Week 9 – Credit risk

MIT Sloan School of Management

2

Outline

• Statistical approach to default risk

• Decomposing the credit spread
• Structural approach to default risk
• Simple binomial example
• Merton model and extensions
• Credit derivatives
Corporate debt

• Divided into two broad credit quality categories by rating agencies

1. Investment grade (IG)
2. Non-investment grade, high yield (HY), or “junk”
• Much higher historical default rates
• Original issue vs. “fallen angels”
• Strong historical return performance

• Default risk manifests itself as:

• Downgrade risk
• Event risk
• E.g., legal changes, surprise actions by managers
• Liquidity risk
• Riskier bonds tend to be less liquidity
3
Default and recovery rates:
key statistics for assessing credit risk
• Default rate
• Probability of a default event
• Usually stated as an annual rate
• What is a default event depends on who is defining it; rating agencies provide definitions

• Recovery rate
• Amount expected to be recovered as a fraction of what is owed
• Sometimes expressed in present value terms
• Can be based on price of a defaulted bond relative to principal

• recovery rate = (1 – loss rate)

• Loss rate is also called “loss given default”
• Term structure of default and recovery
• Expected rates may vary over the life of a security
5

Default rates vary enormously:

over time, by credit rating, and with the business cycle
6

Credit spreads on risky debt

• The “credit spread” (or yield spread) is the difference in the yield to maturity on
a risky bond and on a Treasury bond of similar maturity
• If it is an “options-adjusted spread” it assumes no other embedded options (e.g., no
prepayment or call option)

• For a given credit rating, observed yield spreads vary with maturity
• The yield spread usually increases with maturity over moderate horizons
• Sometimes credit spreads are decreasing at very long maturities
• These patterns partially reflect compositional differences in firms borrowing at different
maturities
8

Decomposition of Credit Spreads

• What determines the YTM spread between Treasury’s and defaultable

securities of the same maturity?

• Conceptually the spread has several components:

( “*” denotes a risk-
• Expected losses Prob*(Default) x neutral variable)
• Premium for market risk () E*(Loss given default)
• “Liquidity premium”
• Non credit features (e.g. tax treatment, embedded options)
BBB
Corporate
Spread
YTM
Treasury’s

maturity
9

A simple valuation model for risky bonds

• This simple model illustrates that expected cash flows are not the promised
cash flows, and the yield to maturity is not the expected return.

• In the model, the difference between the expected return, r, and the YTM, y, is
influenced by the default rate, the recovery rate, and the bond maturity.

• The model is implemented in the spreadsheet default.xls (on webpage)

10

A simple valuation model for risky bonds

• Consider a risky T-period coupon bond, with coupon rate, c, and face value F=1.
• Assumptions
• A constant default rate each period, d
• The probability of no default from time 0 to time t is (1-d)t-1
• A constant recovery rate, g
• An expected return r equal to the risk-free rate plus a premium for market risk,
e.g., r = rf + β(E(rm) – rf)

• Investors discount expected cash flows at r to determine the price of the bond
Basis for
numerator
in bond
Sharpe ratio
11

A simple valuation model for risky bonds

• Default rate d
• Recovery rate g
• Expected return T   ( dg (1  c )  (1  d ) c )  (1  d )T
r P    (1  d ) i 1 

i 1 (1  r ) i  (1  r )T
• Coupon rate c 
• Maturity T
• face value 1; P = price per $1 face value

(1  d ) i 1 is the probability that the bond is still outstanding at time i

(dg (1  c)  (1  d )c)
i
is the pv of the expected coupon plus recovery at time i
(1  r )
(1  d )T
is the pv of the expected $1 principal payment at time T
(1  r )T

T c  1
The yield to maturity takes promised payments as certain: P  

i 1  (1  y )
i  (1  y )T

12

Structural models of credit risk

13

Risky corporate debt and equity payoff diagrams

Notes:

1. Sum of debt and equity is asset

value
2. Underlying in these diagrams is firm
asset value (not equity)
14

Decomposing payoff on risky debt

• A risky zero coupon corporate bond is equivalent to a portfolio that includes:

• a long position in a risk-free zero coupon bond
• a short position in a put option on the assets of the firm, with a strike price equal to the face
value of the bond

Payoff
on
risky
debt

Firm asset value

• A loan guarantee is equivalent to a put option on firm assets. The guarantor

writes the put option in exchange for a fee (premium).
Valuing loan guarantees as put options
Example Today XYZ Co. has a market value of $100 million, and next year the
company’s assets will take on one of two values: $140 million or $70
Binomial guarantee pricing: million:

$140
$100
$70

XYZ also has guaranteed debt with a face value of $90 million
(covering principal and interest), coming due next year.

What is the value of the guarantee (from perspective of

guarantor)?

The payments of the guarantor will be $0 if the assets are worth

$140 or -$20 if the assets are worth $70.

Then the payoffs for the guarantor look like:

$0
P
-$20

This is a written put option on the company’s assets with a strike15

price of $90.
16
Valuing loan guarantees as put options
Assume the price of a 1-year risk-free bond is $95 per $100 face. Its value
is represented by:

$100

$95

$100

The value of the option can be replicated using the information on asset
value and the risk-free bond:

We require that the payments match in the good and bad state of the world:

X100 + Y140 = 0

X100 + Y70 = -20

These two linear equations in two unknowns can be solved for X and Y to
yield:

X = -.4
Y = .2857

The price of this portfolio, based on the $95 price of the bond and the $100
current asset value, is -.4($95) + .2857($100) = -$9.43

(The guarantor has a highly levered position in the assets of the firm!)
17

The Merton Model

• Today is t = 0 and consider a firm with current assets V0 = E0 + D0

• Assume the firm’s (market value) return on assets is log-normally distributed:

• The assets for financed with equity and debt. The debt is a zero coupon bond with face value F
and maturity T
18

The Merton Model: Valuing equity as a call option

Notice the volatility in

this formula is for
assets, not equity
19

The Merton model: Volatility of levered equity

• What is the volatility of levered equity?
• We know the relation between the volatility of a call option and its underlying volatility is given by
its vega. Applying the same formula implies:

V  N (d1 )
• This can also be written as e  
E

• As the value of equity relative to assets falls, its volatility increases

• This is known as the “leverage effect” on equity volatility, first noted by Fischer Black
• Important implication: As a firm becomes distressed, analyzing its expected returns using the
CAPM becomes highly problematic because its beta not well-approximated by a constant
20

The Merton model: Volatility of levered equity

21

The Merton Model: Inferring asset value and volatility

• We have two non-linear equations in two unobservable variables, V0 and σ

E=

And
V0  N (d1 )
e  
E

• Solve simultaneously given other parameters to find V0 and σ

22
The Merton model: Value of debt
What is the value of defaultable debt in the model?

Note: The put option

also represents the
value of a guarantee
on that debt.
 Sample Time Path of Firm Assets

0.08

0.07

0.06

0.05
value

Loss given
0.04
default on this
0.03 path
0.02

0.01

0
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71

Time Debt
Maturity

23
24

The Merton model: Credit spreads

25

The Merton model: Credit spread over time

26

The Merton model: Relative pricing of junior and senior debt

• The Merton model can be used to find the price of bonds with different priorities
• For example, suppose that a firm issues two bonds: one senior and one junior
(also called subordinated), with face value FS and FJ
• At maturity we have the following “waterfall” of payoffs
27
Merton model: Relative pricing of junior and senior debt

The payoffs on more complicated structures can also be valued using Monte Carlo simulation of
the risk-neutral asset process
28

The Merton model: Extensions

• Many extensions of this basic model exist, including:
• Early bankruptcy
• American put option: there is a lower bound Vb to assets so that if V(t) < Vb the firm is bankrupt
• Stationary leverage
• Merton model indicates often counterfactual decline in leverage over time
• Unobservable firm value
• Investors can only rely on noisy accounting information to estimate V(t): the default barrier could be
closer than you think
• Can be modeled by incorporating jumps in asset process
• Stochastic interest rates
• interest rates follow processes like the ones we saw last week
29

Example: KMV model and distance to default

• A generalization of the Merton model is the KMV model, used commercially by Moody’s

• Inputs are stock prices, book value of liabilities, stock volatility, interest rates.

• To get from book liabilities to market liabilities and therefore initial market assets, solve two
simultaneous non-linear equations:

Equity value = F[asset value, asset vol, cap str, r]

Equity volatility = F[asset value, asset vol, cap str, r]

• Unknowns are asset value and asset vol

• “Default point” defined by a rule for assets relative to liabilities

• An abstraction, proxy for the complicated question of when a firm really is likely to default
• E.g., assets fall to < 70% of short-term + ½ long-term book liabilities

• These quantities are used as inputs to find “distance to default.”

Example: KMV model and distance to default
• The likelihood and severity of default depends on asset volatility and leverage
• The “Distance to Default” measure is defined by:

DTD = [(mkt value assets) – (default point)]

[(mkt value assets)(asset volatility)]

• Interpretation: the number of standard deviations the firm is away from default.
• KMV has a proprietary algorithm to map DTD into the “Expected Frequency of Default” (EDF)
• Simple implementations of the model tend to under-estimate default probabilities over short
horizons
• Can address by modeling shocks as having a jump component
• Jumps often modeled as Poisson process
• Also can address with a stochastic default barrier

30
31
Valuing loan guarantees on a binomial tree
• When the default barrier is more complicated and default can occur at any time,
a binomial pricing approach is a natural choice.
• For example, consider the case of America West Airlines (AWA), that received a
loan guarantee after 9/11 from the U.S. government.
• In 2002, AWA was the eighth largest passenger airline in the U.S. Softening economic
conditions had already severely reduced airline revenues. Following 9/11, Moody's
downgraded AWA from B1 in April 2001 to Ca on November 21, 2001.
• After the terrorist attacks, Congress enacted the Air Transportation Safety and System
Stabilization Act, allowing airlines to apply for credit guarantees.
• AWA received final approval from the Air Transportation Stabilization Board (ATSB) for a
government loan guarantee of $380 million in January 2002.
• It paid fees and gave the government warrants in compensation.

• What was that assistance worth on net?

32

Valuing loan guarantees as put options

• This example is based on analysis in “Estimating the Value of Subsidies for Federal Loans
and Loan Guarantees,“ A Congressional Budget Office Study available on its website

• Key inputs:
• Expected return on equity (based on estimated equity beta)
• Volatility equity return (estimated from historical data)
• risk-free rate
• default/prepayment rules
• loan maturity
• Loan coupon rate
• Rest of firm capital structure
33

Valuing Loan Guarantees as Put Options

• Stage 1 estimation: use equity data to find asset stats

• volatility of firm asset value Estimated first stage
• current value of firm assets using Merton model

• average return (physical) on firm assets

• Derived from implied asset beta; used for risk assessment, e.g., VaR

dAt / At  (rA   )dt   AdZ t

• risk-neutral return on firm assets
• Used for pricing guarantees

dAt / At  (r f   )dt   AdZt

Pricing

Auuu
Auu Denotes nodes
where a loss occurs
Au Aduu according to trigger rule.
A0
Aud Loss(t) = F – A(t)
where F is payoff value of
Addu
guaranteed debt
Ad
Add Losses are weighted by
probability of occurrence and
Addd discounted to the present.
Probabilities are “risk-neutral”,
Au  A0 (1  (r f   )t   A t ) so discounting is at risk-free
rate.
Ad  A0 (1  (r f   )t   A t )
34
Pricing
e.g., 0 coupon bond F = 900, T=3; and rf=.05, p*(up) =.5

1520 Denotes nodes

1350 where a loss
occurs.
1200 1285
1000
1050
Loss(t) = F – A(t)

897
910
790 Losses are weighted by
probability of occurrence and
630 discounted to the present.
Probabilities p* are “risk-
neutral”, so discounting is at
risk-free rate.

35
Pricing
0 coupon bond F = 900, T=3; and rf=.05, p(up) =.5

Losses are weighted by

probability of occurrence and
0 discounted to the present.
Probabilities are “risk-neutral”,
0 so discounting is at risk-free
rate.

-3
PV(losses) = pr(3d)(-270)/(1.05)3 +
pr(2d1u)(-3)/(1.05)3 =
$-30.126
-270 Risky bond value = 900/(1.05) 3 –
t=0 t=1 t=2 30.13
30.12634 0.680272 0
62.58503 1.428571
= $777.46 - $30.13 = $747.33
130

36
Valuing loan guarantees as put options

37
Valuing loan guarantees as put options

Note: The same model can be

used to value guarantee fees and
warrants, whose payoffs are also
contingent on the firm’s asset
value.

38
Credit derivatives

39
40

Credit derivatives
• Includes the many types of contracts whose payoffs depend on credit events:
• Loan and bond guarantees
• Credit default swaps (CDS)
• Structured credit products (e.g., CLOs, MBS)
• Total rate of return swaps
• Credit-linked notes
• Trading is mostly over-the-counter
• Used primarily by financial institutions to manage credit risk exposures
• Exponential growth prior to 2008 financial crisis. After that sharp falloff, and now
slower growth
• Excessive CDS exposure brought down insurance giant AIG, which was bailed out
41

Credit derivatives

• Credit derivatives generally involve two counterparties

• Protection buyer
• Protection seller

• There is also an underlying “reference entity” whose behavior determines the

cash flows on the credit derivatives
• E.g., a skipped payment on a debt obligation by Boeing would trigger a payment on Boeing CDS

Question: Under what conditions is the protection buyer hedging? Under what
conditions is the protection buyer speculating?
Credit Default Swaps (CDS)
If insurance fee is paid over time, it’s a “swap”
If insurance fee is paid up front, it’s an “option”
This type of contract can be written on
anything
 e.g., loan, bond, sovereign risk, credit
exposure on derivative contract

42
43

CDS structures
• Fixed maturity
• Fee paid until maturity or default
• Various triggering events:
• Bankruptcy
• Credit event upon merger
• Downgrade
• Failure to pay
• Repudiation
• Restructuring
• Payment acceleration
• Must be verifiable public announcement of the event
• Various alternative settlement rules; contracts can differ:
• Cash settlement = face value – market value at trigger event
• Market value determined by average of dealer quotes
• Physical delivery: deliver defaulted bond for face value (may be multiple deliverables, and
hence cheapest to deliver option)
• Digital settlement: fixed payment in event of trigger event
• Contracts usually governed by ISDA rules
44

CDS structures
• Single name and indices
• “Single name” means debt is from one company
• Indices
• Payoff based on defaults on a pool of bonds
• CDX (e.g., iTraxx)

• Structured products based on index give rise to

• synthetic CDOs
• Nth to default bonds
• Other repackaging of risk

• To read about credit indices, you can download “Credit Indices: A Primer” from IHS Markit’s website
45

Single-name credit default swap

• This is essentially an insurance contract.
• Protection buyer pays periodic premium U on a notional amount of protection
for a period of length T, say on $100 face value of the underlying entity’s debt
• In the event the underlying entity defaults on the debt at time τ and the recovery
amount is Y(τ) per $100 face value, the protection buyer is made whole
• If the contract calls for cash settlement the protection buyer gets 100 - Y(τ)
• If the contract specifies delivery the protection buyer exchanges $100 face value of the bond
which is worth Y(τ) for $100.
46

CDS pricing
• Bottom line: The credit spread is approximately the fair CDS premium.
• As for loan guarantee, writing protection is like being long the risky bond and short a risk-free bond
• Conversely, buying a CDS is like shorting the risky bond and buying a risk-free bond (see picture)
• Simple characterization is only precisely true in special cases; more complicated model is needed to account
for different liquidity, counterparty risk, fixed vs. floating

In event of default, protection buyer

+P P gets difference between promised
payment P and recovery net of
r r r r premium

-recovery
-r-s -r-s -r-s -r-s

-P
47

CDS Pricing
• Approach 1: A CDS is like a credit guarantee. The present value of the insurance can be
estimated using options pricing methods (e.g., structural models like the extensions of the Merton
model discussed earlier).

• Approach 2: Price delivery-settled swap by “no arbitrage” with reference to underlying securities:*
(corresponds to graph on previous slide)
• Assume CDS written on floating rate corporate bond “C” with spread S over risk-free floating rate
bond, originally priced at par.
• Ignoring transactions costs, the same protection is obtained by the CDS buyer by shorting the
risky bond C, and investing in a par value default-free floating rate note.
• Hold portfolio through maturity or credit event.
• Net spread paid is S until termination.
• In event of credit event, liquidate portfolio and get face value of risk-free bond – value of
defaulted bond. Same is CDS payoff!
• This assumes credit event occurs on a coupon reset date, so risk-free bond is priced at par
• It follows that S, the credit spread, is the fair premium rate on the swap.
48

Multi-name CDS

• Suppose an investor holding a portfolio of defaultable bonds is worried about default

• The investor can
(1) Purchase a CDS for each bond in the portfolio; or
(2) Purchase insurance on the portfolio itself
Question: Which is more expensive?
49

Multi-name CDS

• Example: Nth-to-default basket default swaps

• Not all firms in a portfolio will default at the same time, particularly if the credits are diversified
across country, industry, etc.
• It is popular to purchase protection that just pays off after some number of defaults have
occurred
• This is much cheaper than buying protection for each individual credit
• The basket spread (premium) depends on:
• Number of credits: more credits => credit event more likely => more costly to insure
• Credit quality and recovery rates
• Default correlation across underlying reference entities (important!)
50

Summary

• This week we looked at how statistical and structural models can be used to price credit risk

• Statistical models derive default and recovery rates from data on borrower characteristics,
including leverage ratios, credit ratings, tangible collateral, etc.

• Structural models infer default and recovery rates based on the stochastic structure of a
borrower’s assets and a default barrier

• Both approaches predict the effect of credit risk on the value of credit-sensitive securities

• For defaultable bonds, both should produce similar answers if properly calibrated

• The structural approach is more flexible and better suited for pricing more complex credit derivatives

• In practice, hybrids of the two approaches are often used and give better results.