• Credit Risk

– Credit risk and Default Probability

• Credit Ratings
– In the S&P/Fitch rating system, AAA is the best rating. After that
comes AA, A, BBB, BB, B, CCC, CC, and C
– The corresponding Moody’s ratings are Aaa, Aa, A, Baa, Ba, B,
Caa, Ca, and C
– Bonds with ratings of BBB (or Baa) and above are considered to
be “investment grade” (IG bonds); The rest are “junk” bonds
• Default Probabilities
– Predict default and estimate DP
» Altman’s Z-score (using accounting ratios)
» use historical data to estimate historical DP
» use credit spreads (credit risk priced by the market)
» use Merton’s model (modeling equity as a call option on the
assets of the firm)
default correlation plays an important role in determining the
worst-case default rate
– Altman’s Z-score (Manufacturing Companies)
• X1=Working Capital/Total Assets
• X2=Retained Earnings/Total Assets
• X3=EBIT/Total Assets
• X4=Market Value of Equity/Book Value of Liabilities
• X5=Sales/Total Assets

Z = 1.2X1+1.4X2+3.3X3+0.6X4+0.99X5
If the Z > 3.0 default is unlikely; if 2.7 < Z < 3.0 we should be on
alert. If 1.8 < Z < 2.7 there is a moderate chance of default; if Z < 1.8
there is a high chance of default

• 80-90% accuracy in predicting default within a year

• All these ratios are still relevant.
• Perhaps, ML techniques can be used to obtain a better
predictor.
– Historical default probabilities
• Cumulative Average Default Rates (%) (1970-2015, Moody’s)

1 2 3 4 5 7 10
Aaa 0.000 0.011 0.011 0.031 0.087 0.198 0.396
Aa 0.022 0.061 0.112 0.196 0.305 0.540 0.807
A 0.056 0.170 0.357 0.555 0.794 1.345 2.313
Baa 0.185 0.480 0.831 1.252 1.668 2.525 4.033
Ba 0.959 2.587 4.501 6.538 8.442 11.788 16.455
B 3.632 8.529 13.515 17.999 22.071 29.028 36.298
Caa-C 10.671 18.857 25.639 31.075 35.638 41.812 47.843

The table shows the probability of default for companies starting with a
particular credit rating
A company with an initial credit rating of B has a probability of 8.529%
of defaulting by the end of the second year, 13.515% by the end of the
third year, and so on (PD during the third year = 13.515-8.529=4.99%)
• The probability of a B-rated bond defaulting during the third year
conditional on no earlier default is 0.0499/0.916 or 5.45%
– Hazard Rate
• The hazard rate (also called default density), λ(t), at time t, is
defined so that λ(t)∆t is the conditional default probability for a short
period between t and t+∆t
• If V(t) is the probability of a company surviving to time t
V(t+∆t) – V(t) = -𝜆𝜆(t)V(t) ∆t
dV(t)/dt = -𝜆𝜆(t)V(t)
𝑡𝑡
− ∫0 𝜆𝜆 𝜏𝜏 𝑑𝑑𝜏𝜏 �
This leads to 𝑉𝑉 𝑡𝑡 = 𝑒𝑒 = 𝑒𝑒 −𝜆𝜆𝑡𝑡
1 𝑡𝑡
𝜆𝜆̅ = 𝑡𝑡 ∫0 𝜆𝜆 𝜏𝜏 𝑑𝑑𝜏𝜏 is the average hazard rate between 0 and t.
• Q(t) is the probability of defaults by time t,
�
𝑄𝑄 𝑡𝑡 = 1 − 𝑉𝑉 𝑡𝑡 = 1 − 𝑒𝑒 −𝜆𝜆𝑡𝑡 ; 𝜆𝜆̅ = −𝑙𝑙𝑙𝑙(1 − 𝑄𝑄 𝑡𝑡 )/𝑡𝑡
– Recovery Rate
• The recovery rate for a bond is usually defined as the price of the
bond immediately after default as a percent of its face value
Recovery Rates; Moody’s: 1982 to 2015
Class Mean(%)

First Lien Bond 53.4

Second Lien Bond 49.7

Senior Unsecured 37.6

Senior Subordinated 31.1

Subordinated 31.9

Junior Subordinated 24.2

• Recovery rates tend to decrease as default rates

increase (both for mortgage default and corporate
bond default)
– Default Probability Using Credit Spreads
• Credit Default Swap
Like an insurance contract that pays off in case of a default
• If the reference entity (a country or company) defaults,
– Physical settlement: the buyer of the CDS has the right to sell to
the seller the bonds issued by the reference entity for their face
value
– Cash settlement: based on the difference between the face value
and the cheapest-to-deliver bond after the credit event (based on
the auction of the bonds within a certain maturity range)
– The spread payments stop after the default
• Example: Notional principal (e.g. $100 million) and maturity specified;
Protection buyer pays a fixed rate of 120 bps on the notional principal
(the CDS spread)
– Insurance premium 1.2 million per year
– If a default occurs and supposed the cheapest-to-deliver bond
issued by the reference entity is worth 40 cents per dollar, the
seller of CDS has to make a payment equal to $60 million (the
insurance is to bring the value back up to $100 million).
– Relating credit spread to the hazard rate
• Suppose s(T) is the credit spread for maturity T, s(T) should
compensate for the average loss rate.
• The average loss rate between time zero and time T is
approximately
𝜆𝜆̅ 𝑇𝑇 1 − 𝑅𝑅 = 𝑠𝑠 𝑇𝑇
where R is the recovery rate
Thus
𝜆𝜆̅ 𝑇𝑇 = 𝑠𝑠(𝑇𝑇)/ 1 − 𝑅𝑅
• This estimate is quite accurate in most situations
• Example: 1-year and 2-year bonds yield 150 and 180 more than the
risk-free rate, respectively. The recovery rate is estimated at 40%.
– Average 1 year hazard rate = 0.015/(1-0.4) = 2.5%
– Average 2 year hazard rate = 0.018/(1-0.4) = 3.0%
– Average 2nd year hazard rate = 3.5%
– Matching Bond Prices
• For more accuracy we can work forward in time choosing hazard
rates that match bond prices, using the bootstrap method
• Example:
– Given risk-free rate = 5% and
1-year, 2-year, 3-year bond yields: 6.5%, 6.8%, 6.95%
bond face value = 100, coupon rate 8% (semi-annual comp.)
– We can obtain
» bond prices (from the yields): $101.33, $101.99, $102.47
𝐵𝐵 = ∑𝑛𝑛𝑖𝑖=1 𝑐𝑐𝑖𝑖 𝑒𝑒 −𝑦𝑦𝑡𝑡𝑖𝑖 ; ci --- cash flow at time ti;
y --- bond yield
» bond prices (if they were risk-free, discounted using the
risk-free rate of 5%) would be: $102.83, $105.52, $108.08
– The present value of the expected default losses (the difference
between bond prices assuming they are risk-free and the actual
bond prices): $1.50, $3.53, and $5.61 for 1-year, 2-year, and 3-
year bonds
• Example (cont.)
– Consider a 1-year bond and assume that the recovery rate is 40% ---
we want to relate the hazard rate λ1 to the expected default loss of
$1.50
– Assume that the defaults can happen only at the midpoints of 6-
month intervals (default times are in 3 months and 9 months)
» This is a reasonable numerical approximation --- it would be
exact if the present value of the loss linearly depends on the time
of default; essentially, it is the first-order approximation. Another
way to improve is to consider more default points
– The risk-free value of the bond at the 3-month point is
4e-0.05x0.25+104e-0.05x0.75 = $104.12
– The present value of the loss if there is a default at the 3-month point
is (104.12-40)e-0.05x0.25 = $63.33
– The risk-free value of the bond at the 9-month point is
104e-0.05x0.25 = $102.71
– The present value of the loss if there is a default at the 9-month point
is (102.71-40)e-0.05x0.75 = $60.40

x x
6 12
• Example (cont.)
– The probability of default in the first 6 months: 1 − 𝑒𝑒 −0.5𝜆𝜆1
– The probability of default during the following 6 months:
1 − 𝑒𝑒 −𝜆𝜆1 − 1 − 𝑒𝑒 −0.5𝜆𝜆1 = 𝑒𝑒 −0.5𝜆𝜆1 − 𝑒𝑒 −𝜆𝜆1
– The hazard rate λ1 must satisfy
1 − 𝑒𝑒 −0.5𝜆𝜆1 × 63.33 + (𝑒𝑒 −0.5𝜆𝜆1 −𝑒𝑒 −𝜆𝜆1 ) × 60.40 = 1.50
This gives λ1 =2.46%
– This is compared to 𝜆𝜆̅ 𝑇𝑇 = 𝑠𝑠(𝑇𝑇)/ 1 − 𝑅𝑅 = 1.5/(1-0.4)=2.5%
– With λ1 determined, we can obtain hazard rate λ2 …
» The hazard rate for the second year depends both on λ1
and λ2
– Risk-Free Rates
• The risk-free rate for estimating credit spreads and default
probabilities is traditionally the LIBOR/swap zero rate. The risk-free
rate implied from the CDS (the difference between the risky bond
yield and the CDS spreads) is about 10 bps below the LIBOR/swap
rate and is close to the OIS rate (SOFR in US, SORA in Singapore).
– LIBOR is being phased out starting at the end of 2021
• Asset swaps (structured, for example, to swap the coupon payment
of a bond with LIBOR plus spread) provide a direct estimate of the
spread of bond yields over LIBOR.

– Real World vs Risk-Neutral Default Probabilities

• The default probabilities backed out of bond prices or credit default
swap spreads are risk-neutral default probabilities
• The default probabilities backed out of historical data are real-world
default probabilities
– Comparing two DP estimates

Cumulative 7-year default Average 7- year credit

Rating
probability(%): 1970-2015 spread (bp): 1996-2007
Aaa 0.198 35.74
Aa 0.54 43.67
A 1.345 68.68
Baa 2.525 127.53
Ba 11.788 280.28
B 29.028 481.04
Caa 41.812 1,103.70
Data from Moody’s and Merrill Lynch
– Comparing two DP estimates (cont..)

Rating Historical Hazard Rate Hazard Rate from bonds Ratio Difference
(% per annum)1 (% per annum)2
Aaa 0.028 0.596 21.0 0.568
Aa 0.077 0.728 9.4 0.651
A 0.193 1.145 5.9 0.952
Baa 0.365 2.126 5.8 1.761
Ba 1.792 4.671 2.6 2.879
B 4.898 8.017 1.6 3.119
Caa 7.736 18.395 2.4 10.659

Historical Hazard Rate: Calculated as −[ln(1-Q(7))]/7 where Q(7) is the

Moody’s 7-year default rate (𝑄𝑄 7 = 1 − 𝑒𝑒 −𝜆𝜆 7 ×7 ). For example, in the
case of Aaa companies, Q(7)=0.00198, then

𝜆𝜆 7 = -ln(0.99802)/7=0.00028 or 2.8bps.

For investment-grade companies, the historical hazard rate ≈ Q(7)/7.

Hazard Rate from bonds: Calculated as s/(1-R) where s is the bond yield
spread and R is the recovery rate (assumed to be 40%).
 – Expected Excess Return (extra risk premium) on Bonds
Rating Bond Yield Spread of risk-free Spread to Extra Risk
Spread over rate used by market compensate for Premium
Treasuries over Treasuries default rate in the (bps)
(bps) (bps)1 real world (bps)2
Aaa 78 42 2 34
Aa 86 42 5 39
A 111 42 12 57
Baa 169 42 22 105
Ba 322 42 108 172
B 523 42 294 187
Caa 1146 42 464 640
• Use the average spread (42bps) of our benchmark risk-free rate
over Treasuries.
Bond spread over treasuries = Bond spread over RFR + 42 bps
For example, for Baa: 169 ≈ 127 + 42
• Spread to compensate for default rate in the real world: historical
hazard rate times (1-R) where R is the recovery rate. For example,
in the case of Baa, 22bps is 0.6 times 36.5bps; the extra risk
premium = 169 – 42 – 22 = 105 bps
– Possible Reasons for the Extra Risk Premium
• Corporate bonds are relatively illiquid
• The subjective default probabilities of bond traders may be much
higher than the estimates from Moody’s historical data
• Bonds do not default independently of each other. This leads to a
systematic risk that cannot be diversified away. As a result, bond
returns are highly skewed with limited upside. This may be priced in
by the market
– On an individual bond, there might be a 99.75% chance of a
7% return in a year (no default), and a 0.25% chance of a -60%
return in the year (default). The big systematic tail risk (due to
default correlation) can’t be diversified away.
– Which DP Estimates Should We Use?
• Use risk-neutral estimates for valuing credit derivatives and
estimating the present value of the cost of default
• Use real-world estimates for calculating credit VaR and scenario
analysis
– Using Equity Prices: Merton’s Model

• Merton’s model regards equity as an option on the assets of the firm

• In a simplified model, the equity value is
ET = max(VT −D, 0)
where VT is the value of the firm and D is the debt repayment required

• To use the Black-Scholes-Merton option pricing model for the value

of the firm’s equity E0 today. We need the value of its assets today,
V0, and the volatility of its assets, σV

E 0 = V0 N ( d 1 ) − De − rT N ( d 2 )
where
ln (V0 D) + ( r + σ V2 2) T
d1 = ; d 2 = d 1 − σV T
σV T
 • Volatilities are related by (Ito’s Lemma)
𝜕𝜕𝜕𝜕
dV = μVVdt + σVVdw, 𝑑𝑑𝑑𝑑 = … 𝑑𝑑𝑑𝑑 + 𝜕𝜕𝜕𝜕 𝜎𝜎𝑉𝑉 𝑉𝑉𝑑𝑑𝑑𝑑 = … 𝑑𝑑𝑑𝑑 + 𝜎𝜎𝐸𝐸 𝐸𝐸𝐸𝐸𝑤𝑤
∂E
σ E E0 = σV V0 = N ( d 1 ) σV V0
∂V
– Note the delta from BSM: N(d1)
• The two equations can be used to solve for V0 and σV, Given E0 and
𝜎𝜎𝐸𝐸 from the market
• The probability (risk-neutral) of default: Prob(VT < D) = N(-d2)

– An example
• A company’s equity is $3 millions and the volatility of the equity is
80%
• The risk-free rate is 5%, the debt is $10 million and the time to debt
maturity is 1 year
• Solving the two equations (numerically) yields V0=12.40 and
σv=21.23%
• The probability of default is N(−d2) = 12.7%
– Risk-neutral vs. Real World
• The average growth rate of the value of the assets, V, is greater in
the real world than in the risk-neutral world
• This means that V has more chance of dropping below the default
point in a risk-neutral world than in the real world
• This explains why risk-neutral default probabilities are higher than
real-world default probabilities
– Implementation of Merton’s Model
• Choose time horizon and calculate D: cumulative obligations to time
horizon --- This is termed by Moody’s KMV credit risk analytics the
“default point”.
• Use Merton’s model to calculate a theoretical probability of default
• Use historical data or bond data to develop a one-to-one mapping of
theoretical probability from Merton’s model into either the real-world
or risk-neutral probability of default. Use this mapping to predict
future default rates.
– This is possible because Merton’s model and its extension
produce a good ranking of default probabilities.
– Distance to Default (how far away from default)
• Used to describe the output from Merton’s model. As the distance
declines, the company becomes more likely to default. One
definition of the distance to default is simply d2 (N(-d2) is the
theoretical default probability)
𝑉𝑉 𝜎𝜎 2
ln 𝐷𝐷0 + 𝑟𝑟− 𝑉𝑉 𝑇𝑇
2
𝑑𝑑2 = 𝜎𝜎𝑉𝑉 𝑇𝑇

– Extensions
• Use distance to default as the key factor and add new factors
(related to fundamentals of the company or industry) – CRI,
Bloomberg
• How to handle default correlation? --- still open questions
– Default correlation on supply chains is very important. But how
to analyze it?
• Gaussian Copula Application: Credit Default Correlation
– The credit default correlation between two companies is
a measure of their tendency to default at about the same
time
– Default correlation is important in risk management
– It is important for determining the worst-case default rate
(tail risk of a loan portfolio).

• Model for Loan Portfolio Value (Vasicek)

– We map the time to default for company i, Ti, to a new variable
Ui and assume (ai to be the same)
Ui = aF + 1 − 𝑎𝑎 2 Zi, where F and the Zi have independent
standard normal distributions
– The copula correlation is ρ=a2
• Analysis
– The aim
• To determine the probability distribution of the default rate, given the
average default probability PD and default correlation ρ
– Steps to derive the probability distribution
• Map the time to default Ti to Ui
• Assume Ui can be decomposed into a systematic component F and
an idiosyncratic component specific to the company i.
• Obtain the probability of default conditional on macro/systematic
variable F: Q(U|F), or Q(T|F)
– Default probabilities for different years are different simply
because the variable F is different. In our model, the lower the
F(this may indicate a bad economic condition) the higher the
probability (lower F lower U lower T).
• Given the probability distribution of F (here we assume a normal
distribution for F; this is part of the model assumption), we can then
derive the probability distribution of the default rate
– Finally, using the historical data of default rate and the method of
maximum likelihood, we can estimate PD and ρ
– Analysis (cont…)

• To analyze the model, we

– assume the probability distribution of the default rate is the same for
all companies ({𝑎𝑎𝑖𝑖 } are the same)
– calculate the probability that, conditional on the value of F, Ui is less
than some value U

• The company defaults by time T (Ti is less than T) when 𝑈𝑈𝑖𝑖 < U, or
𝑈𝑈−𝑎𝑎𝐹𝐹
𝑍𝑍𝑖𝑖 < [Note: Ui = 𝑎𝑎F + 1 − 𝑎𝑎 2 Zi ]
1−𝑎𝑎2
» Ti and Ui are related by the percentile-to-percentile mapping:
N(Ui)=Q(Ti), where Q(T)=N(U)=PD (𝑈𝑈 = 𝑁𝑁 −1 (𝑃𝑃𝑃𝑃)) is the
unconditional probability of default by time T (if we consider 1-
year default rate, T= 1 year).
– Analysis (cont…)

• Conditional on F the probability of default by time T is

𝑈𝑈 − 𝑎𝑎𝑎𝑎
𝑄𝑄 𝑇𝑇 𝐹𝐹 = 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑇𝑇𝑖𝑖 < 𝑇𝑇 𝐹𝐹 = 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑈𝑈𝑖𝑖 < 𝑈𝑈 𝐹𝐹 = 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝑍𝑍𝑖𝑖 < 𝐹𝐹
1 − 𝑎𝑎 2
𝑈𝑈−𝑎𝑎𝑎𝑎
= 𝑁𝑁
1−𝑎𝑎2
𝑁𝑁−1 (𝑃𝑃𝑃𝑃)− 𝜌𝜌𝐹𝐹
= 𝑁𝑁 [Note that 𝑎𝑎 = 𝜌𝜌]
1−𝜌𝜌
– Low values of F give high default probabilities
– If 𝜌𝜌 = 0, Q(T|F)=PD (Q(T))
– Analysis (cont…)

• Given a distribution of F, we can derive a distribution of the default rate

Cumulative Probability distribution of the default rate,
G(y) = P(Q(T|F) < y)
𝑁𝑁−1 𝑃𝑃𝑃𝑃 − 𝜌𝜌𝐹𝐹 𝑁𝑁−1 𝑃𝑃𝑃𝑃 − 1−𝜌𝜌𝑁𝑁−1 (𝑦𝑦)
Q(T|F) < y  𝑁𝑁 < y  𝐹𝐹 > ≡ 𝑥𝑥
1−𝜌𝜌 𝜌𝜌

Assuming F follows the standard normal distribution,

𝐺𝐺 𝑦𝑦 = 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝐹𝐹 > 𝑥𝑥 = 1 − 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝐹𝐹 < 𝑥𝑥
𝑁𝑁−1 𝑃𝑃𝑃𝑃 − 1−𝜌𝜌𝑁𝑁−1 𝑦𝑦
=1 − 𝑁𝑁 𝑥𝑥 = 1 − 𝑁𝑁 𝜌𝜌
1−𝜌𝜌𝑁𝑁−1 𝑦𝑦 −𝑁𝑁−1 𝑃𝑃𝑃𝑃
G y = 𝑁𝑁 (Note that 1 − 𝑁𝑁 𝑥𝑥 = 𝑁𝑁(−𝑥𝑥))
𝜌𝜌
– Analysis (cont…)

In risk management, we are interested in the worst-case default rate given a

certain confidence level, X (the probability that the default rate is less than
WCDR)

1−𝜌𝜌𝑁𝑁−1 𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊 −𝑁𝑁−1 𝑃𝑃𝑃𝑃

G(WCDR) = X, or 𝑁𝑁 = 𝐗𝐗
𝜌𝜌

𝑵𝑵−𝟏𝟏 (𝑷𝑷𝑷𝑷)+ 𝝆𝝆𝑵𝑵−𝟏𝟏 (𝑿𝑿)

WCDR(T,X)= 𝑵𝑵
𝟏𝟏−𝝆𝝆
• Analysis (cont…)
– How to estimate PD and ρ?
• The probability density function for the default rate is (taking the
derivative of G(y))

2
−1 −1
1 − 𝜌𝜌 1 1 − 𝜌𝜌𝑁𝑁 𝑦𝑦 − 𝑁𝑁 (𝑃𝑃𝑃𝑃)
𝑔𝑔 𝑦𝑦 = 𝑒𝑒𝑒𝑒𝑒𝑒 (𝑁𝑁 −1 𝑦𝑦 )2 −
𝜌𝜌 2 𝜌𝜌

• Maximum likelihood estimate: Use the default rates ranging from

0.088% to 4.996% between 1970 and 2016 for all rated companies;

– Maximizing the sum of the logarithms (log-likelihood) of

∑𝑖𝑖 ln(𝑔𝑔 𝑦𝑦𝑖𝑖 ) for the data we get PD=1.32% and ρ = 0.098
– Probability Distribution for Default Rate

70

60

50

40

30

20

10

0
0 0.01 0.02 0.03 0.04 0.05 0.06
Default rate

• default correlation is important in determining the worst-case

default rate (WCDR), which describes the tail risk
With 𝞀𝞀=0.1, PD=1%  WCDR=7.7% (given 99.9% confidence
level). When 𝞀𝞀=0, WCDR=PD