including

applicant) > age 18 or older• # household members < age 17• # household members with salaried job• # enterprises run by HH members

5. Data: Contact particulars

• Phone number to contact at home:– Is it a neighbor’s phone?– Is it a cell phone?

• Phone number to contact at business:– Is it a neighbor’s phone?– Is it a cell phone?– Is it different from home phone number?

• e-mail address

5. Data: Household assets

• Home tenure (owner, renter, other)• Year took up current residence• Number of rooms (excluding bathroom, kitchen)• Hectares of non-homestead, arable land owned• Roof and floor construction materials• Household services

– Electricity– Drinking water– Cooking fuel– Type of toilet

• Vehicles (working)– Automobile, tractor, truck, or bus– Motorcycle– Bicycle

5. Data: Household assets (cont.)

• Appliances (working)– Refrigerator– Color television – Washing machine (clothes)– Context-specific?

• Formal savings account– Date opened– Passbook or time– Balance

• Freq. of receipt of remittances

5. Data: Business ‘demographics’

• Sector (manufacturing, services, trade, agriculture, other)

• Specific type of business (list of 30–50)• Year founded• Formalization• Tenure of locale (owned, rented, other)• Person-months of full-time-equivalent

workers per year– Applicant– Family members (excluding applicant)– Non-family members

5. Data: Monthly business fin. flows

• Sales• Expenses• Installments paid on other debts:

– Business– Household

5. Data: Business financial stocks

• Cash and savings-account balances• Inventory• Fixed assets• Debts

– Formal– Informal

5. Data: Repayment record

• For each scheduled installment:– Date due– Date paid-off

• Identity of the lender:– Branch– Loan officer:

• At disbursement• Current

5. Data: Aspects of loan contract

• Date:– Application submitted– Loan disbursed– Loan paid-off

• Amount:– Disbursed– Average installment

• # installments• Frequency of installments• Refinanced status• Type of guarantee(s)• Value of guarantee(s)• Identity of co-signer(s)

6. Steps in a scoring project

1. Make sure you are ready2. Define ‘Bad’3. Plan to improve data quality4. Build scorecard5. Integrate scoring in MIS6. Pilot7. Expand, monitor, maintain

1. Make sure you are ready

• Efficient, stable lending technology?• Adequate MIS and data base?• Upper management champion?• Manage like any large change project:

– Form a strategic plan– Convene an Advisory Committee– Involve people from Credit and IT– Designate a ‘Scoring Manager’

• Build-in feedback processes

2. Define ‘Bad’

• When do arrears become ‘costly’?– When do arrears preclude repeat loans?– Read ‘Credit Manual’, but also talk with

loan officers and examine incentives

• Choose a risk that policy can affect:– Not borrower death– Loans become unprofitable long before

they become uncollectible

• Start simple, but think big. Get a simple scorecard accepted before adding more or increasing complexity

4. Build scorecard

• Make your own or hire expertise• If possible, use data-based

weights; else, use expert weights

• Ask users to review weights• Regardless of scorecard type,

always back-test to see if it works

5. Integrate scoring in the MIS

• Goal: Make scoring easy to use– MIS computes and displays forecast

risk and realized risk in all standard screens and reports familiar to users• Grafts scoring onto the daily routine• Avoids ‘extra’ work for user

– Helps managers, loan officers see that scoring works

– Tells why a case has high or low risk– Users will look for excuse not to use

6. Pilot

• Write ‘Scoring Manual’ with policy based on back-testing

• Train loan officers, branch managers– Concepts, results of back-testing– Scoring in MIS screens and reports– Affects on performance incentives

• Pilot in 2–3 branches for 6–12 months– Hold hands on all cases for a few days– Provide weekly (then monthly) follow-up– Lasts for an average loan cycle

• Plan formal opportunities for feedback

7. Expand, monitor, maintain

• Based on feedback, adjust MIS implementation and ‘Scoring Manual’

• Second round of training in branches– Testimonials from pilot users– Show lower arrears, higher profits

• Weekly follow-up (then monthly)– Track overrides– Constantly test predictive power

• Manage stubborn branches, officers

Summary: Challenges of scoring

• Keys are data, training, and project management, not computers and statistics– Users must see that scoring works

and does not make them work more– Scoring is not a project but a

process with profound changes– MIS integration helps win over

users

• Always back-test

© 2005 by Microfinance Risk Management, L.L.C., p. 1 Scoring for SMEs ‘SME Financing: Issues and...

Documents

Transcript of © 2005 by Microfinance Risk Management, L.L.C., p. 1 Scoring for SMEs ‘SME Financing: Issues and...