Understanding and Using

Systems for Measuring Loss

Given Default
Greg M. Gupton
(212) 553-1493
[email protected]
m
1
 Topics
• Overview
• Data
• LossCalc Methodology
• Validation and Testing
• Performance Over Time
• Web-based Delivery
• Frequently Asked Questions

2
 Credit Risk Depends Critically
on Loss Given Default...
Both expected and “unexpected” losses vary
with and impact:
– pricing
– provisioning
– capital calculation and allocation
– credit risk measurement & management

Note: LGD, Loss in the Event of Default (LIED), Default severity,

1-Recovery Rate are all equivalent

3
 Topics
• Overview
• Data
• LossCalc Methodology
• Validation and Testing
• Performance Over Time
• Web-based Delivery
• Frequently Asked Questions

4
 Two ways to measure LGD

(Method 1) (Method 2)
Secondary Market Realized Net Present
Quotes on Defaulted Value of Funds
Debt: Recovered:
• Larger firms • Middle mkt. firms
• Syndicated loans • Mid-tier lending
• Corporate bonds • Mostly of loans
• Preferred Stock • Few (if any) bonds,
etc.

5
 LossCalc is Based on Post-
Default Debt Prices
• Prices are good predictors of eventual recovery
• Allows use of larger database of prices
– taken from dealer quotes, data vendors, exchanges,
and other sources
– consistent with Moody’s published studies
– thousands of observations
– includes bonds, loans, & preferred stock
• Prices taken 15 to 60 days after default

6
 LossCalc’s Data Set
We chose 1981-2001 data:
• Better address today’s market
– Example: Pre-packaged Chapter 11 filings
• Extensive data: 1,800+ observations from more than 900 defaults
• Includes two full credit cycles
• Composition:
• Public (50%) and private (50%) firms
• Rated (60%) and unrated (40%) instruments
• Firm size (assets): $5 million to $38 billion
• Obligation size: $680,000 to $2 billion

7
 Topics
• Overview
• Data
• LossCalc Methodology
• Validation and Testing
• Performance Over Time
• Web-based Delivery
• Frequently Asked Questions

8
 LGD Transformations
Beta Distribution Fit to LGD
Relative Frequency

LGD
9
LossCalc Factors: Average Recoveries by Debt/Seniority

Secured Unsecured Senior Senior Senior Subordinated Junior Preferred

Secured Unsecured Subordinated Subordinated
Bank Loans Bonds Stock

10
Traditional “Cutting Edge” LGD Analytics

Source: Gupton, G., “Bank Loan Loss Given Default,” Moody’s Special Comment, November 2000

11
 The LossCalc Predictors
Traditional debt type/seniority lookup:
1) Average recoveries by Debt+Seniority
Firm-specific capital structure:
2) Is there more senior debt above?
3) Leverage ratio: (Total Assets / Total Liabilities )
Industry:
4) LGD index by industry
5) Industry indicator for banking (Low)
Macro economic:
6) RiskCalc PD Index (North American Population)*
7) Leading economic indicators
8) Moody’s Bankrupt Bond Index (MBBI)
9) Trailing speculative-grade default rates
12
 LossCalc Factors

Debt Type & Seniority

Macro Economic

Industry

Firm Specific
Capital Structure

0% 10% 20% 30% 40%

13
 Industries Behave Differently

• Industries enter and exit

recessions at different
points and with different
intensities.
• ‘Consumer Products’
enters after ‘Industrial’
Jan-89

Jul-89

Jan-90

Jul-90

Jan-91

Jul-91

Jan-92

Jul-92

Jan-93

Jul-93

Jan-94

Jul-94

Jan-95

Jul-95

Jan-96

Jul-96

Jan-97

Jul-97

Jan-98

Jul-98

Jan-99

Jul-99

Jan-00

Jul-00

Jan-01

Jul-01

Consumer Products Grand Total Industrial

and doesn’t have the
same upside.

14
LossCalc & Default Rates

15
 Topics
• Overview
• Data
• LossCalc Methodology
• Validation and Testing
• Performance Over Time
• Web-based Delivery
• Frequently Asked Questions

16
 Performance Testing

What we Measured:
• Prediction mean squared error
– Differences between predicted vs. actual losses?
• The correlation of predictions with actual losses
– i.e., did LossCalc rank predictions in the right order?
• Ability of model to predict larger than expected losses

» All taken out of sample

17
 Walk-Forward Testing
Training set of firms taken at t0 Validation set of original firms in
training sample but taken at t1

B Validation set of new firms not in

training sample and taken at t1
1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 Resampling Performance Statistics
Error

0.6
A

0.4
0.2
0.0
B
-2 -1 0 1

Out-Of-Sample Validation

Error 100%
90%

0.30
80%

0.25
70%

Defaults (%)
A

0.20
60%
NN
50%

0.15
Merton
40%

0.10
Altman 5V
30%

Result Set
Altman 4V

0.05
20% Shumway

0.0
10% Zemijewski

}{
-3 -2 -1 0 1 2 3
0%
0% 20% 40% 60% 80% 100%
Population (%)
Error

0.4
A

0.3
0.2
Non-Defaulters Defaulters
B

0.1
Training sample

0.0
-3 -2 -1 0 1 2 3

86 % [5 %] 14 % [5 %]
28 % [5 %] 72 % [5 %]
  Cross-validation sample
85 % [5 %] 15 % [5 %]
  33 % [5 %] 67 % [5 %]
 

Error

0.5
A

0.4


0.3
0.2
0.1
B 

0.0
-3 -2 -1 0 1 2 3

18
 Topics
• Overview
• Data
• LossCalc Methodology
• Validation and Testing
• Performance Over Time
• Web-based Delivery
• Frequently Asked Questions

19
 Responding to the Economy
$80.00

$70.00

$60.00

$50.00

$40.00

$30.00

$20.00

$10.00

$0.00
1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001
Historical Recovery Average Average LossCalc Estimate Historical Lookup

20
 Topics
• Overview
• Data
• LossCalc Methodology
• Validation and Testing
• Performance Over Time
• Web-based Delivery
• Frequently Asked Questions

21
22
 Summary
• LGD is key for understanding credit risk
• LossCalc incorporates 1) macroeconomic, 2)
industry, 3) firm specific, and 4) exposure specific
information
• Significantly outperforms traditional methods
• Provides immediate and 1 year estimates for
exposures with different maturities
• Requires only readily available data
• Based on largest & expanding recovery database

23