Financial Innovation and Default Rates

Samuel MaurerHoai-Luu Nguyen

Asani SarkarJason Wei

January 2, 2009

DAY AHEAD CONFERENCE 2009, SAN FRANCISCO

These views belong to the authors, and do not necessarily those of the Federal Reserve Bank of New York or the Federal Reserve System.

2

Motivation

• Measured default rates have been unusually low relative to history and economists’ predictions

Moody's Global Speculative Grade Forecasts

0.01

0.02

0.03

0.04

Jan-0

7

Mar-

07

May

-07

Jul-0

7

Sep-0

7

Nov-07

Jan-0

8

Mar-

08

May

-08

Jul-0

8

Sep-0

8

Nov-08

Def

ault

rat

e (%

)

Moody_dec06 Moody_mar Moody_may Moody_jun

Moody_jul Moody_oct global_spec

3

Why Models Over-Predicted Defaults

• Business cycle (not properly accounted for in models)

• Structural Break in default model relationships

• Omitted variable in default prediction model: financing

• Expanded menu of financing for distressed firms since 2004

– Traditional financing (bank loans; CP issuance)

– Structured financing (High-yield CLO, CDO issuances)• Rescue financing for distressed firms without access to traditional

financing• Substitute loans for bonds, increasing flexibility

– Structured financing vehicles (CLO managers)• Bring in new sources of capital• Major buyers of leveraged loans

4

Growth in Traditional Financing

0

50,000

100,000

150,000

200,000

250,000

90 92 94 96 98 00 02 04 06

CP_ISS

500,000

600,000

700,000

800,000

900,000

1,000,000

1,100,000

90 92 94 96 98 00 02 04 06

CILOAN_OUT

0E+00

1E+10

2E+10

3E+10

4E+10

5E+10

90 92 94 96 98 00 02 04 06

HY_ISS

5

Growth in Structured Financing

CDO Issuance: 1995-2006

0100200300400500600

1995

1997

1999

2001

2003

2005

Year

CD

O is

suan

ce (

$b)

All CDO Issuance

Global Leveraged Loan CLO Issuance

0

50

100

150

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

In $

Bil

lion

Leveraged Loan CLO

6

Implications for Default Rates• Default delayed or avoided, depending on investment opportunities

• Structural model: Default occurs if firm value V below a threshold V*

– Financing may increase V (no change in recovery rates) or

– Lower V* (lower recovery), perhaps due to the greater flexibility of leveraged loans compared to traditional financing (e.g. less covenants or PIK terms)

Standard Merton-type Framework

0

20

40

60

80

100

120

140

160

1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97

Time

Fir

m V

alu

e as

Pct

of

So

lven

cy L

evel

Path A Path B High Threshold Low Threshold

7

The Number of Covenants had been on the Decline

Mean Number of Covenants Per Contract - Senior Unsecured Debt

0

1

2

3

4

5

6

7

8

9

10

11

1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Offering Year

No

. of

Co

ven

ants

Speculative Grade

Investment Grade

8

Analysis: Financing and Default Rates

• Year-rating-cohort analysis: Is there a reduction in the proportion of bonds defaulting early?

• Prediction model for monthly aggregate default rates– Evidence of structural break in most recent sample

• Structured financing, distance to default and default rates

• Traditional financing, distance to default and default rates

9

Results

• Evidence is consistent with “delayed defaults”– Proportion of early defaults are historically low

• Evidence of structural break in most recent sample

• Financing measures significantly related to distance to default and default rates

10

Contributions• Empirically link default rates to financing

– A new channel for endogenous default boundary (flexibility of loan terms)– Chen (2007): simultaneously determines firm’s capital structure and default

decisions– Rajan, Seru, Vig (2008): Mortgage models under-predicted defaults in

recent years due to lowered incentive to monitor– Mian and Sufi (2008): expansion of mortgage supply led to defaults– Keys et al (2008): securitization reduces screening by lenders– Leland and Toft (1996), Anderson, Sundaresan, Tychon (1996):

endogenizes default boundary

• New channel for financial innovations to affect the economy– Lower long-run default rates by reducing the compensation required for

bearing credit risk• Spread credit risk to less risk-averse investors• Reduce macroeconomic and financial volatility

– May have delayed bankruptcy in the short-run by making more distress financing available and on better terms

11

Data

• Moody’s data on annual cumulative default rates by yearly cohort for speculative-grade issuers, 1980-2006

• 3 ways to default– Missed/delayed payment of principal/interest– Legal blocks to timely payment (e.g. bankruptcy)– Distressed exchange

• Cumulative issuer-weighted default rates at the end of year, for bonds outstanding as of the beginning of the year

12

Cumulative Default Rates By Year

13

Low Percent of Early Defaults in Recent Years Relative to Prior Expansion Years

14

Predicting Monthly Default Rates

Outcome variable: Changes in the default rate

Moody default rates: trailing 12-month rates

Change in rates potentially affected by default events up to 12 month past

• Calls for including several lags of explanatory variables

15

Determinants of Default Rate (All Variables in Changes)

12*

12*)*5.0()/( 2

ttt

LVLnDdefault

• Distance to Default• Number of standard deviations of asset growth by which the asset level > firm’s liabilities

V: firm value; L: liability measure (st debt + 0.5*lt debt)

μ: mean asset growth σ: standard dev of asset growth

• Macro conditions•10 year – 3 month term spread•Unemployment rates •Consumer expectations

• Credit quality •High yield – IG credit spreads

• Stock returns

16

Results Distance to default Macroeconomic

Conditions Credit Quality and

Stock Returns Growth in Corporate

Leverage Explanatory Variable

Estimate t-stats Estimate t-stats Estimate t-stats Estimate t-stats

Dependent variable: D Intercept -0.02 -0.71 -0.02 -0.86 -0.06 ** -2.53 -0.09** -2.50

DDEF, Lag1 0.36 1.34 0.18 0.57 0.50* 1.86 0.52* 1.81

TERM, Lag12 --- --- -0.17** -2.10 -0.18** -2.15 -0.16* -1.83

CON EXP , Lag1 --- --- -0.01* -1.88 -0.01* -1.70 -0.01* -1.77 LEV_GR , Lag1 --- --- --- --- --- --- 0.01 1.06

VARIABLES WITH MULTIPLE LAGS UEM, 3 Lags

+, SIG --- 1 1 2 -, SIG 0 0 0

CQ, 10 Lags +, SIG --- --- 6 6 -, SIG 0 0

SRET, 6 Lags +, SIG --- --- 3 2 -, SIG 0 0 12 Lags of DEF?D included?

YES YES

YES YES

Adj -R2 0.38 0.41 0.48 0.48

17

In-Sample Fit of Model

• Since 2006, the actual changes generally less than predicted

18

Stability Tests

• Has historical relationship between default rates and fundamentals changed?

• Statistical break tests: evidence of a break in 2003– Factor breakpoint test (10% significance)– CUSUM test (5% significance)

• No further breaks in sample from 2004

19

Distance to Default and Financing

Standard Merton-type Framework

0

20

40

60

80

100

120

140

160

1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97

Time

Fir

m V

alu

e as

Pct

of

So

lven

cy L

evel

Path A Path B High Threshold Low Threshold

20

Distance to Default and Structured Financing: Results

Growth in high-yield CLO issuance

Growth in aggregate CDO

issuance Explanatory Variable

Estimate t-stats Estimate t-stats

Dependent variable: ∆ DDEF Intercept 0.53 0.89 0.29 0.84 LL_GR, Lag1 0.37 0.86 --- --- LL_GR, Lag2 0.46** 2.36 --- --- LL_GR, Lag3 0.05** 4.15 --- --- LL_GR, Lag4 0.04** 3.85 --- ---

CDO_GR, Lag1 --- --- 0.94* 2.03 CDO_GR, Lag2 --- --- 0.86** 4.69 CDO_GR, Lag3 --- --- 0.67** 2.98 2 Lags of ∆ DDEF included?

YES YES

Wald test: All lags of financial innovation are zero Chi-sq p-value 0.00 0.00 Adj-R2 0.91 0.93

21

Distance to Default and Traditional Financing: Results

CP Issuance C&I Loans DDEF DDEF Explanatory Variable

Estimate t-stats Estimate t-stats

Intercept 1.61** 3.43 4.24 -0.08** ∆DDEF: Fitted, Lag1 --- --- --- 0.52**

∆DDEF: Resid, Lag1 --- --- --- -1.90* CP_GR, Lag1 2.29 0.46 --- --- CP_GR, Lag2 3.19 0.55 --- --- CP_GR, Lag3 -0.86 -0.17 --- --- CP_GR, Lag4 -4.58 -1.04 --- --- CP_GR, Lag5 -7.76** -2.42 --- --- CP_GR, Lag6 -5.79* -1.99 --- --- CIL_GR, Lag1 --- --- -29.00 -0.88 CIL_GR, Lag2 --- --- -27.94 -0.77 CIL_GR, Lag3 --- --- -76.01** -2.62 CIL_GR, Lag4 --- --- -30.81 -1.13 CIL_GR, Lag5 --- --- -45.53* -1.98 OTHER CONTROLS INCLUDED? YES YES

All lags=0? Chi-sq p-value 0.02 0.03

Sum of lags=0? Chi-sq p-value 0.38 0.03 Adj-R2 0.90 0.91

22

Financing and Default Rates

• Two channels

• Indirect: Effect on defaults via its effect on default boundary– Fitted DDEF: part explained by financing– Residual DDEF: part orthogonal to financing

• Direct

23

Structured Financing and Default Rates

No financing Growth in high-yield CLO issuance

Growth in CDO issuance

Explanatory Variable

Estimate t-stats Estimate t-stats Estimate t-stats

Intercept -0.08** -2.62 -0.04 -1.60 -0.01 -0.54 ∆DDEF: Fitted, Lag1 0.41* 1.75 0.29* 2.03 1.08** 4.73

∆DDEF: Resid, Lag1 -1.60* -1.93 -1.10* -1.81 -2.60** -2.42 LL_GR, Lag1 --- --- -0.04** -2.75 --- --- LL_GR, Lag2 --- --- 0.01 1.21 --- --- LL_GR, Lag3 --- --- 0.05** 3.96 --- --- LL_GR, Lag4 --- --- -0.00** -2.48 --- --- CDO_GR, Lag1 --- --- --- --- -0.07** -3.89 CDO_GR, Lag2 --- --- --- --- 0.02 0.83 CDO_GR, Lag3 --- --- --- --- -0.01 -0.63 CDO_GR, Lag4 --- --- --- --- -0.04* -2.55 CDO_GR, Lag5 --- --- --- --- -0.08** -3.93 CDO_GR, Lag6 --- --- --- --- -0.04** -2.41 SRET,Lag1 -0.48 -0.44 -1.95** -3.32 -1.81** -2.79 OTHER CONTROLS INCLUDED? YES YES YES Walt test: All lags of financial innovation are jointly zero Chi-sq p-value --- 0.00 0.00 Walt test: Sum of all lags of financial innovation is zero Chi-sq p-value --- 0.59 0.00 Adj-R2 0.24 0.65 0.61

24

Traditional Financing and Default Rates, 2005-2007

Explanatory Variable

Est. t-stats Estimate t-stats

Intercept -0.11** -2.95 -0.08** -3.12 ∆DDEF: Fitted, Lag1

0.53** 3.79 0.52** 2.09

∆DDEF: Resid, Lag1 -1.89** -2.86 -1.90* -2.05 CP_GR, Lag1 0.16 0.83 --- --- CP_GR, Lag2 0.40** 2.18 --- --- CP_GR, Lag3 0.23 1.47 --- --- CP_GR, Lag4 0.07 0.37 --- --- CP_GR, Lag5 -0.21 -1.27 --- --- CP_GR, Lag6 -0.24* -1.91 --- --- CIL_GR, Lag1 --- --- 0.43 0.48 OTHER CONTROLS INCLUDED? YES YES All lags=0? Chi-sq p-value 0.00 0.48 Sum of all lags=0? Chi-sq p-value 0.47 0.48 Adj-R2 0.35 0.23

25

Conclusions

• Proportion of early defaults are historically low– Consistent with “delayed defaults” (since defaults are now

already turning up)

• Structural break in recent sample, coincides with period of rapid growth in financing

• Including finance in prediction model is informative– Significantly related to distance to default– Component of distance to default explained by financing

positively related to defaults– Residual component negatively related to defaults