4.3.2 Measuring Credit Risk

2. Measuring Credit Risk

2.1 Default Probability

Issuer default rate is the number of bonds that have defaulted in a given year divided by the number of issues outstanding.

Dollar default rate is the total par value of bonds that have defaulted in a given year divided by the total par value of all bonds outstanding.

Cumulative default probability: an issuer with a certain rating will default within one year, within two years, within three years and so on.
$$\text{Cumulative survival rate}=1-\text{cumulative PD}$$

Unconditional default probability: the probability of a bond defaulting between time $t_1$ and $t_2$
$$P{D}_K^{\text{Uncond}}=P{D}_{t+k}^{\text{cumulated}}-P{D}_t^{\text{cumulated}}$$

Conditional default probability: if the firm survives to the end of year $n$, what is the probability that it will default during year $n+1$
$$P{D}_{t:k}^{\text{Cond}}=\frac{(De{f}_{t+k}-De{f}_t)}{{\text{Cumulative survival}}_t}$$

What is the cumulative survival rate within four years for a B-rates Bond?

For the same rate, what is the unconditional PD that the bond will default during the fifth year? And the conditional PD during the fifth year assume no earlier default?

Cumulative survival rate: $100\%-15.87\%=84.13\%$

Unconditional PD: $18.32\%-15.87\%=2.45\%$

Conditional PD: $2.45\%/84.13\%=2.91\%$

2.2 Estimate PD

2.2.1 Estimate PD - Default Intensity Model

Poisson Distribution is used to model number of default events over time
$$f(x)=P(X=x)=\frac{(\lambda t{)}^x{e}^{-\lambda t}}{x!}$$

No default within $T$ years
$$P(X=0)=\frac{(\lambda t{)}^0{e}^{-\lambda t}}{0!}=e^{-\lambda t}$$

$\lambda$ is hazard rate, which is the rate at which default are happening. We can use it to calculate unconditional default probabilities.

$$\text{Cumulative survival rate}=e^{-\lambda t};\text{Cumulative PD}=1-e^{-\lambda t}$$

$$P{D}_K^{\text{Uncond}}=P{D}_{t+k}^{\text{cumulated}}-P{D}_t^{\text{cumulated}}=e^{-\lambda t}-e^{-\lambda (t+k)}$$

Suppose that the hazard rate is constant at $1\%$ per year. Please calculate the probability of a default by the end of the third year and unconditional probability of a default occurring during the fourth year.

$\text{Cumulative PD}=1-e^{-\lambda t}=1-e^{-1\%\times 3}=2.9554\%$

$\text{Unconditional PD}=(1-e^{-\lambda (t+k)})-(1-e^{-\lambda t})=(1-e^{-1\%\times 4})-(1-e^{-1\%\times 3})=0.9656\%$

2.2.2 Estimate PD - KMV Model(Merton Model)

We can take credit risk as an option. Consider a firm with total value $V$ that has one bond due in one year with face value $D=100$.

One year later	Total value of firm (V)	Face value of bond (D)	Equity
Scenario 1	$500>D$	$100$	$400$
Scenario 2	$300>D$	$100$	$200$
Scenario 3	$100=D$	$100$	$0$
Scenario 4	$70<D$	$70$	$0$

Equity is a call option on the assets of the firm with a strike price equal to the face value of the debt.

Model includes factors

• The amount of debt in the firm’s capital structure
• The market value of the firm’s equity
• The volatility of the firm’s equity

$$c_t=S_{t}N(d_1)-X{e}^{-rt}N(d_2)$$

$$S_t=V_{t}N(d_1)-D{e}^{-rt}N(d_2)$$

$$PD=1-N(d_2)$$

2.3 Exposure

Exposure is the amount at risk during the life of the financial instrument.

Exposure at default(EAD) is the amount of money lender can lose in the event of a borrower’s default.

2.4 Loss Given Default

Loss given default(LGD) is the amount of creditor loss in the event of a default.

The recovery rate(RR) for a bond defined as the value of the bond shortly after default and it is expressed as a percentage of its face value.

The loss given default provides the same information of loss give default, and it it the percentage recovery rate subtracted from $100\%$.

$$LGD=1-RR$$

$$RR=\frac{\text{Recoveryamount}}{\text{Exposure}}=1-\frac{LGD}{\text{Exposure}}$$

Recovery rates are negatively correlated with default rates.

• Recessionary period: default rates on bonds are high and recovery rates are low.
• Economy is doing well: default rates on bonds are low and recovery rates are high.

2.5 Expected Loss vs. Unexpected Loss

Expected loss(EL) is the amount a bank can expect to lose over a given period of time as a result of credit events.

A bank can manage expected loss by setting lending rates.

$$EL=EAD\times PD\times LGD$$

$$\text{Expected Loss}(\%)=PD\times LGD$$

Expected default rate = 1.5%	Expected loss(%)=0.9%
Recovery rate = 40%
Margin to cover its expenses	1.6%
Average funding cost	1%
Interest rate it charges on its loans	0.9% + 1.6% + 1% = 3.5%

Unexpected loss is the amount a bank cannot anticipates as a result of credit events.

The unexpected loss is high percentile of the loss distribution minus the expected loss.

The bank’s capital is a cushion that covers the unexpected loss.

2.6 Credit Loss Distribution

$EA{D}_i$: The amount borrowed in the $i_{th}$ loan (assumed constant throughout the year).
$P{D}_i$: The probability of default for the $i_{th}$ loan.
$LG{D}_i$: The loss rate in the event of default by the $i_{th}$ loan (assumed known with certainty)
${\rho }_{i,j}$: The correlation between losses on the $i_{th}$ and $j_{th}$ loan.
${\sigma }_i$: The standard deviation of loss from the $i_{th}$ loan.
${\sigma }_p$: The standard deviation of loss from the portfolio.

For an individual loan:
The mean loss (expected loss) is
$$EL=P{D}_i\times EA{D}_i\times LG{D}_i$$

The standard deviation of the credit loss is
$${\sigma }_i^2=E({\text{loss}}^2)-[E(\text{loss}){]}^2=(P{D}_i-P{D}_i^2)(LG{D}_i\times EA{D}_i{)}^2$$

$$\to {\sigma }_i=\sqrt{P{D}_i-P{D}_i^2}(LG{D}_i\times EA{D}_i)$$

For a loan portfolio:
The mean loss(expected loss) is

$$E{L}_p=\sum _{i=1}^{n}E{L}_i=\sum _{i=1}^{n}P{D}_i\times EA{D}_i\times LG{D}_i$$

The variance of the credit loss is

$${\sigma }_p^2=\sum _i\sum _j{\rho }_{i,j}{\sigma }_i{\sigma }_j$$

We assume all $PD$, $EAD$, $LGD$ and $\rho$ are the same and constant for all loans:

$${\sigma }_n^2=n{\sigma }_i^2+n(n-1)\rho {\sigma }_i^2$$

Suppose a bank has a portfolio with $100,000$ loans, and each loan is USD $1$ million and has a $1\%$ probability of default in a year. The recovery rate is $40\%$ and correlation between loans is $0.1$. What is the standard deviate of individual loan credit loss, and the mean and standard deviate of portfolio credit loss?

Standard deviate of individual loan credit loss:
${\sigma }_i=\sqrt{1\%\times 99\%}(1\times 60\%)=0.0597\text{million}$

Mean of portfolio credit loss:
$EL=100,000\times 1\times 1\%\times 60\%=60\text{million}$

Standard deviate of portfolio credit loss:
${\sigma }_p^2=100,00\times 0.0597^2+100,000\times 99,999\times 0.1\times 0.0597^2=3,564,41\to {\sigma }_p=1.888\text{million}$

3. Capital for Bank’s Credit Risk

3.1 Economic Capital v.s. Regulatory Capital

Regulatory capital is the capital bank regulators (also known as bank supervisors) require a bank to keep.

• Separate capital calculations are added to give the total capital requirements.
• Internal ratings-based (IRB, Basel II) - Vasicek Model
• The Basel Committee sets one year $X=99.9\%$ for regulatory capital in the internal ratings-based approach. It occur only once every thousand years.

Economic capital is a bank’s own estimate of the capital it requires.

• Correlations between the risks are often considered.
• CreditMetrics model.
• When banks determine economic capital, they tend to be even more conservative.
• An AA-rated corporation with $PD=0.02\%$ → setting X as high as $99.98\%$
  

Consider a bank rated as AA. One of its key objectives will almost certainly be to maintain its AA credit rating. If an AA-rated corporation has a default probability of about $0.02\%$ in one year. The $99.9$ percentile of the default rate distribution is therefore around $14.89\%$. The $99.98$ percentile of the distribution is around $22.31\%$. The expected default rate for all rated companies is $1.305\%$. The recovery rate is $25\%$. Please calculate the regulatory capital and economic capital.

$EL=(1-25\%)\times 1.305\%=0.98\%$

$\text{Regulatory capital}=0.75\times 14.89\%-0.98\%=10.19\%$

$\text{Economic capital}=0.75\times 22.31\%-0.98\%=15.75\%$

3.2 Regulatory Capital — Vasicek Model

Vasicek model is used by regulators to determine capital for loan portfolios. It uses the Gaussian copula model to define the correlation between defaults.

The Basel Committee sets $X=99.9\%$ for regulatory capital in the internal ratings-based approach.
$$(WCDR-PD)\times EAD\times LGD$$
WCDR (worst case default rate)

Gaussian copula model: a Gaussian copula creates a joint probability distribution between two or more variables which are both normal distributed variables.

One-factor correlation model: Now suppose we have many variables, $V_i(i=1,2,...)$. Each $V_i$ can be mapped to a standard normal distribution $U_i$ in the way we have described.
$$Ui=a_{i}F+\sqrt{1-a_i^2}{Z}_i$$

• $F$ is a factor common to all the $U_i$
• $Z_i$ is the component of $U_i$ that is unrelated to the common factor $F$ (idiosyncratic). The $Z_i$ corresponding to the different $U_i$ are uncorrelated with each other.
• $F\sim N(0,1),Z_i\sim N(0,1)$
• $a_i$ are parameters with values between $-1$ and $+1$.
• $U_i\sim N(0,1)$
• $\rho =\frac{E(U_i{U}_j)-E(U_i)E(U_j)}{SD(U_i)SD(U_j)}\to \rho =a_i{a}_j$
  • $SD(U_i)=0,SD(U_j)=0$
  • $E(U_i)=0,E(U_j)=0$

Vasicek model - Unconditional default distribution
Assume the probability of default( $PD$ ) is the same for all companies in a large portfolio.
The $a_i$ are assumed to be the same for all $i$. Setting $a_i=a$
$$U_i=aF+\sqrt{1-a^2}{Z}_i$$

The binary probability of the default distribution for company $i$ for one year is mapped to a standard normal distribution $U_i$. Company $i$ defaults if:
$$U_i\le N^{-1}(PD)$$

$PD=1\%$, company $i$ default if $U_i\le N^{-1}(0.01)=-2.33$

Vasicek model - Conditional default distribution
$$U_i=aF+\sqrt{1-a^2}{Z}_i$$

The factor $F$ can be thought of as an index of the recent health of the economy.

• If $F$ is high, the economy is doing well and all the $U_i$ will tend to be high (making defaults unlikely).
• If $F$ is low, all the $U_i$ will tend to be low so that defaults are relatively likely.
  $$U_i\sim N(aF,1-a^2)\stackrel{\rho =a^2}{⟶}{U}_i\sim N(\sqrt{\rho }F,1-\rho )$$

Vasicek model
The default rate conditional $n$ the factor $F$:

$99.9\%$ percentile worst case default rate(WCDR)

$$WCDR=N\left(\frac{N^{-1}(PD)-\sqrt{\rho }{N}^{-1}(0.001)}{\sqrt{1-\rho }}\right)$$

$\text{Capital requirement}=(WCDR-PD)\times EAD\times LGD$

A bank has a USD $100$ million portfolio of loans with a PD of $0.75\%$. Assume a correlation parameter of $0.2$. The recovery rate in the event of a default is $30\%$. What is the required regulatory capital using $99.9$ percentile of the default rate given by the Vasicek model?

$N^{-1}(0.001)=-3.0902and{N}^{-1}(0.0075)=-2.432$

$WCDR=N\left(\frac{N^{-1}(PD)-\sqrt{\rho }{N}^{-1}(0.001)}{\sqrt{1-\rho }}\right)=N\left(\frac{-2.4324+\sqrt{0.2}\times 3.0902}{\sqrt{1-0.2}}\right)=N(-1.17)=12.1\%$

$\text{Capital requirement}=(WCDR-PD)\times EAD\times LGD=(12.1\%-0.75\%)\times 100\times 70\%=7.9\text{million}$

3.3 Economic Capital — CreditMetric

CreditMetrics is the model banks often use to determine economic capital. Under this model, each borrower is assigned an external or internal credit rating.

• Step 1: The bank’s portfolio of loans is valued at the beginning of a one-year period.
• Step 2: Use Monte Carlo simulation to model how ratings change during the year.
• Step 3: The portfolio is revalued.
• Step 4: The credit loss is calculated as the value of the portfolio at the beginning of the year minus the value of the portfolio at the end of the year.

CreditMetrics considers the impact of rating changes as well as defaults.

3.4 Risk Allocation — Euler’s Theorem

Euler’s theorem: can be used to divide many of the risk measures used by risk managers into their component parts.

If a risk measure meets homogeneity
$$Q_i=\frac{\Delta F}{\Delta X_i/X_i}(\text{Decomposition})\to F=\sum _{i=1}^n{Q}_i(\text{Combination})$$

• $\Delta X_i$ is a small change in variable $i$, $\Delta X/X_i$ is a proportional change.
• $\Delta F$ is the resultant small change in $F$.
• $Q_i$ is the risk component decomposition.

Suppose that the losses from loans A and B have standard deviations of $ $2$ and $ $6$. The correlation between two loans is $0.5$. The standard deviations of portfolio is $ $7.2111$. Please calculate the dollar contribution of loan A and loan B to the whole portfolio risk.

Loan A:
If size of loan A is increase by $1\%$ ($\Delta X/X_i=1\%$). The SD of loan A is $2\times 1.01=2.02$

$\Delta {\sigma }_p=\sqrt{2.02^2+6^2+2\times 2.02\times 6\times 0.5}-7.2111=0.01388$

Contribution Lona A $=\frac{\Delta F}{\Delta X_i/X_i}=0.01388/1\%=1.388$

Loan B:
If size of loan B is increase by $1\%$ ($\Delta X/X_i=1\%$). The SD of loan B is $6\times 1.01=6.06$

$\Delta {\sigma }_p=\sqrt{2^2+6.06^2+2\times 2\times 6.06\times 0.5}-7.2111=0.05826$

Contribution Lona A $=\frac{\Delta F}{\Delta X_i/X_i}=0.05826/1\%=5.826$

3.5 Challenges to Measuring Credit Risk for Derivatives

Derivatives also give rise to credit risk:

• The value of the contract in the future is uncertain.
• Since the value of the contract can be positive or negative, counterparty risk is typically bilateral.
• Netting agreements: all outstanding derivatives with a counterparty may be considered a single derivative in the event that the counterparty defaults.

3.5 Challenges to Quantifying Credit Risk

PD: Banks are faced with the problem of making both through-the-cycle estimates (to satisfy regulators) and point- in-time estimates (to satisfy their auditors).
LGD: the recovery rate is negatively correlated with the default rate.
EAD: derivative transactions need a relatively complex calculation (may contain wrong-way risk).
Correlations: are difficult to estimate.

Credit risk is only one of many risks facing a bank.