bond and cds pricing with stochastic recovery : moody's...

93
Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation Model Albert Cohen Actuarial Sciences Program Department of Mathematics Department of Statistics and Probability C336 Wells Hall Michigan State University East Lansing MI 48823 [email protected] [email protected] http://actuarialscience.natsci.msu.edu SFU Actuarial Seminar - June 2014 Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation Model Simon Fraser University 1 / 42

Upload: lethuan

Post on 26-Jul-2018

223 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Bond and CDS Pricing with Stochastic Recovery :Moody’s PD-LGD Correlation Model

Albert Cohen

Actuarial Sciences ProgramDepartment of Mathematics

Department of Statistics and ProbabilityC336 Wells Hall

Michigan State UniversityEast Lansing MI

[email protected]@stt.msu.edu

http://actuarialscience.natsci.msu.edu

SFU Actuarial Seminar - June 2014

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 1 / 42

Page 2: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Outline

1 Review of Some Structural ModelsFundamentals of Structural ModelsMerton ModelBlack-Cox ModelMoody’s KMV Model

2 Stochastic RecoveryComplete Recovery ModelPartial Recovery ModelCorrelated Asset-Recovery ModelBond Price - 2d Black-Cox ModelCDS - 2d Black-Cox ModelConclusions

Joint work with Nick Costanzino (nick [email protected], OTPP and

U of Toronto - RiskLab)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 2 / 42

Page 3: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Single-Name Default Models

Single-name default models typically fall into one of three main categories:

Structural Models. Attempts to explain default in terms offundamental properties, such as the firms balance sheet and economicconditions (Merton 1974, Black-Cox 1976, Leland 1994 etc).

Reduced Form (Intensity) Models. Directly postulates a model forthe instantaneous probability of default via an exogenous process λt(Jarrow-Turnbul 1995, Duffie-Singleton 1999 etc) via:

P[τ ∈ [t, t + dt)|Ft ] = λtdt

Hybrid Models. Incorporates features from structural andreduced-form models by postulating that the default intensity is afunction of the stock or of firm value (Madan-Unal 2000,Atlan-Leblanc 2005, Carr-Linetsky 2006 etc).

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 3 / 42

Page 4: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Single-Name Default Models

Single-name default models typically fall into one of three main categories:

Structural Models. Attempts to explain default in terms offundamental properties, such as the firms balance sheet and economicconditions (Merton 1974, Black-Cox 1976, Leland 1994 etc).

Reduced Form (Intensity) Models. Directly postulates a model forthe instantaneous probability of default via an exogenous process λt(Jarrow-Turnbul 1995, Duffie-Singleton 1999 etc) via:

P[τ ∈ [t, t + dt)|Ft ] = λtdt

Hybrid Models. Incorporates features from structural andreduced-form models by postulating that the default intensity is afunction of the stock or of firm value (Madan-Unal 2000,Atlan-Leblanc 2005, Carr-Linetsky 2006 etc).

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 3 / 42

Page 5: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Single-Name Default Models

Single-name default models typically fall into one of three main categories:

Structural Models. Attempts to explain default in terms offundamental properties, such as the firms balance sheet and economicconditions (Merton 1974, Black-Cox 1976, Leland 1994 etc).

Reduced Form (Intensity) Models. Directly postulates a model forthe instantaneous probability of default via an exogenous process λt(Jarrow-Turnbul 1995, Duffie-Singleton 1999 etc) via:

P[τ ∈ [t, t + dt)|Ft ] = λtdt

Hybrid Models. Incorporates features from structural andreduced-form models by postulating that the default intensity is afunction of the stock or of firm value (Madan-Unal 2000,Atlan-Leblanc 2005, Carr-Linetsky 2006 etc).

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 3 / 42

Page 6: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Single-Name Default Models

Single-name default models typically fall into one of three main categories:

Structural Models. Attempts to explain default in terms offundamental properties, such as the firms balance sheet and economicconditions (Merton 1974, Black-Cox 1976, Leland 1994 etc).

Reduced Form (Intensity) Models. Directly postulates a model forthe instantaneous probability of default via an exogenous process λt(Jarrow-Turnbul 1995, Duffie-Singleton 1999 etc) via:

P[τ ∈ [t, t + dt)|Ft ] = λtdt

Hybrid Models. Incorporates features from structural andreduced-form models by postulating that the default intensity is afunction of the stock or of firm value (Madan-Unal 2000,Atlan-Leblanc 2005, Carr-Linetsky 2006 etc).

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 3 / 42

Page 7: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Single-Name Default Models

Single-name default models typically fall into one of three main categories:

Structural Models. Attempts to explain default in terms offundamental properties, such as the firms balance sheet and economicconditions (Merton 1974, Black-Cox 1976, Leland 1994 etc).

Reduced Form (Intensity) Models. Directly postulates a model forthe instantaneous probability of default via an exogenous process λt(Jarrow-Turnbul 1995, Duffie-Singleton 1999 etc) via:

P[τ ∈ [t, t + dt)|Ft ] = λtdt

Hybrid Models. Incorporates features from structural andreduced-form models by postulating that the default intensity is afunction of the stock or of firm value (Madan-Unal 2000,Atlan-Leblanc 2005, Carr-Linetsky 2006 etc).

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 3 / 42

Page 8: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

What Exactly Are Structural Models Supposed To Do?

Price both corporate debt and equity securities (perhaps evenequity derivatives and CDS!) Important for buyers/investors,sellers/firms, and advisors

Estimate default probabilities of firms. Useful to commercialbanks, investment banks, rating agencies etc.

Determining optimal capital structure decisions. Essential part ofeconomic capital estimation (Moody’s Portfolio Manager andRiskFrontier)

Analyzing most corporate decisions that affects cash flows. Canaid in determining most value-maximizing decisions

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 4 / 42

Page 9: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

What Exactly Are Structural Models Supposed To Do?

Price both corporate debt and equity securities (perhaps evenequity derivatives and CDS!) Important for buyers/investors,sellers/firms, and advisors

Estimate default probabilities of firms. Useful to commercialbanks, investment banks, rating agencies etc.

Determining optimal capital structure decisions. Essential part ofeconomic capital estimation (Moody’s Portfolio Manager andRiskFrontier)

Analyzing most corporate decisions that affects cash flows. Canaid in determining most value-maximizing decisions

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 4 / 42

Page 10: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

What Exactly Are Structural Models Supposed To Do?

Price both corporate debt and equity securities (perhaps evenequity derivatives and CDS!) Important for buyers/investors,sellers/firms, and advisors

Estimate default probabilities of firms. Useful to commercialbanks, investment banks, rating agencies etc.

Determining optimal capital structure decisions. Essential part ofeconomic capital estimation (Moody’s Portfolio Manager andRiskFrontier)

Analyzing most corporate decisions that affects cash flows. Canaid in determining most value-maximizing decisions

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 4 / 42

Page 11: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

What Exactly Are Structural Models Supposed To Do?

Price both corporate debt and equity securities (perhaps evenequity derivatives and CDS!) Important for buyers/investors,sellers/firms, and advisors

Estimate default probabilities of firms. Useful to commercialbanks, investment banks, rating agencies etc.

Determining optimal capital structure decisions. Essential part ofeconomic capital estimation (Moody’s Portfolio Manager andRiskFrontier)

Analyzing most corporate decisions that affects cash flows. Canaid in determining most value-maximizing decisions

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 4 / 42

Page 12: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

What Exactly Are Structural Models Supposed To Do?

Price both corporate debt and equity securities (perhaps evenequity derivatives and CDS!) Important for buyers/investors,sellers/firms, and advisors

Estimate default probabilities of firms. Useful to commercialbanks, investment banks, rating agencies etc.

Determining optimal capital structure decisions. Essential part ofeconomic capital estimation (Moody’s Portfolio Manager andRiskFrontier)

Analyzing most corporate decisions that affects cash flows. Canaid in determining most value-maximizing decisions

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 4 / 42

Page 13: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

How Do We Mathematically Define Default?

Several possible triggers for default:

Zero Net Worth Trigger. Asset value falls below debt outstanding.But... firms often continue to operate even with negative networth. Might issue more stock and pay coupon rather thandefault.

Zero CashflowTrigger. Cashflows inadequate to cover operating costs.But... zero current net cash flow doesn’t always imply zeroequity value. Shares often have positive value in thisscenario. Firm could again issue more stock rather thandefault.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 5 / 42

Page 14: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

How Do We Mathematically Define Default?

Several possible triggers for default:

Zero Net Worth Trigger. Asset value falls below debt outstanding.But... firms often continue to operate even with negative networth. Might issue more stock and pay coupon rather thandefault.

Zero CashflowTrigger. Cashflows inadequate to cover operating costs.But... zero current net cash flow doesn’t always imply zeroequity value. Shares often have positive value in thisscenario. Firm could again issue more stock rather thandefault.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 5 / 42

Page 15: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

How Do We Mathematically Define Default?

Several possible triggers for default:

Zero Net Worth Trigger. Asset value falls below debt outstanding.But... firms often continue to operate even with negative networth. Might issue more stock and pay coupon rather thandefault.

Zero CashflowTrigger. Cashflows inadequate to cover operating costs.But... zero current net cash flow doesn’t always imply zeroequity value. Shares often have positive value in thisscenario. Firm could again issue more stock rather thandefault.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 5 / 42

Page 16: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Simplest Framework for Structural Models

Assume that the firm’s assets At is the sum of equity Et and debt Dt .

Assumptions on the assets process must be postulated.

Assumptions on the debt structure must be postulated.

A default mechanism τ and payout structure f (Aτ , τ) must bepostulated.

The main idea, pioneered by Merton, is that the firms equity Et isthen modeled as a call option on the assets At of the firm at expiry T :

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]This connects equity Et with debt value Dt . In particular if Dt has atractable debt structure it provides a methodology to compute defaultprobability, bond price, credit spreads, recovery rates, etc.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 6 / 42

Page 17: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Simplest Framework for Structural Models

Assume that the firm’s assets At is the sum of equity Et and debt Dt .

Assumptions on the assets process must be postulated.

Assumptions on the debt structure must be postulated.

A default mechanism τ and payout structure f (Aτ , τ) must bepostulated.

The main idea, pioneered by Merton, is that the firms equity Et isthen modeled as a call option on the assets At of the firm at expiry T :

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]This connects equity Et with debt value Dt . In particular if Dt has atractable debt structure it provides a methodology to compute defaultprobability, bond price, credit spreads, recovery rates, etc.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 6 / 42

Page 18: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Simplest Framework for Structural Models

Assume that the firm’s assets At is the sum of equity Et and debt Dt .

Assumptions on the assets process must be postulated.

Assumptions on the debt structure must be postulated.

A default mechanism τ and payout structure f (Aτ , τ) must bepostulated.

The main idea, pioneered by Merton, is that the firms equity Et isthen modeled as a call option on the assets At of the firm at expiry T :

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]This connects equity Et with debt value Dt . In particular if Dt has atractable debt structure it provides a methodology to compute defaultprobability, bond price, credit spreads, recovery rates, etc.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 6 / 42

Page 19: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Simplest Framework for Structural Models

Assume that the firm’s assets At is the sum of equity Et and debt Dt .

Assumptions on the assets process must be postulated.

Assumptions on the debt structure must be postulated.

A default mechanism τ and payout structure f (Aτ , τ) must bepostulated.

The main idea, pioneered by Merton, is that the firms equity Et isthen modeled as a call option on the assets At of the firm at expiry T :

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]This connects equity Et with debt value Dt . In particular if Dt has atractable debt structure it provides a methodology to compute defaultprobability, bond price, credit spreads, recovery rates, etc.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 6 / 42

Page 20: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Simplest Framework for Structural Models

Assume that the firm’s assets At is the sum of equity Et and debt Dt .

Assumptions on the assets process must be postulated.

Assumptions on the debt structure must be postulated.

A default mechanism τ and payout structure f (Aτ , τ) must bepostulated.

The main idea, pioneered by Merton, is that the firms equity Et isthen modeled as a call option on the assets At of the firm at expiry T :

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]This connects equity Et with debt value Dt . In particular if Dt has atractable debt structure it provides a methodology to compute defaultprobability, bond price, credit spreads, recovery rates, etc.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 6 / 42

Page 21: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Simplest Framework for Structural Models

Assume that the firm’s assets At is the sum of equity Et and debt Dt .

Assumptions on the assets process must be postulated.

Assumptions on the debt structure must be postulated.

A default mechanism τ and payout structure f (Aτ , τ) must bepostulated.

The main idea, pioneered by Merton, is that the firms equity Et isthen modeled as a call option on the assets At of the firm at expiry T :

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]This connects equity Et with debt value Dt . In particular if Dt has atractable debt structure it provides a methodology to compute defaultprobability, bond price, credit spreads, recovery rates, etc.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 6 / 42

Page 22: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Simplest Framework for Structural Models

Assume that the firm’s assets At is the sum of equity Et and debt Dt .

Assumptions on the assets process must be postulated.

Assumptions on the debt structure must be postulated.

A default mechanism τ and payout structure f (Aτ , τ) must bepostulated.

The main idea, pioneered by Merton, is that the firms equity Et isthen modeled as a call option on the assets At of the firm at expiry T :

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]

This connects equity Et with debt value Dt . In particular if Dt has atractable debt structure it provides a methodology to compute defaultprobability, bond price, credit spreads, recovery rates, etc.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 6 / 42

Page 23: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Simplest Framework for Structural Models

Assume that the firm’s assets At is the sum of equity Et and debt Dt .

Assumptions on the assets process must be postulated.

Assumptions on the debt structure must be postulated.

A default mechanism τ and payout structure f (Aτ , τ) must bepostulated.

The main idea, pioneered by Merton, is that the firms equity Et isthen modeled as a call option on the assets At of the firm at expiry T :

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]This connects equity Et with debt value Dt . In particular if Dt has atractable debt structure it provides a methodology to compute defaultprobability, bond price, credit spreads, recovery rates, etc.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 6 / 42

Page 24: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Set-up

Assume a probability space (Ω,F ,P)

Underlying asset is modeled as GBM under Physical Measure P:

dAt = µAAtdt + σAAtdW At

Default is implicitly assumed to coincide with the event

AT < B

:

τMerton = T 1AT<B +∞1AT≥B

This results in a turnover of the company’s assets to bondholders ifassets are worth less than the total value of bond outstanding.

At maturity the bond value (payoff) is:

payoffMerton = BT = AT1τ=T + B1τ=∞

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 7 / 42

Page 25: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Set-up

Assume a probability space (Ω,F ,P)

Underlying asset is modeled as GBM under Physical Measure P:

dAt = µAAtdt + σAAtdW At

Default is implicitly assumed to coincide with the event

AT < B

:

τMerton = T 1AT<B +∞1AT≥B

This results in a turnover of the company’s assets to bondholders ifassets are worth less than the total value of bond outstanding.

At maturity the bond value (payoff) is:

payoffMerton = BT = AT1τ=T + B1τ=∞

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 7 / 42

Page 26: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Set-up

Assume a probability space (Ω,F ,P)

Underlying asset is modeled as GBM under Physical Measure P:

dAt = µAAtdt + σAAtdW At

Default is implicitly assumed to coincide with the event

AT < B

:

τMerton = T 1AT<B +∞1AT≥B

This results in a turnover of the company’s assets to bondholders ifassets are worth less than the total value of bond outstanding.

At maturity the bond value (payoff) is:

payoffMerton = BT = AT1τ=T + B1τ=∞

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 7 / 42

Page 27: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Set-up

Assume a probability space (Ω,F ,P)

Underlying asset is modeled as GBM under Physical Measure P:

dAt = µAAtdt + σAAtdW At

Default is implicitly assumed to coincide with the event

AT < B

:

τMerton = T 1AT<B +∞1AT≥B

This results in a turnover of the company’s assets to bondholders ifassets are worth less than the total value of bond outstanding.

At maturity the bond value (payoff) is:

payoffMerton = BT = AT1τ=T + B1τ=∞

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 7 / 42

Page 28: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Set-up

Assume a probability space (Ω,F ,P)

Underlying asset is modeled as GBM under Physical Measure P:

dAt = µAAtdt + σAAtdW At

Default is implicitly assumed to coincide with the event

AT < B

:

τMerton = T 1AT<B +∞1AT≥B

This results in a turnover of the company’s assets to bondholders ifassets are worth less than the total value of bond outstanding.

At maturity the bond value (payoff) is:

payoffMerton = BT = AT1τ=T + B1τ=∞

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 7 / 42

Page 29: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Bond Price

Consequently,

Bt = e−r(T−t)Et [BT ] = e−r(T−t)Et [AT1τ=T + B1τ=∞]

= e−r(T−t)

[∫ B

0A · Pt [AT ∈ dA] + B

∫ ∞B

Pt [AT ∈ dA]

]= e−r(T−t)Et [AT | AT < B] · Pt [AT < B] + e−r(T−t)B · Pt [AT ≥ B]

= Be−r(T−t) − e−r(T−t)Et [(B − AT ) | AT < B] · Pt [AT < B]

:= Be−r(T−t) − Be−r(T−t)Et [Loss | Default] · Pt [Default].

Here, Loss denotes the loss per unit of face in the event

AT < B.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 8 / 42

Page 30: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Bond Price

Consequently,

Bt = e−r(T−t)Et [BT ] = e−r(T−t)Et [AT1τ=T + B1τ=∞]

= e−r(T−t)

[∫ B

0A · Pt [AT ∈ dA] + B

∫ ∞B

Pt [AT ∈ dA]

]= e−r(T−t)Et [AT | AT < B] · Pt [AT < B] + e−r(T−t)B · Pt [AT ≥ B]

= Be−r(T−t) − e−r(T−t)Et [(B − AT ) | AT < B] · Pt [AT < B]

:= Be−r(T−t) − Be−r(T−t)Et [Loss | Default] · Pt [Default].

Here, Loss denotes the loss per unit of face in the event

AT < B.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 8 / 42

Page 31: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model PD and LGD

BMertont,T = Be−r(T−t)N (d2) + AtN (−d1)

PDMerton = Pt [AT < B] = N (−d2)

LGDMerton =1

BEt [B − AT |AT < B] = 1− er(T−t) At

B

N (−d1)

N (−d2)

d1 =ln(At/B) + (r + 1

2σ2A)(T − t)

σA√

T − t= d+

d2 =ln(At/B) + (r − 1

2σ2A)(T − t)

σA√

T − t= d−

(1)

Real-world probabilities from risk-neutral probabilities via µ→ r (WangTransform:)

PDPMerton = N(N−1(PDQMerton)− µA − r

σA

√T

)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 9 / 42

Page 32: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model PD and LGD

BMertont,T = Be−r(T−t)N (d2) + AtN (−d1)

PDMerton = Pt [AT < B] = N (−d2)

LGDMerton =1

BEt [B − AT |AT < B] = 1− er(T−t) At

B

N (−d1)

N (−d2)

d1 =ln(At/B) + (r + 1

2σ2A)(T − t)

σA√

T − t= d+

d2 =ln(At/B) + (r − 1

2σ2A)(T − t)

σA√

T − t= d−

(1)

Real-world probabilities from risk-neutral probabilities via µ→ r (WangTransform:)

PDPMerton = N(N−1(PDQMerton)− µA − r

σA

√T

)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 9 / 42

Page 33: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model PD and LGD

BMertont,T = Be−r(T−t)N (d2) + AtN (−d1)

PDMerton = Pt [AT < B] = N (−d2)

LGDMerton =1

BEt [B − AT |AT < B] = 1− er(T−t) At

B

N (−d1)

N (−d2)

d1 =ln(At/B) + (r + 1

2σ2A)(T − t)

σA√

T − t= d+

d2 =ln(At/B) + (r − 1

2σ2A)(T − t)

σA√

T − t= d−

(1)

Real-world probabilities from risk-neutral probabilities via µ→ r (WangTransform:)

PDPMerton = N(N−1(PDQMerton)− µA − r

σA

√T

)Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 9 / 42

Page 34: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Credit Spread

The yield-to-maturity credit spread Y (t,T ) is defined as the spreadover the risk free rate r which prices the bond:

Bt = Be−(r+Y (t,T ))(T−t)

Solving for Y yields

Y (t,T ) =1

T − tln

(B

Bt

)− r

For the Merton Model:

YMerton(t,T ) = − 1

T − tln

[N (d2) +

At

Ber(T−t)N (−d1)

]

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 10 / 42

Page 35: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Credit Spread

The yield-to-maturity credit spread Y (t,T ) is defined as the spreadover the risk free rate r which prices the bond:

Bt = Be−(r+Y (t,T ))(T−t)

Solving for Y yields

Y (t,T ) =1

T − tln

(B

Bt

)− r

For the Merton Model:

YMerton(t,T ) = − 1

T − tln

[N (d2) +

At

Ber(T−t)N (−d1)

]

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 10 / 42

Page 36: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Credit Spread

The yield-to-maturity credit spread Y (t,T ) is defined as the spreadover the risk free rate r which prices the bond:

Bt = Be−(r+Y (t,T ))(T−t)

Solving for Y yields

Y (t,T ) =1

T − tln

(B

Bt

)− r

For the Merton Model:

YMerton(t,T ) = − 1

T − tln

[N (d2) +

At

Ber(T−t)N (−d1)

]

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 10 / 42

Page 37: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Credit Spread

The yield-to-maturity credit spread Y (t,T ) is defined as the spreadover the risk free rate r which prices the bond:

Bt = Be−(r+Y (t,T ))(T−t)

Solving for Y yields

Y (t,T ) =1

T − tln

(B

Bt

)− r

For the Merton Model:

YMerton(t,T ) = − 1

T − tln

[N (d2) +

At

Ber(T−t)N (−d1)

]

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 10 / 42

Page 38: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Credit Spread

The yield-to-maturity credit spread Y (t,T ) is defined as the spreadover the risk free rate r which prices the bond:

Bt = Be−(r+Y (t,T ))(T−t)

Solving for Y yields

Y (t,T ) =1

T − tln

(B

Bt

)− r

For the Merton Model:

YMerton(t,T ) = − 1

T − tln

[N (d2) +

At

Ber(T−t)N (−d1)

]

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 10 / 42

Page 39: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Parameter Estimation

To compute the desired quantities, model asks for A and σA as inputs.

These are unobservable from the markets.

The solution is to imply them from the market.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 11 / 42

Page 40: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Parameter Estimation

To compute the desired quantities, model asks for A and σA as inputs.

These are unobservable from the markets.

The solution is to imply them from the market.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 11 / 42

Page 41: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Parameter Estimation

To compute the desired quantities, model asks for A and σA as inputs.

These are unobservable from the markets.

The solution is to imply them from the market.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 11 / 42

Page 42: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Parameter Estimation

To compute the desired quantities, model asks for A and σA as inputs.

These are unobservable from the markets.

The solution is to imply them from the market.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 11 / 42

Page 43: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Parameter Estimation

There are many ways to do this, but one is to solve the equations

Emarkett = EMerton

t = AtN (d1)− Be−r(T−t)N (d2)

σmarketE Emarket

t = σAAtN (d1).(2)

Note that these equations above are obtained via

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]σVσA

=

∣∣∣∣∣ A∂V∂A

V (A, t)

∣∣∣∣∣ .

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 12 / 42

Page 44: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Parameter Estimation

There are many ways to do this, but one is to solve the equations

Emarkett = EMerton

t = AtN (d1)− Be−r(T−t)N (d2)

σmarketE Emarket

t = σAAtN (d1).(2)

Note that these equations above are obtained via

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]σVσA

=

∣∣∣∣∣ A∂V∂A

V (A, t)

∣∣∣∣∣ .

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 12 / 42

Page 45: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Parameter Estimation

There are many ways to do this, but one is to solve the equations

Emarkett = EMerton

t = AtN (d1)− Be−r(T−t)N (d2)

σmarketE Emarket

t = σAAtN (d1).(2)

Note that these equations above are obtained via

Et = V (A, t) = E[e−r(T−t)(AT − B)+ | At = A

]σVσA

=

∣∣∣∣∣ A∂V∂A

V (A, t)

∣∣∣∣∣ .

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 12 / 42

Page 46: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Criticisms

The assumption that the firm can only default at debt maturity isempirically violated.

Model tends to underestimate default probabilities and credit spreads,especially for short maturities. Indeed

limT→t+

1

T − tPt [AT ≤ B] = 0

Does not take into account empirical evidence that recovery iscorrelated to probability of default.

Does not take into account liquidity issues.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 13 / 42

Page 47: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Criticisms

The assumption that the firm can only default at debt maturity isempirically violated.

Model tends to underestimate default probabilities and credit spreads,especially for short maturities. Indeed

limT→t+

1

T − tPt [AT ≤ B] = 0

Does not take into account empirical evidence that recovery iscorrelated to probability of default.

Does not take into account liquidity issues.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 13 / 42

Page 48: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Criticisms

The assumption that the firm can only default at debt maturity isempirically violated.

Model tends to underestimate default probabilities and credit spreads,especially for short maturities. Indeed

limT→t+

1

T − tPt [AT ≤ B] = 0

Does not take into account empirical evidence that recovery iscorrelated to probability of default.

Does not take into account liquidity issues.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 13 / 42

Page 49: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Criticisms

The assumption that the firm can only default at debt maturity isempirically violated.

Model tends to underestimate default probabilities and credit spreads,especially for short maturities. Indeed

limT→t+

1

T − tPt [AT ≤ B] = 0

Does not take into account empirical evidence that recovery iscorrelated to probability of default.

Does not take into account liquidity issues.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 13 / 42

Page 50: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Merton Model Criticisms

The assumption that the firm can only default at debt maturity isempirically violated.

Model tends to underestimate default probabilities and credit spreads,especially for short maturities. Indeed

limT→t+

1

T − tPt [AT ≤ B] = 0

Does not take into account empirical evidence that recovery iscorrelated to probability of default.

Does not take into account liquidity issues.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 13 / 42

Page 51: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Black-Cox Model Setup

Same asset dynamics as in Merton model:

dAt = µAAtdt + σAAtdW At

Default can happen at times other than maturity, in particular whenthe assets fall below a prescribed default point DP:

τBlack−Cox = mininft ≥ 0 : At ≤ DP, τMerton

In the original Black-Cox paper, DP was taken to beDP = Ce−γ(T−t).

The payoff to the bondholder is

payoffBlack−Cox = Aτ1τ≤T + B1τ>T

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 14 / 42

Page 52: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Black-Cox Model Setup

Same asset dynamics as in Merton model:

dAt = µAAtdt + σAAtdW At

Default can happen at times other than maturity, in particular whenthe assets fall below a prescribed default point DP:

τBlack−Cox = mininft ≥ 0 : At ≤ DP, τMerton

In the original Black-Cox paper, DP was taken to beDP = Ce−γ(T−t).

The payoff to the bondholder is

payoffBlack−Cox = Aτ1τ≤T + B1τ>T

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 14 / 42

Page 53: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Black-Cox Model Setup

Same asset dynamics as in Merton model:

dAt = µAAtdt + σAAtdW At

Default can happen at times other than maturity, in particular whenthe assets fall below a prescribed default point DP:

τBlack−Cox = mininft ≥ 0 : At ≤ DP, τMerton

In the original Black-Cox paper, DP was taken to beDP = Ce−γ(T−t).

The payoff to the bondholder is

payoffBlack−Cox = Aτ1τ≤T + B1τ>T

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 14 / 42

Page 54: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Black-Cox Model Setup

Same asset dynamics as in Merton model:

dAt = µAAtdt + σAAtdW At

Default can happen at times other than maturity, in particular whenthe assets fall below a prescribed default point DP:

τBlack−Cox = mininft ≥ 0 : At ≤ DP, τMerton

In the original Black-Cox paper, DP was taken to beDP = Ce−γ(T−t).

The payoff to the bondholder is

payoffBlack−Cox = Aτ1τ≤T + B1τ>T

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 14 / 42

Page 55: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Black-Cox Model Setup

Same asset dynamics as in Merton model:

dAt = µAAtdt + σAAtdW At

Default can happen at times other than maturity, in particular whenthe assets fall below a prescribed default point DP:

τBlack−Cox = mininft ≥ 0 : At ≤ DP, τMerton

In the original Black-Cox paper, DP was taken to beDP = Ce−γ(T−t).

The payoff to the bondholder is

payoffBlack−Cox = Aτ1τ≤T + B1τ>T

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 14 / 42

Page 56: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Black-Cox Model Setup

Same asset dynamics as in Merton model:

dAt = µAAtdt + σAAtdW At

Default can happen at times other than maturity, in particular whenthe assets fall below a prescribed default point DP:

τBlack−Cox = mininft ≥ 0 : At ≤ DP, τMerton

In the original Black-Cox paper, DP was taken to beDP = Ce−γ(T−t).

The payoff to the bondholder is

payoffBlack−Cox = Aτ1τ≤T + B1τ>T

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 14 / 42

Page 57: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Black-Cox Model Setup

Same asset dynamics as in Merton model:

dAt = µAAtdt + σAAtdW At

Default can happen at times other than maturity, in particular whenthe assets fall below a prescribed default point DP:

τBlack−Cox = mininft ≥ 0 : At ≤ DP, τMerton

In the original Black-Cox paper, DP was taken to beDP = Ce−γ(T−t).

The payoff to the bondholder is

payoffBlack−Cox = Aτ1τ≤T + B1τ>T

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 14 / 42

Page 58: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

KMV Model Setup

Moody’s starts by defining a “distance-to-default” DD:

DD =ln A

DP + (µA − σ2A/2)(T − t)

σA√

T − t

which is just d2 in the Merton model with B replaced with DP and rreplaced with µA.

Default point DP chosen as the short term debt plus half oflong-term debt:

DP = ST +1

2LT

Then calibrate DD to empirical distribution Ψ of realized defaultsfrom proprietary database:

PDMoody′s = Ψ(−DD)

Compared to Merton’s estimate of PD (i.e. PDMerton = N (−d2))prediction power is far superior.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 15 / 42

Page 59: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

KMV Model Setup

Moody’s starts by defining a “distance-to-default” DD:

DD =ln A

DP + (µA − σ2A/2)(T − t)

σA√

T − t

which is just d2 in the Merton model with B replaced with DP and rreplaced with µA.

Default point DP chosen as the short term debt plus half oflong-term debt:

DP = ST +1

2LT

Then calibrate DD to empirical distribution Ψ of realized defaultsfrom proprietary database:

PDMoody′s = Ψ(−DD)

Compared to Merton’s estimate of PD (i.e. PDMerton = N (−d2))prediction power is far superior.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 15 / 42

Page 60: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

KMV Model Setup

Moody’s starts by defining a “distance-to-default” DD:

DD =ln A

DP + (µA − σ2A/2)(T − t)

σA√

T − t

which is just d2 in the Merton model with B replaced with DP and rreplaced with µA.

Default point DP chosen as the short term debt plus half oflong-term debt:

DP = ST +1

2LT

Then calibrate DD to empirical distribution Ψ of realized defaultsfrom proprietary database:

PDMoody′s = Ψ(−DD)

Compared to Merton’s estimate of PD (i.e. PDMerton = N (−d2))prediction power is far superior.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 15 / 42

Page 61: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

KMV Model Setup

Moody’s starts by defining a “distance-to-default” DD:

DD =ln A

DP + (µA − σ2A/2)(T − t)

σA√

T − t

which is just d2 in the Merton model with B replaced with DP and rreplaced with µA.

Default point DP chosen as the short term debt plus half oflong-term debt:

DP = ST +1

2LT

Then calibrate DD to empirical distribution Ψ of realized defaultsfrom proprietary database:

PDMoody′s = Ψ(−DD)

Compared to Merton’s estimate of PD (i.e. PDMerton = N (−d2))prediction power is far superior.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 15 / 42

Page 62: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

KMV Model Setup

Moody’s starts by defining a “distance-to-default” DD:

DD =ln A

DP + (µA − σ2A/2)(T − t)

σA√

T − t

which is just d2 in the Merton model with B replaced with DP and rreplaced with µA.

Default point DP chosen as the short term debt plus half oflong-term debt:

DP = ST +1

2LT

Then calibrate DD to empirical distribution Ψ of realized defaultsfrom proprietary database:

PDMoody′s = Ψ(−DD)

Compared to Merton’s estimate of PD (i.e. PDMerton = N (−d2))prediction power is far superior.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 15 / 42

Page 63: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

American Airlines Default

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 16 / 42

Page 64: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

MF Global Default

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 17 / 42

Page 65: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Bond Price under Correlated A-R Model

We assume

Existence of a risk-neutral measure P with risk-free rate rf .

Default τ is adapted to the filtration Ft = σ(At ,Rt).

A recovery process Rt .

Bond has maturity T .

It follows that

Bt(ω) = Be−rf (T−t)Pt [τ > T ] + Et

[e−rf (T−t)RT1τ=T

]+ Et

[e−rf (τ−t)Rτ1τ<T

].

(3)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 18 / 42

Page 66: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Complete Recovery Upon Default and No Default

If it happens that the entire arbitrage-free value of the bond is recoveredupon default, i.e. under risk-free rate rf and risk-neutral measure P,

Et [Rs ] = Be−rf (T−s) (4)

then the entire no-default value is retained:

Bt(ω) = Be−rf (T−t)Pt [τ > T ] + Et

[e−rf (T−t)B1τ=T

]+

∫ T

te−rf (s−t)Be−rf (T−s)Pt [τ ∈ ds]

= Be−rf (T−t)

(5)

Of course, if Pt [τ > T ] = 1, then Pt [τ ∈ ds] = 0 for all s ∈ [t,T ] and soBt = Be−rf (T−t) once again.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 19 / 42

Page 67: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Example-Basic Hazard Rate Model

As an initial example to motivate bond pricing formula, assume that bothdefault time and recovery are independent of the Asset filtration, anddefault time is in fact memoryless:

Pt [τ > u] = e−λ(u−t) for some λ > 0 and all u ≥ t ≥ 0.

Et [e−rf sRs ] = e−rf tRt .

It follows that

Bt(ω) = Be−rf (T−t)e−λ(T−t) + Rt(ω) · (1− e−λ(T−t)) and

Y (t,T )(ω) = − 1T−t ln

[e−λ(T−t) + Rt(ω)

Be−rf (T−t) (1− e−λ(T−t))]

limT→t+ Y (t,T )(ω) = λ(

1− Rt(ω)

B

).

Note that the short term credit spread is positive a.e. as long asPt

[Rt < B

]= 1.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 20 / 42

Page 68: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Bond Price under Correlated A-R Model

Define the quantities:

At , the asset value at time t > 0

Rt , the recovery amount at time t > 0. We model recovery as a”shadow” asset, and thus follows the same dynamics as the asset.

The dynamics for the asset and recovery are, under a risk neutral measureP,

dAt

At= rf dt + σAdW A

t

dRt

Rt= rf dt + σRdW R

t .

(6)

The risk drivers are correlated via

E[dW At dW R

t ] = ρA,Rdt. (7)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 21 / 42

Page 69: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Bond Price under Correlated A-R Model

Define the quantities:

At , the asset value at time t > 0

Rt , the recovery amount at time t > 0. We model recovery as a”shadow” asset, and thus follows the same dynamics as the asset.

The dynamics for the asset and recovery are, under a risk neutral measureP,

dAt

At= rf dt + σAdW A

t

dRt

Rt= rf dt + σRdW R

t .

(6)

The risk drivers are correlated via

E[dW At dW R

t ] = ρA,Rdt. (7)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 21 / 42

Page 70: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Bond Price under Correlated A-R Model

Define the quantities:

At , the asset value at time t > 0

Rt , the recovery amount at time t > 0. We model recovery as a”shadow” asset, and thus follows the same dynamics as the asset.

The dynamics for the asset and recovery are, under a risk neutral measureP,

dAt

At= rf dt + σAdW A

t

dRt

Rt= rf dt + σRdW R

t .

(6)

The risk drivers are correlated via

E[dW At dW R

t ] = ρA,Rdt. (7)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 21 / 42

Page 71: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Two Factor Merton Model - Default at Maturity

If we allow for a shadow recovery process as in equation (6), then we areleft with an exchange option where the payoff at expiry is

G (AT ,RT ) = RT1AT<B + B1AT≥B. (8)

Define

Default =

AT ≥ Bc

α := δ +1

2σ2R(1− ρ2

A,R) + γ(

rf −1

2σ2A

)+γ2σ2

A

2

γ := ρA,RσRσA

δ := (rf −1

2σ2R)− γ(rf −

1

2σ2A).

(9)

Note that ρA,R = γ = 1 implies that δ = 0 and α = rf .

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 22 / 42

Page 72: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Two Factor Merton Model - Default at Maturity

Consequently,

Bt = e−rf (T−t)Et

[RT1AT<B + B1AT≥B

]= Et

[e−rf (T−t)RT1AT<B

]+ Be−rf (T−t) · Pt [AT ≥ B]

= Rte(α−rf )(T−t)Φ

(ln B

At− (rf − 1

2σ2A)(T − t)− γσ2

A(T − t)

σA√

T − t

)

+ Be−rf (T−t) ·

(1− Φ

(ln B

At− (rf − 1

2σ2A)(T − t)

σA√

T − t

))(10)

Note that we retain the classical 1d Merton price if ρA,R = γ = 1.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 23 / 42

Page 73: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Two Factor Merton Model - Default at Maturity

Recovery metrics:

Et [Loss | Default] =Et [B − RT | AT < B]

B

= 1− Rt

Be−α(T−t)

Φ

(ln B

At−(rf− 1

2σ2A)(T−t)−γσ2

A(T−t)

σA√T−t

)

Φ

(ln B

At−(rf− 1

2σ2A)(T−t)

σA√T−t

)

Pt [Default] = Φ

(ln B

At− (rf − 1

2σ2A)(T − t)

σA√

T − t

)

Y2D Merton(t,T ) =1

T − tln

(1

1− Pt [Default]Et [Loss | Default]

)

(11)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 24 / 42

Page 74: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Consistency: 2d vs 1d Merton

Finally, it should be noted that since

limγ→1

dγ = d1

limρ→1,γ→1

α = rf

limρ→1,γ→1,R0→A0

P[At = Rt ] = 1,

(12)

it follows that

limρ→1,γ→1,Rt→At

Y2d Merton(t,T ) = YMerton(t,T ). (13)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 25 / 42

Page 75: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Two Factor Black-Cox Model - Stopping Times

Let DP define our default point written into the bondholder covenant onthe asset A, and define τDP as the first time A reaches this default point.

Furthermore, define 2dBC default time τ as

τ =

τDP : τDP < TT : τDP ≥ T ∩

AT < B

∞ : otherwise.

Equivalently,

Default =τDP > T ,AT ≥ B

c. (14)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 26 / 42

Page 76: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Two Factor Black-Cox Model - Stopping Times

Let DP define our default point written into the bondholder covenant onthe asset A, and define τDP as the first time A reaches this default point.

Furthermore, define 2dBC default time τ as

τ =

τDP : τDP < TT : τDP ≥ T ∩

AT < B

∞ : otherwise.

Equivalently,

Default =τDP > T ,AT ≥ B

c. (14)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 26 / 42

Page 77: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Weak Covenant Model: DP < B

Theorem

B2d Black Coxt,T = Be−rf (T−t) · Pt [AT ≥ B, τDP > T ]

+[Rt − Et [e

−rf (T−t)RT1AT≥B,τDP>T]]

:= Be−rf (T−t)[1− u(w , y ,T − t)

]+ v(Rt ,w , y ,T − t)

y = lnDP

At≤ 0

w = lnB

At≤ 0

µ = rf −1

2σ2A

γ = ρA,RσRσA.

(15)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 27 / 42

Page 78: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Weak Covenant Model: DP < B

u(w , y ,T − t) = N

w − µ(T − t)√σ2A(T − t)

+ e

2µy

σ2A N

−w − 2y − µ(T − t)√σ2A(T − t)

v(Rt ,w , y ,T − t) = RtN

w − (µ+ γσ2A)(T − t)√

σ2A(T − t)

+ Rte

2γy+ 2µy

σ2A N

−w − 2y − (µ+ γσ2A)(T − t)√

σ2A(T − t)

.

(16)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 28 / 42

Page 79: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Weak Covenant Model Price - Sketch of Proof

We can also complete this calculation using the fact that e−rf tRt is a localmartingale, and so for the bounded stopping time τ := min τDP ,T,

e−rf tRt = Et [e−rf τRτ ] = Et [e

−rf τRτ1τ≤T] + Et [e−rf τRτ1τ>T]

= Et [e−rf τDP RτDP

1τDP≤T] + Et [e−rf TRT1τDP>T]

(17)

It follows that

Bt = Rt + Be−rf (T−t) · Pt [AT ≥ B, τDP > T ]

− Rt

Aγte−(rf−δ− 1

2σ2R(1−ρ2

A,R))(T−t) ·∫ ∞DP

aγPt

[AT ∈ da, τDP > T

](18)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 29 / 42

Page 80: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Weak Covenant Model Price - Sketch of Proof

We calculate the remaining integral using the identities from the DoubleLookbacks paper. We begin with a standard Brownian motion W on aprobability space and define

Xt = µt + σWt

τa = min t : Xt = aX t = min

0≤s≤tXs

g(x , y , t, µ) :=1

σ√

tφ(x − µt

σ√

t

)(1− exp

(4y 2 − 4xy

2σ2t

)) (19)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 30 / 42

Page 81: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Weak Covenant Model Price - Sketch of Proof

The authors of the Double Lookback paper prove that

P [Xt ∈ dx ,X t ≥ y ] = g(−x ,−y , t,−µ)dx

=1

σ√

tφ(−x + µt

σ√

t

)(1− exp

(4y 2 − 4xy

2σ2t

))

=1

σ√

tφ(− (x − µt)

σ√

t

)(1− exp

(− 4y(x − y)

2σ2t

))(20)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 31 / 42

Page 82: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Weak Covenant Model Price - Sketch of Proof

By setting

µ := rf −1

2σ2A

σ := σA

(21)

it follows that

Xt = lnAt

A0

X t = min0≤s≤t

lnAs

A0

(22)

and so for y = ln DPA0

,

At = A0eXt

τDP > T =

XT > ln

(DP

A0

)=τXy > T

.

(23)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 32 / 42

Page 83: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Weak Covenant Model Price - Sketch of Proof

Using this notation, we combine the above to retain,

∫ ∞DP

aγ · Pt

[AT ∈ da, τDP > T

]= Et

[AγT1τDP>T

]

= E0

[eγ

(ln (At)+XT−t

)1τXy >T−t

]

=

∫ ∞y

eγ(x+lnAt)P0

[XT−t ∈ dx , τXy > T − t

]

= (At)γ

∫ ∞y

1√2πσ2

A(T − t)e

(− (x−µ(T−t))2−2σ2

Aγ(T−t)

2σ2A

(T−t)

)dx

− (At)γ

∫ ∞y

1√2πσ2

A(T − t)e

(− (x−µ(T−t))2+4y(x−y)−2σ2

Aγ(T−t)

2σ2A

(T−t)

)dx .

(24)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 33 / 42

Page 84: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Consistency: Reduction to 1-D Black-Cox Model

As a result of the closed form above for Bt,T , the consistency of the modelas T → t follows quickly:

Lemma

limT→t+ B2d Black Coxt,T = B.

Proof.

Without loss of generality, we assume t = 0. As y ,w < 0,

limT→0+

[N

−w − µT√σ2AT

− e2µy

σ2A N

−w − 2y − µT√σ2AT

] = 1

limT→0+

[N

w − (µ+ γσ2A)T√

σ2AT

+ e2γy+ 2µy

σ2A N

−w − 2y − (µ+ γσ2A)T√

σ2AT

]

= 0.(25)

and our result follows.

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 34 / 42

Page 85: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Consistency: Reduction to 1-D Black-Cox Model

B1d Black Coxt,T = lim

(Rt ,ρA,R ,σR)→(At ,1,σA)B2d Black Coxt,T

= Be−rf (T−t)N

− ln BAt− µ−(T − t)√σ2A(T − t)

− Be−rf (T−t)

(DP

At

) 2µ−

σ2A N

− ln BAtDP2 − µ−(T − t)√

σ2A(T − t)

+ At

[N

ln BAt− µ+(T − t)√σ2A(T − t)

+(DP

At

) 2µ+

σ2A N

− ln BAtDP2 − µ+(T − t)√

σ2A(T − t)

].µ+ = rf +

1

2σ2A

µ− = rf −1

2σ2A.

(26)Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 35 / 42

Page 86: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Credit Spread: 2-D Black-Cox Model

By appealing to the martingale nature of recovery, and Bayes’ Theorem,

Y2dBC (t,T ) =1

T − tln

(1

1− Et [Loss | Default] · Pt [Default]

)Et [Loss | Default] = Et

[ B − RT

B| Default

]= 1− Et [e

−rf (T−t)RT | Default]

Be−rf (T−t)

= 1−Rt − Et [e

−rf (T−t)RT1 No Default]

Be−rf (T−t)Pt [Default]

= 1− v(Rt ,w , y ,T − t)

Be−rf (T−t)[1− u(w , y ,T − t)]

Pt [Default] = 1− u(w , y ,T − t) = PD1dBCt,T .

(27)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 36 / 42

Page 87: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

CDS: 2-D Black-Cox Model

Without loss of generality, we can define the CDS Premium at time 0.Under the usual (EPP) framework for continuous payment, premium rateP0,T

=E0[e−rf τDP (B − RτDP

)1τDP≤T] + E0[e−rf T (B − RT )1AT<B,τDP>T]∫ T0 e−rf s P0[τDP > s]ds

=E0[e−rf τDP B1τDP≤T] + E0[e−rf T B1AT<B,τDP>T]− v(R0,w , y ,T )∫ T

0 e−rf s P0[τDP > s]ds

=B(E0[e−rf τDP 1τDP≤T] + e−rf T P0[AT < B, τDP > T ]

)− v(R0,w , y ,T )∫ T

0 e−rf s P0[τDP > s]ds(28)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 37 / 42

Page 88: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

CDS: 2-D Black-Cox Model

It follows that

P0,T =B(h(A0,T ) + e−rf T [F (T )− u(w , y ,T )]

)− v(R0,w , y ,T )∫ T

0 e−rf sF (s)ds

= rfB(h(A0,T ) + e−rf T [F (T )− u(w , y ,T )]

)− v(R0,w , y ,T )

1− e−rf TF (T )− h(A0,T )

F (s) = N

−w + µs√σ2As

− e2µy

σ2A N

−w + 2y + µs√σ2As

h(A0,T ) = E0[e−rf τDP 1τDP≤T]

=A0

[N( ln B

A0−(rf + 1

2σ2A)T√

σ2AT

)−(DPA0

) 2(rf + 12σ

2A)

σ2A N

(2 ln DPA0−ln B

A0+(rf + 1

2σ2A)T√

σ2AT

)]DP

.

(29)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 38 / 42

Page 89: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

CDS: 2-D Black-Cox Model

It follows that

P0,T =B(h(A0,T ) + e−rf T [F (T )− u(w , y ,T )]

)− v(R0,w , y ,T )∫ T

0 e−rf sF (s)ds

= rfB(h(A0,T ) + e−rf T [F (T )− u(w , y ,T )]

)− v(R0,w , y ,T )

1− e−rf TF (T )− h(A0,T )

F (s) = N

−w + µs√σ2As

− e2µy

σ2A N

−w + 2y + µs√σ2As

h(A0,T ) = E0[e−rf τDP 1τDP≤T]

=A0

[N( ln B

A0−(rf + 1

2σ2A)T√

σ2AT

)−(DPA0

) 2(rf + 12σ

2A)

σ2A N

(2 ln DPA0−ln B

A0+(rf + 1

2σ2A)T√

σ2AT

)]DP

.

(29)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 38 / 42

Page 90: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Conclusions and Next Papers?

Have extended 2d models to include clsoed form solutions (bothMerton and Black-Cox)

Recovery is unbounded. Historically possible (foreclosure and bankfees.)

Can include bounded recovery in this framework, work underway.

Can extend to other products, such as Credit Linked Notes.

Calibration to market data for both CDS and Bond prices (anytakers?)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 39 / 42

Page 91: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Acknowledgements

Gary Parker (SFU)

Yang Wang (MSU)

Keith Promislow (MSU)

Emil Valdez (MSU)

Steven Shreve (CMU)

Dmitry Kramkov (CMU)

Joe Campolieti (WLU)

Peter Carr (Morgan Stanley)

All Of You!!

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 40 / 42

Page 92: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

References

V.V. Acharya, T.S. Bharath, & A. Srinivasan, Does industry-widedistress affect defaulted firms? Evidence from creditor recoveries,Journal of Financial Economics 85 (2007) 787-821.E.I. Altman, B. Brady, A. Resti, & A. Sironi, The Link BetweenDefault and Recovery Rates: Theory, Empirical Evidence andImplications, The Journal of Business ,Vol. 78, No. 6, November2005.Black, F., & Cox, J.C., Valuing Corporate Securities: Some Effects

of Bond Indenture Provisions, Journal of Finance 31, 1976, 351-367.Levy, A. and Hu, A., Incorporating Systematic Risk in Recovery:Theory and Evidence, Modeling Methodology, Moody’s KMV, 2007.G. Giese, The Impact of PD/LGD Correlations on Credit RiskCapital, Risk, April 2005, pp 79-85.He, H, Keirstead, W.P., and Rebholz, J., Double Lookbacks,Mathematical Finance, Vol. 8, No 3 1998 (201-228).Merton, R., On the Pricing of Corporate Debt: the Risk Structure ofInterest Rates, Journal of Finance 29, 1974, 449-470.Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 41 / 42

Page 93: Bond and CDS Pricing with Stochastic Recovery : Moody's …users.math.msu.edu/users/albert/CreditTalk.pdf · Moody’s PD-LGD Correlation Model ... Department of Statistics and Probability

Where it all started!! Albert and Nick (circa 1999)

Albert Cohen (MSU) Bond and CDS Pricing with Stochastic Recovery : Moody’s PD-LGD Correlation ModelSimon Fraser University 42 / 42