Chapter – 7: Risks of Financial Institutions

Interest rate risk

This type of risk arises when the maturities of the assets and liabilities do not match. It can be of
two types:
1. Reinvestment risk
2. Refinancing risk
Key features:
1. Mere size matching or maturity matching does not result in immunization. We need to match
duration.
2. Immunization is a dynamic phenomenon. Matching duration at a point in time will solve
problems to some extent for that time, but once the situation changes, the whole portfolio
will have to be balanced once again. It is not possible to hedge 100%, but institutions strive
to hedge for a specific moment.
3. Interest rate risk often arises from business model.

Credit risk
Credit risk arises from partial/full repayment of assets (loans, bond etc).
Key features: The experience related to credit risk is not similar across different types of loans and
credit, for example Commercial and Industrial, Consumer, Real Estate and such. Different loan
types have different level of credit risk associated with them.
Commercial and Industrial
 Generally unsecured
 Most of the loans are given under commitments (LCA = Loan Commitment Agreement)
where unused parts usually entail fees.
 Generally less number of default events. However, the amount of loan given is large. For
this reason, one or two defaults may jeopardize the entire financial solvency of the bank.
Therefore, banks generally focus on improving the quality and amount of loan disbursement
among a select few clients rather than increasing the number of clients.
Consumer Loans
Consumer loans generally entail more credit risk. Credit card default risk is much higher compared
to corporate loans. However, the value of the defaults are very low as well.
Real Estate Loans
The risk sometimes drops down to almost zero with the rising prices of collateralized assets.
Generally very long term loans (20-35 years) are disbursed under this heading.
Market risk
It arises due to unanticipated changes in the value of stock, fixed income securities, derivatives.

Banking Book
 Cash  Deposit
 Loan  Bond
 Land  Equity
 Fixed Assets

Trading Book
 Stock (long)  Stock (short)
 Financial Instrument (long)  Financial Instrument (short)
 Foreign Exchange (long)  Foreign Exchange (short)
 Derivatives (long)  Derivatives (short)

Generally each bank has some extra money. They use it for investing in security markets. In
Bangladesh, banks generally used to engaged in such practices themselves, but now they have to
open different subsidiaries. In other words, to engage in securities trading, banks have to open
separate subsidiaries that would exclusively engage in trading. However, at the end of day, the
financial situation of the subsidiaries impact the overall financial situation of the banks as they are
included in the group financial accounts. These subsidiaries maintain trading books. Market Risk
may appear outside of the core banking activities of the bank.

Loan concentration risk

Loan concentration risk is the risk arising out of the situation where Financial institutions have
much of their investments tied to (or concentrated into) a few sectors of the economy.
The nature of the economy is such that despite working to ensure a balanced portfolio, the portfolio
becomes concentrated into few sectors. The experience of Brac Microcredit was that they had to
create markets for the recepients of the credit as well. When they received license for banking, they
went for concentrating on SME banking due to its specialist capabilities.
A large amount of credit flowing to the Industry in Bangladesh flows to RMG as the overall
economy of Bangladesh is dependent on RMG. Loan becomes concentrated into those sectors.
Systematic problem in this sector may trigger a large scale financial crisis for the banks.

Off-balance sheet risk

Risks arising out of contingent liabilities.
 Letter of guarantee
 Loan Commitment
Commercial L/C
Income generated from commercial L/Cs come into the income statement of the bank. If the client
defaults, the bank has to release the products from the port and realize the value of the products
through selling them in the market. However, everyone knows that it is a bank which is making the
deal, who is inexperienced and incompetent to some degree in this market. Therefore, the
bargaining power of the bank deteriorates and the bank ends up making a loss on the sale. The loss
generated from such situation will go to the balance sheet.
Loss=L/C Amount − Recovered Amount

Foreign Exchange Risk

The risk arising out of volatility in the foreign exchange market. This is very relevant at this
moment.
Transaction exposure
Each banks maintain a basic level of inventory of foreign currencies. Instability in the value of the
inventory caused by volatility in the foreign exchange rates may be a source of risk.
Translation exposure
This is generally applicable for the multinational banks. These banks operate in our country and
after the financial year translate their earnings to dollar. This may result in a loss.
Speculation and hedging using FX
Volatility in the foreign exchange market will result in loss or gain in derivatives.

Liquidity risk
This is one of the most fundamental risks of the bank. Depositors may withdraw their money at any
time from current and savings accounts. A situation may arise where the clients are asking to
withdraw their money but the banks are failing to meet their demands. Banks generally build an
understanding of the trends and can reasonably estimate the possible withdrawals and possible
inflow during a period. They also become prepared for festive seasons when there is usually a large
pressure for withdrawing funds. However, if the demands exceed the bank’s expectations, it
becomes risky for the bank.
Insurance companies managing liquidity risk: Some people are buying policies using money. On
the other hand, some are discontinuing their policies and asking for surrender value. Also,
unexpected events result in cash outflows for the company. Insurance companies try to reasonably
estimate cash inflows and outflows within a time frame.
Mutual funds managing liquidity risk: A person asking to withdraw his funds will get as much
money as distribution according to pro-rata basis permits.

Sovereign risk / country risk

Imbalance between the revenue and expenditure of the government results in government
bankruptcy or insolvency.
Historical examples:
1. Russia paying 5 cents on dollar in 1994-95 on its sovereign debt.
2. Greece paying 54% on its sovereign debt.
Sometimes, dramatic ideological changes in government in some countries may lead to repudiation
of foreign loans. In repudiation, the government refuses to pay back the foreign loan incurred by the
previous government which it has overthrown or supplanted.
Also, countries may block foreign remittance to ensure financial stability within their borders. This
happened during 1997 Asian financial crisis.

Technology risk
1. Failure to achieve economies of scale with technology
2. Failure to provide service due to random server crash.

Some points
 Good banks may go bankrupt due to panic in the market
 Country risk may overshadow the diversification benefits achieved by investing in multiple
countries.
 Government may become insolvent as well, leading to defaulting on sovereign debt.
 Chapter – 8: Interest rate risk
Interest rate risk measurement model
1. Repricing gap model
2. maturity gap model
3. duration gap model

Repricing gap model

In Repricing gap model, we identify the RSA (rate sensitive assets) and RSL (Rate sensitive
liabilities) then we calculate the income effect of change in interest rate on these selected RSA and
RSL.
Repricing Gap= Rate Sensitive Assets ( RSA ) − Rate Sensitive Liabiilities ( RSL )

Assets Amount Liabilities Amount

1 Month Treasury bill 100 Equity 100
1 year T bond 100 Passbook Savings 50
30 year FRM (Fixed Rate 300 Demand Deposit 50
Mortgage)
15 year FRM (rate adjusted 50 1 year CD 200
every 9 months)
Cash 50 6 months banker’s acceptance 100
6 month time deposit 100
Total 600 Total 600

Requirement:
 calculate the 6 month repricing gap
 calculate the 1 year repricing gap
 calculate the impact of 1% interest rate change in both the maturity brackets.

Answer
Having a 1 month treasury bill means that if the market interest changes after a month the bank will
not be able to enjoy the interest rate yielded by it any longer. In the 6 months period, this is the only
asset that can be classified as rate sensitive asset in the given scenario. The others have longer
maturity and therefore will be considered as locked in for the period concerned.
Among the rate sensitive assets, only those assets, the yields on which is expected to change in the
given time frame, shall be considered.
Equity is never a rate sensitive liability. Rate sensitive liabilities are only those on which the bank
will have to pay interest. 1 year certificate of deposit is not rate sensitive for a 6 months period. 6
months banker’s acceptance is rate sensitive for a six months time period. 6 months time deposit is
also so.
RSA = 1 month treasury bill = 100
RSL = 6 months banker’s acceptance + 6 months time deposit = 100 + 100 = 200
Gap = RSA – RSL = 100 – 200 = -100
Why demand deposit should be kept as a part of RSL
Argument against:
1934 marked the year when US started to regulate the banking industry. Regulation Pew gave
instructions about how much interest should be charged on different securities and deposits. It
specified that demand deposits should not carry interest and prohibited the banks from paying
interest on that account. They said that demand deposit is only there to facilitate transactions. If a
demand deposit does not carry interest, there is no question to consider it as a rate sensitive
liabilities.
In 2010 through Wall Street Reform, Regulation Pew was abolished. Now banks are not expressly
prohibited, banks may pay interest on demand deposits if they wish to. Now it has become the
decision of the banks to decide whether or not to pay interest.

Argument for:
A bank may not have to pay any interest on demand deposit, but it constitutes nearly 30-40% of the
total deposit financing of the bank. Many people consider demand deposit to be part of core deposit.
The level of demand deposit is somewhat predictable, enabling many banks to actually make use of
them to provide loans. To prevent people from withdrawing demand deposit from the bank, the
bank has to come up with new schemes that ensure that the money remains in the bank. In other
words, the bank has to ultimately replace demand deposit with schemes that carry higher interest.

Passbook savings
Arguments for not adding:
In the rural areas of the United States, during the cowboy era, the human settlement was scattered
and rare. In that situation, to ensure capital creation Post Office provided the people with passbooks
that provided clear information about how much they have saved and stuff. Till 1986, the interest
rate on passbook savings remained inflexible due to regulation (how much banks can pay as interest
rate was capped). Even after reduction of regulation, the interest rate remained inflexible. For this
reason, many people advocate that passbooks should not be considered as a RSL.

Arguments for adding:

When interest rate increases in the market, people carrying passbook savings withdraw money to
invest in other lucrative assets. Therefore, there is a indirect effect. Therefore, it should be
considered as a RSL.
We are not going to consider Passbook savings as a RSL.

1 Year Gap
Gap = RSA – RSL = (100 + 100 + 50) – (100) = 150

interest rate increases by 1%

Interest rate Interest income Interest Expense Gap Net effect

movement
Rise Rise Rise Negative (Amount Negative (as my
of Liability high, expense is rising
amount of asset is more compared to
less) my income as the
liability is larger
compared to
assets)
Rise Rise Rise Positive (Amount Positive
of liabilities low,
amount of assets
high)
Fall Fall Fall Negative Positive (Expense
will fall by a
larger amount
compared to
income)
Fall Fall Fall Positive Negative

Effect of 1% increase in interest rate within the six month time frame = 1% * Six month gap = 0.01
* (-100) = -1
Weaknesses of repricing gap
Market Value Effects: In the valuation formula of a bond, we do consider the discount rate that is
prevailing in the market. In the repricing model, no provision is made for the change in the value of
bonds or other assets through change in valuation factors. It is at most a incomplete model to assess
the interest rate exposure. It only tries to capture the income effect of the movement of market
interest rate, and ignores the balance sheet implications resulting from interest rate change.
Over-aggregation: Central banks around the world asks for report on Interest Rate Gaps from the
commercial banks. They ask for gap analysis of a specific period. It is very possible to have positive
or negative gaps in different period duration. 1 month gap may be negative, 2 month gap may be
positive and again 3 month gap may be negative. Therefore, only finding the gap for the reporting
purposes might not paint the right picture. This is a manageable problem. Find smaller gaps on top
of central bank requirement to find the right, or a more accurate, picture.
Runoff Problems: 30 year FRM is not sensitive in context of 1 or 2 year gap calculation. However,
the bank does earn interest on them yearly, and those cash flows have to be reinvested. These cash
flows may be rate sensitive. These large assets may carry reinvestment or refinancing risks.
Off Balance Sheet Implication: From a balance sheet point of view, it might seem that the bank
has eliminated all possible interest rate risk. However, the bank can always make or lose money
from off-balance sheet items. One example is interest rate futures. This is also influenced by interest
rate movement. This gap analysis do not consider these elements in assessing the risks.

Maturity model of interest rate risk evaluation

Issue:
 Price Yield relationship.
 Weighted average maturity gap
 Interpretation of maturity gap on interest rate risk
 weaknesses of maturity model.
Price yield relationship
Math
Maturity = 1 year
Face value = $100
Coupon rate = 10%
current market interest rate = 10%

110
Price = = 100
1.10
110
Price (if the interest rate in the market becomes 11%) = = 99.09
1.11

Case 2
Everything is same, except the maturity of the security is 2 years.
10 110
Price @ 10% rate of interest = + = 100 (the price remains same as the yield is equal to
1.10 1.102
the market interest rate perhaps. If the interest rate changes, the impact will be more severe
compared to that of 1 year maturity bond.
Case 3
If the maturity increases to 3 years, the price becomes more sensitive to the changes to the interest
rate. A small change in interest rate leaves a severe impact on the price.
Observation
  the price yield relationship is negative.
 The negative price yield relationship is more pronounced with longer maturity security.
 However, the value will drop at a decreasing rate as the maturity goes up. The value drop
increases at a decreasing rate. The fall in price with respect to maturity will increase at a
decreasing rate.
How to calculate maturity gap

Asset Liabilities
Loan O 3 years $100 Deposit 1 5 years $100
Loan A 5 years $100 Deposit 2 1 year $100
Corporate 10 years $200
bond
$400 $200

Weighted average maturity of assets= ( 100

400 ) ∗ 3+(
100
400 ) ∗ 5+(
400 )
200
∗ 10=7 years

A similar calculation of liabilities will reveal the weighted average maturity of liabilities to be 3
years.
Maturity gap = weighted average asset maturity – weighted average liabilities maturity
=7–3
= 4 years
Understanding the impact of maturity gap on equity
In the following situation, the market interest rate is 10%.

Asset Liabilities
Asset @10% 3 years $100 Liability 1 year $90
@10%
Equity $10
Total $100 $100

Now let’s assume the average market interest rate rises to 11%.

New value of the asset = ( 1.11

10
)+( 1.11
10
)+( 1.11
2
110
)=97.55
3

90 +9
New value of the Liability = =89.19
( 1.11 ) 1
Assets = Liabilities + Equity
Equity = Assets - Liabilities
New situation will become like the following

Asset Liabilities
Asset @10% 3 years 97.55 Liability 1 year 89.19
@10%
Equity 8.37
Total 97.55 97.55

8.37 −10
Fall in equity price due to 1% rise in interest rate = =16 % ( almost )
10
Savings institutions and stuff
these institutions exclusively focus on the real estate sector, where the maturity gap is very high.
They are very specialized in this specific industry. For them, Asset maturity of asset would be very
high, around 30 years. Whereas, the liabilities carry comparable maturity to banking institutions (for
example 1 year). In this situation, they start with a very precarious situation, even slight change in
interest rate will force them into bankruptcy. Their maturity gap in the example is nearly 29 years,
which exposes them to significant risk. During 1980s, US experienced rapid increase in inflation
and interest rate, interest rate rose by around 3-4%. For that reason, around 2/3 of the savings
institutions went bankrupt. Their equity was wiped out. This teaches us that institutions should
carefully keep the maturity gap low, so that temporary shocks can be absorbed well.

Problems of maturity model

Having maturity gap at 0 will help to preserve equity, but won’t fully immunize institution from
risks.
Case 1: Asset and Liability maturity are same
market interest rate at 10%

Asset Liabilities
Asset @10% (1 year) 100 Liability @10% (1 90
year)
Equity 10
100 100

Let’s assume market interest rate rises to 11%, then the situation becomes like this

Asset Liability
Asset @10% (1 year) 98.11 Liability @10% (1 89.9
year)
Equity 9.89
98.11 98.11
Despite having maturity gap at 0, my equity value dropped. This is because in reality there is no
fully levered company in real life. Even for banks, it has to comply with numerous regulations that
encourages it to increase the level of equity capital. As equity has been used, the value of equity
dropped in the context.

Case 2: Both maturity gap and amount gap is zero

Asset Liability
Loan @15% (1 year) 100 Liability @15% (1 year 100
CD) (After 1 year the
payment structure: bank will have to pay
Within six months, the 115)
loan recipient will pay
back half the principal
with interest. After that
at the end of the year,
he will pay back the
remaining half along
with interest for the
remaining period.

Repayment:
After 6 months
Principal = 50
interest payment = 100 * 15% * (½) = 7.5

After 1 year
Principal = 50
Interest payment = 50 * 15% (½) = 3.25

I will have to reinvest the cashflows received. The reinvestment rate in this scenario may be
different compared to the interest rate originally discussed. If the reinvestment rate was higher, I
may have extra profits, however, if the reinvestment rate was lower, I will incur a loss.

I will incur a loss in this scenario at a lower reinvestment rate, because the payment structure of the
assets and liabilities is different. Therefore, matching maturity along with dollar amount will not
fully immunize me, I will be exposed to possibilities of loss due to payment or cash flows structure.

Takeaways
 maturity gap model is an upgrade over repricing gap model, since it accounts for the change
in market value due to interest rate change.
 It intuitively suggests that asset and liability portfolio maturity should be matched, but as all
of assets cannot be funded by liabilities, it cannot guarantee full immunization.
 At the extreme even after matching maturity and dollar value, cash flow patterns may still
thwart the planning of the institution and yield a loss on the project.

Duration model of interest rate risk evaluation

Duration is the weighted average maturity of any fixed income security, where weight represents
the present value of cashflow.
Duration measures the interest rate sensitivity of a fixed income security. The higher the duration,
the higher will be the interest rate sensitivity.
Duration also measures the recovery time of investment.
Example situation:
1 year corporate loan (15% interest), face value $100; payment structure at the end of 6th month
50% principal and interest...
Asset duration calculation

Time Cash flow PV of cash flow

½ years 57.5 53.49
1 years 53.75 46.51
100

Asset duration = 0.5 * (53.49/100) + 1 * (46.51/100) = .73 years

Liability duration calculation

Time Cash flow PV of cash flow

1 year 115 115
= 100
( 1.15 )1

Liability duration = 1 year

Duration gap = Asset duration – liability duration = .73 – 1 = -0.27 years
Example math 1
Question: Eurobonds pay coupons annually. Suppose a Eurobond matures in 6 years, the annual
coupon is 8%, the face value of the bond is $1000, and the current yield to maturity is also 8
percent. What is the duration?
Solution:

Time (t) Cash flow (CF) Present value discount Present value of PVCF × t
factor (PVDF) = cash flow, PVCF

1 CF × PVDF
( 1.08 )n
1 80 0.93 74.07 74.07
2 80 0.86 68.59 137.18
3 80 0.79 63.51 190.53
4 80 0.74 58.80 235.20
5 80 0.68 54.45 272.25
6 1080 0.63 680.58 4,083.48
Sum 1000 4,992.71

Duration=
∑ ( PVCF ×t ) = 4992.71 =4.992 years
∑ ( PVCF ) 1000

In case of semiannual compounding, just divide the interest rate by 2 and use it for semiannual
compounding. In this case it would be 0.08/2 = 0.04. On the other hand, each half year period shall
be counted as a 1 period (explanation: n = 1, then 2, then 3).

Question: Suppose a zero coupon bond matures in 2 years, the face value of the bond is $1000, and
the current yield to maturity is also 8 percent. What is the duration?

Time (t) Cash Flow PVDF PVCF PVCF*t

1 0 .93 0 0
2 1000 .86 860 1720
Sum 860 1720

For any zero coupon bond, duration of the bond equals the maturity of the bond (Maturity =
Duration)
1720
Duration = =2 years
860
Interest rate sensitivity
6 year Eurobond with an 8% coupon and 8% yield. We determined in that its duration was
approximately 4.993 years.
 What is the % change in bond’s price if interest rate increases by 2%.
 what is the dollar change in price if interest rate increases by 2%

Solution:
Percent Price change due to interest rate change =
Δ Price 1
=%changeinprice=( − Duration ) × ΔR
Price ( 1+ R )
Δ Price ΔR
→ =− Duration×
Price 1+ R

when the rate increases, ΔR shall be positive. If the rate decreases, ΔR shall be negative.
If the dollar amount of change is asked to be calculated, just use the given formula, only this time
take the ‘Price’ from LHS to the RHS.
1
Change in dollar value of price = Δ Price=( − Duration ) × Δ R× Price
( 1+ R )

Features of Duration
 Duration increases if the maturity increases. In this situation, duration increases at a
decreasing rate.
 If the market interest rate increases, duration decreases.
 When coupon increases, duration decreases. This is because when a bond pays a higher
coupon rate, the recovery rate of the initial outlay increases. Therefore, the investor recoups
the initial investment faster.

Calculating duration at Portfolio level

Market Value Duration of Market value Duration of

(Dollars) individual (Dollars) individual
assets (years) liabilities
(years)
Asset 1 100 m 4 Liability 1 50 m 1
Asset 2 200 m 6 Liability 2 50 m 1
Asset 3 300 m 8 Liability 3 400 m 3
Sum 600 500

(
Asset Portfolio Duration = 4 ×
100
600 )( )( )
+ 6×
200
600
+ 8×
300
600
=6.67 years

Liability Portfolio Duration = ( 1 × ) +( 1 × ) + (3 ×

500 )
50 50 400
=2.6 years
500 500

Portfolio Duration Gap = Asset Duration – Liability Duration = 6.67 – 2.60 = 4.07 years
Duration and interest rate risk management (with respect to a single security and a single
transaction)
Context: An Insurance company has to pay an obligation after 5 years. The value of the obligation is
$1469. Insurance company has two options
1. Lock in a zero coupon bond (5 years); YTM = 8%, that exactly matches the obligation
value.
2. 6 year maturity 8% coupon bearing bond, YTM = 8%; Face value = 1000
Option 1:
The duration of the obligation will be the same as its maturity as it doesn’t have any payment within
the period. Therefore, the duration of the obligation = 5 years.
For the zero coupon bond, maturity = duration. Therefore, the duration of the zero coupon bond = 5.
If nothing is mentioned, the face value of the zero coupon bond is 1000.
1000
Market Price of the Zero coupon bond= =680
( 1.08 ) 5
Available fund for me to invest = $1000
1000
Number of bonds that can be purchased= =1.47
680
Total payoff 1.47 bonds after 5 years=1000 × 1.47=1470
The future value exactly matches the obligation in question. I am not gaining, nor losing.
Option 2:
First scenario, YTM remains 8%:
There are three possible cash flows
1. coupons (this is certain)
2. reinvestment income (this is uncertain)
3. proceeds from sale (I plan to sell the bond after 5 years instead of holding it until maturity)
Coupon income = 1000 × 0.08=80

Future value of coupon income= ( 0.08

80
) × [ ( 1.08 ) − 1]=469.33
5

Reinvestment income=Total Income −Total coupon payment=469.33 − ( 80×5 ) =69.33

1000+80
Selling price of the bond = =1000
1.08
(only one year remains for the buyer, therefore he will expect one year’s coupon payment along
with principal on maturity)
Based on similar calculations as shown above, the following table can be prepared:

Particulars YTM = 8% YTM = 9% YTM = 7%

Selling price 1000 990.80 1009.35
Coupon Payment 400 400 400
Reinvestment income 69.33 78.78 60.06
Total payoff 1469.33 1469.6 1469.41

(matches exactly) (the increase in (the decrease in

reinvestment income is reinvestment income
offset by the reduction offset by the increase in
in selling price of the selling price of the
 bond after 5 years) bond after 5 years)

This proves that if a company matches the duration of a single transaction, it contributes
theoretically to immunization.

Duration and interest rate risk management for the overall balance sheet
A=L+ E
→ Δ A= Δ L+ Δ E
→ Δ A − Δ L= Δ E
Now, we know:
ΔP 1
=− D × ×ΔR
P 1+ R
1
→ ΔP=− D × × Δ R× P
1+ R
As price can imply both assets and liabilities, this implies that:
ΔR
Δ A=− D A × ×A
1+ R
ΔR
Δ L=− D L × ×L
1+ R
Here, we’ve replaced P with A and L in respective cases.

So, now,
Δ A − Δ L= Δ E

[
→ − D A×
ΔR
1+ R ][
× A − − DL ×
ΔR
1+R ]
× L =Δ E

[
→ − D A×
ΔR
1+ R ][
× A + DL ×
ΔR
1+ R ]× L =Δ E
ΔR
→
1+ R
[ ( − D A × A ) + ( D L × L ) ]= Δ E
Δ R × A [ (− D A × A )+( D L × L )]
→ × =Δ E
1+ R A

→
1+ R [
Δ R × A − DA × A DL× L
×
A
+ ]A
=Δ E

→
ΔR× A
1+ R [ ]
× − DA+ DL × = Δ E
L
A

→
ΔR× A
1+ R (
× [ − D A + DL × K ] = Δ E Assuming = K
L
A )
Therefore ,
ΔR×A
Δ E= × [− DA+ DL× K ]
1+R
ΔR
Wrapping up , Δ E= × A × [ − DA+ DL× K ]
1+ R
From this, we can understand that ΔR depends on 3 issues. Among them, Leverage adjusted
duration or [ − D A + D L × K ]is the most practical policy choice to reduce exposure to interest rate
risk. Asset size or ‘A’ is a suboptimal policy target because no bank wants to reduce its asset size, it
is simply not practical to expect banks to make their corporate strategies in a way that limits their
level of assets.
The equation derived above for finding Δ Ecan be used to find the change in equity value if the
interest rate changes.
Now the question is, how to use the leverage adjusted duration gap or [ − D A + D L × K ]for some level
of immunization from interest rate risk. The ways are given below

Element Action that will reduce Practical consideration

interest rate risk
DA Decrease Good policy
DL Increase Good Policy
K Increase Not a practical one as it will
involve raising the level of
Liabilities (L) compared to
level of Assets (A). Regulators
are increasingly asking for
more equity in Bank capital.
This makes it hard to
implement.

Example:
Let’s consider an example of a balance sheet:
When interest rate = 10%,

Duration Value Duration Value

Asset 5 100 Liability 3 90
Equity 10
100 100
Now if the interest rate changes to 11%
ΔR
Δ A=− Duration × ×A
1+ R
0.01
→ Δ A=− 5 × ×100
1+0.10
→ Δ A=− 4.55
Therefore ,new level of assets=100 − 4.55=95.45
Again,
ΔR
Δ L=− Duration × ×L
1+ R
0.01
→ Δ L=−3 × × 90
1+0.10
→ Δ L=−2.45
Therefore , New level of liability=90− 2.45=87.55
Now the balance sheet will stand as,

Assets 95.45 Liabilities 87.55

Equity (95.45 – 87.55) 7.90
95.45 95.45

7.90 − 10 −2.10
percent change of value of equity= = =21%
10 10
The 21% decline of the value of equity was a result of only 1% increase in interest rate.
Limitations of Duration model
 Duration matching requires balance sheet restructuring. Balance sheet restructuring used to
quite costly 15 years ago. However, many new methods of balance sheet restructuring has
emerged in the recent years. Balance sheet restructuring can be either indirect or direct.

Direct method Securitization of loans carrying large duration and turning them
into smaller duration loans. The same can be done for assets as
well. In United States this market is becoming more and more
vibrant. This is not very costly.
Indirect method Futures, forward etc derivatives can be used restructure the
balance sheet without much of a cost and much of a hassle. In
this method, D Aand D Lcan be matched without directly messing
with the balance sheet.

 Duration matching is a dynamic problem.

Example:
Pension fund obligation = 1469
Asset: 6 years, 8% coupon rate, 1000 face value.
From previous calculation the Duration came as 4.993 years.
Now, let’s assume one year has passed.
Liability duration = 5 – 1 = 4 years. (zero coupon bond)
The same simple method cannot be done for the asset as it has some contractual parts and some non
contractual parts. We shall have to recalculate the duration for the asset.

Time Cash Flow PVCF PVCF * t

1 80 74.77 74.77
2 80 69.87 139.24
3 80 65.30 195.9
4 80 61.03 244.12
5 1080 770.04 3850.2
Sum 1041.04 4504.77

1041.04
Therefore, Duration= =4.33 years
4504.77
My duration has not matched due to non-contractual elements, exposing me to risk. Therefore,
duration is not a static problem that I can solve once and forget about it, rather I would have to
match it again and again. The cost of matching it once is not very high, however the cumulative cost
of matching it again and again may become prohibitive.

Duration one year ago 4.993 years

Duration now 4.33 years

One way to immunize against it will be buying a zero coupon bond that reduces the average
duration of the asset to 4 years (thereby matching it with the liability duration).
4.33 ×0.5+3.67 ×0.5=4 years
In the given above given example, the bank is buying a zero coupon bond carrying a duration of
3.67 years at such a quantity that it becomes 50% (0.5) of the total assets. However, there is no one
way of doing it. There are infinite number of ways this equation could have been constructed, for
different amount of weights, for different level of duration. The goal is to bring the average duration
to 4 years from the current 4.33 years.
 The relationship between yield and price is not linear, rather it is convex.
FV = 1000
Maturity = 4 years
Coupon rate (annual)= 8%
Market Yield = 10%
Note : 100 bp=0.01=1 %
Time Cash Flow PVDF PVCF PVCF*t
1 80 0.9090 72.7272 72.7272
2 80 0.8264 66.1157 132.2314
3 80 0.7513 60.1051 180.3155
4 1080 0.6830 737.6545 2950.6181
936.6026 3335.8923

3335.8923
Duration= =3.5617
936.6026
Now, let’s assume that interest rate changes by 20 bps, or 0.2%.
Price change predicted by Duration model
ΔP ΔR
=− Duration×
P 1+ R
ΔR
Δ P=− Duration× ×P
1+ R
0.002
Δ P=−3.5617 × × 936.6026
1+0.10
Δ P=− 6.0652
Actual price change found through discounted cash flow model
Actual Price change=New Price − Old price

Actual price change=

[
80
0.102
× 1−
1
(
1.102
4 +
1000
1.102) ]
4 − 936.602

Actual Price change=−6.03

Absolute deviation should be shown for error approximation, direction is not needed.
Again, now consider market interest rate rises by 5% from previous 10%.
Price change predicted by the duration model
ΔP ΔR
=− Duration×
P 1+ R
ΔR
Δ P=− Duration× ×P
1+ R
0.05
Δ P=−3.5617 × × 936.6026
1+0.10
Δ P=−151.6297
Actual price change found through discounted cash flow model
Actual Price change=New Price − Old price

Actual price change=

[ (
80
0.15
× 1−
1
1.15 4 + )
1000
1.15 4 ]
−936.602

Actual Price change=−136.44

Again, assume the market interest rate decreases by 5%.
Price change predicted by Duration model
ΔP ΔR
=− Duration×
P 1+ R
ΔR
Δ P=− Duration× ×P
1+ R
− 0.05
Δ P=−3.5617 × × 936.6026
1+0.10
Δ P=151.6297
The actual price change measured using discounted cashflow model
Actual Price change=New Price − Old price

Actual price change=

[ (
80
0.05
× 1−
1
1.05 4 + )
1000
1.05 4 ]
−936.602

Actual Price change=169.79

The error can be summarized in the following table:

Situation of the interest Duration gap predicted Actual Change Error (absolute value)
rate change
20 bp or 0.2% increase - 6.0652 - 6.03 0.0392
5% increase - 151.6297 - 136.44 15.19
5% decrease 151.6297 169.79 18.1603

For this convex relationship, the price change predicted by duration model will not be the same as
the price change that will actually take place based on the bond valuation method (
Price change=New price − Old price ). The difference between the two methods of finding the price
change will be considered as an error.
First find the new value of the bond based on the discounting cash flow valuation model. Then find
the difference between the new price and the old price. You will find the difference. On the other
hand, using the duration model, find the price change that were expected by the model. You will
find some difference between the two. When finding error we must find the absolute value, the
positive or negative issue doesn’t matter.
We can make some observations from this:
First, Duration model is good for processing small movements of interest rate. However, if the
movement is larger, duration model experiences larger error rate. Furthermore, two equal movement
of interest rate at opposite direction should bring the same level of change in price predicted by the
duration model, this is not the reality when we use the discounted cash flow model.
Duration assumes the price yield relationship to be linear. However, the relationship is actually a
curvature (convex). That’s why small changes do not produce much of a large error, but large errors
do produce a very large error.
Second, the convex relationship is also responsible for equal but opposite interest rate changes not
producing symmetric effect. Value loss with interest rate increase is lower than the approximate
value increase for the equal level of interest rate fall.
 Chapter 10: Credit Risk
Types of Loans

Dimension Wholesale Real Estate Consumer Loans

(commercial and
industrial)
Maturity Short term and long Long term Short term
term
Collateral Short term = Secured Unsecured
Unsecured
Long term = secured.
Loan Commitment Yes No No
agreement

Institutions
1. Commercial banks
2. Savings Institutions
3. Trade Union
Commercial Loans
Revolving: Can withdraw the same loan multiple times if the person pays back within the stipulated
period.
Non-revolving: Particular asset is used to give loans.
(Credit cards usually carry shorter terms. Auto-Loans are small, therefore the maturity is low as
well. One interesting thing to notice here is that loan for buying the same car in Japan and
Bangladesh may carry different maturities. In Japan, the price of the car is low, therefore the amount
lent is low as well. This contributes to lower maturity of the loans. On the other hand, when the car
comes to Bangladesh, it has to pay a very high customs duty and tax, thereby pushing up the price
many times. Therefore, to accommodate higher prices, the loan has to carry higher maturity as
well.)
Wholesale Loans (Commercial and industrial)
Short Term: Given to fulfill the needs of working capital management.
Long Term: Given for building up larger investments.
Real Estate
This type of loan is by default long term in average (28.4 years).

Collateral
Real Estate
This is the best loan in a bullish market as the asset price is continuously increasing. Even if the
prices do not increase, the home remains as a collateral. Therefore, it might be regarded as highly
secured.
Consumer Loans
It generally doesn’t carry collateral. The interest rate is much higher as the loan is unsecured.
Commercial and Industrial
Short term one is unsecured but the long term one is secured.
Special Lien
The special lien is when a specific asset is tagged for saying that if the company fails it will sell the
specific asset to pay installments. Otherwise it will be general lien.
It is easier to create special lien in industries where parts of the machinery can be sold separately.
Therefore, it becomes easier for receiving collateralized loans for that industry.
Loan Commitment
The parties can’t say specifically but can make a reasonable estimate about use. The bank keeps that
estimated amount in bank, for withdrawing and using as much as the party wants. The used part of
the loan will entail interest charges. The remaining amount carries a small penalty to be paid to the
bank.

Bangladeshi perspective
Wholesale Corporate loans
Real Estate + Consumer Loans Retail Loans

The banks has to give special treatment to SME and Women Entrepreneurs due to regulations.

Potential and expected return of the loan (from Banker’s

perspective)
Explanation of terms:

Terms Discussion
Base Rate The bank has to collect money before giving loans. The Base Rate is the
weighted average cost of funding (WACC) of the bank.
Credit Risk Premium This can change from borrower to borrower. It can be negative, zero or
positive.
Loan Origination Fee Before disbursement there are some costs of documentation.
After disbursement the following costs arise:
a) check history
b) complementary card
c) transaction
Compensating balance A part of the loan will not be disbursed by the bank. Bank will keep that
money in a demand deposit where the interest rate is very low (1%-2%).
 The bank can use this demand deposit to give more loans. However, that
will be subject to reserve ratio as well.
The effect of compensating balance is the reduction of Loan to Value ratio.
If the customer defaults, the bank will still have this amount with it.

Math:

Particulars Symbol Value

Prime Lending Rate/ Base rate BR 6%
Credit Risk Premium Φ 2%
Loan Origination Fee OF 1%
Compensating Balance b 8%
Reserve Ratio RR 10%

Requirements
a) Calculate the promised return (K) of the loan
b) Provided that there is 2% chance of default, what is the expected return of the loan?

Loan Amount 100

Disbursement[ Loan amount × ( 1 − b ) ] 92 [ note : 100 × ( 1 − 0.08 ) =92 ]
(the bank is retaining compensating balance
while disbursing loan)
Demand Deposit (Coming from compensating 8
Balance)
Loan from the above demand deposit 7.2 [ note : 8 × ( 1− 0.10 ) =7.2 ]
[ Demand Deposit amount × ( 1 − RR ) ]
(the bank is maintaining the reserve ratio while
giving out loans from the demand deposit)
Remaining in reserve 0.8 [ note : 8 −7.2=0.8 ]
[ Demand deposit − Loans disbursed ]

Answer to the Question no (a):

 OF + ( BR+Φ )
1+ K =1+
1− [ b ( 1 − RR ) ]
0.01+ ( 0.06+0.02 )
→ 1+ K=1+
1 − [ 0.08 ( 1 − 0.10 ) ]
0.01+0.08
→ 1+ K=1+
1 − [ 0.08 × 0.90 ]
0.09
→ 1+ K=1+
1 − 0.072
→+ K =1.096
→=1.096 −1
→ K =0.096
→ K =9.60 %
Effect of different variables on K

BR (up) K (up)
Φ (up) (acts as a pricing tool of the bank) K (up)
OF (up)(acts as a pricing tool of the bank) K (up)
b (up) (acts as a pricing tool of the bank) K (up)
RR (up) K (down)

Bank uses the three variables (pointed out in the table) for pricing the loans.

Answer to the question no (b):

Here, Probability of payment, P = 1- Probability of default = 1 – 0.02 = 0.98
1+ E ( R ) = [ ( 1+ K ) × P ] + [ nonpayment value × ( 1 − P ) ]
→ 1+ E ( R ) = [ ( 1+0.096 ) ×0.98 ] + [ 0 × ( 1 − 0.98 ) ]
→ E ( R ) =7.41 %
Here, the nonpayment value has been assumed to be 0. However, if the loan is collateralized, the
nonpayment value will be greater than 0 as some amount will be recovered.

Models of credit risk

a) K means clustering: Within the group variations less, across the group variations high.
Group members are homogeneous. Groups are heterogeneous.
b) KNN Algorithm: KNN means K nearest neighbor.
c) Elbow Curve
d) Silhouette Coefficient

Core dataset as examples

Borrower Expense to Income Ratio Number of Credit Cards

A 75 0
B 70 1
C 80 0
D 85 2
E 60 3
F 65 0
G 90 1
H 80 1

K means clustering
1st iteration:

Borrower Expense to Number of Distance Distance Distance Initial

Income Credit C1 C2 C3 Clustering
Ratio Cards
A 75 0 0 5.09 5 1 (The
Distance cell
containing
the smallest
among the
three. Here,
C 1carries the
smallest
value,
therefore the
Initial
clustering is
1.)
B 70 1 5.09 0 10.05 2
C 80 0 5 10.05 0 3
D 85 2 10.198 15.03 5.39 3
E 60 3 15.29 10.198 20.29 2
F 65 0 10 5.09 15 2
G 90 1 15.03 20 10.05 3
H 80 1 5.09 10 1 3

Here,
Initial Centroid: (Here, Centroid means central tendency. We have to have a general idea about the
central tendencies of the groups). We would have to assume it.

C 1(Assuming Equal to A) (75,0)

C 2(Assuming Equal to B) (70,1)
C 3(Assuming Equal to C) (80,0)

“K” means the number of clusters that we are going to have. K has to be greater than 1, as having
one group doesn’t make sense. K can’t be an even number as well, because then it might create
problems with prediction. Therefore, K should be an odd number, greater than 1.
In this situation, we are assuming that K = 3.

Distance is calculated using Euclidean Distance Method.

Distance= √ ( A first data− C first data ) + ( A second Data − C second data )

2 2

Example ,
Distance of A with consideringC 1= √ ( 75 −75 ) +( 0 −0 ) = √ 0=0
2 2

0 has been added for the C 1 column for ' A ' row .

When finding Distance with regard to C 1, use C 1data for C data.

In this situation, it is possible to do it using hand as the number of dataset is less. If the number of
dataset increases, it will become increasingly impossible to do by hand. In real life, managers use
coding like Python, R etc.
New Centroid
C 1=( 75,0 )(There is only one set in Cluster 1)

C 2=( 65 ,1.33 )(Take the sets belonging to Cluster 2 and take their average)

C 3= ( 8.375 ,1 ) (Take the sets belonging to Cluster 3 and take their average)

We will have to repeat the 1st iteration again (it will be called 2nd iteration) using the new centroid.
This will continue until everything matches exactly. When it does, it will be called ultimate
clustering.
In Bangladesh, in the credit risk manual, it has been said to divide the customer base according to 8
credit risk clusters. Through clustering, we can understand the characteristics of the significant
groups, like for an ‘Excellent’ group, what is there yearly income, how many credit cards they hold
etc. When the dataset increases, so does the accuracy and insight provided by the model.
This method is called unsupervised learning, as no prediction is being made. We are not trying to
allocate one dataset to any cluster. Instead, we are just trying to ensure an effective clustering
process.

Note
If I have 10,000 data, the highest number of clusters I can have is 10,000 clusters. The lowest
number of clusters would be 1. In case of 10,000 clusters, the sum of squares would be zero, as the
data would be equivalent to mean (each cluster would contain only one data, that would be the
mean and that would be data). In case of 1 cluster, the sum of squares would be the highest. We
don’t want either, we want the number of clusters to be somewhere between 1 and 10,000 that will
ensure best possible use.
Note
“Kaggle” is a website that provides free access to demographic and other data with regard to credit
and stuff. Data can be downloaded from there.

KNN Algorithm
Here, Akash is applying for a loan. He has 1 credit cards, his expense to income ratio is 80.
This data is going to act as the centroid in this situation.

Borrower Expense to Number Label Distance Ranking

Income of Credit (using the same (Rank the lowest
Ratio Cards formula as used in K distance as 1, after
means Clustering that go on like
model) that)
A 75 0 Non Default 5.09 3
B 70 1 Non Default 10.07 5
C 80 0 Non Default 1 2
D 85 2 Default 5.09 4
E 60 3 Default 20.09 8
F 65 0 Non Default 15.03 7
G 90 1 Non Default 10 6
H 80 1 Non Default 0 1

Here, we assume K = 3. This means that we are going to evaluate the closest 3 cases (by ranking,
the lower the ranking the closer to Akash’s case) to Akash’s case.
Here, Rank 1 is Non Default, Rank 2 is Non Default, Rank 3 is Non Default. Based on that we can
give the verdict that Akash is not likely to default.

This method is called supervised learning, as prediction is being made.

Silhouette Coefficient
The value of the coefficient will be within the range of -1 and +1. The more the value approaches
negative, the worse the clustering is. The more the value approaches positive, the better the
clustering is.

LPM model (Linear Probability Model)

It is basically a regression model, where the dependent variable is a binary one. The dependent
variable can also be thought of as a dummy variable.

Dependent variable values

Default 1
Non Default 0
The values can be flipped as well.
One important prerequisite for using the model is having a significant amount of data regarding the
variables used in the model. Another prerequisite for using the model is that there should be a level
of default risk within the model. I take a huge dataset, but none of the people have default risk, this
is not acceptable.
Math 1:
We have developed an LPM model

Probability of default 0.3 × Levarage level − 0.4 × sales growth rate

Leverage of the firm under evaluation 1.2
Sales Growth rate of the firm under evaluation 0.02

Probability of Default (based on the equation given before) =

=0.3 × 1.2−0.4 × 0.02
=0.368
Limitations of LPM Model
a) Probability is a bounded number (it remains within 0% to 100%). But in an LPM, we can
predict values well above or well below the range. This is because the LPM model assumes
a linear relationship between the variables. But in reality, the relationship is more like an S
curve.

b) LPM estimator is conceptually inefficient due to heteroscadasticity. (this is a big problem. It

can never be homoscadastic)
Addressing the limitations of LPM model
The first limitation of LPM model (probability is bounded number) can be addressed through
making use of another model.
Logit regression
Let’s assume Probability of Default, PD = y
The Linear Probability Model,
y= β 0 + β 1 x ........... [ this isthe LPM Model ]
we may construct this as ,
e β +β x
0 1

y= β +β x [ equation .... ( i ) ]
e 0
+1 1

why did we construct this in this way? The explanation is given below.

Now CRM defines a concept which is known as the Odds Ratio. The Odds ratio appears like the
following:
y
Odds Ratio=
1− y
elaborating on the odds ratio using equation (1),
 β0 +β 1 x
e
β 0 +β 1 x
y e +1
=
1− y e
β +β x
0 1

1− β +β x
e 0
+11

λ
e
λ
e +1
= [ assuming λ= β 0 + β 1 x ]
eλ
1− λ
e +1
eλ
e λ +1
= λ λ
e +1 −e
e λ +1
λ
e
λ
e +1
=
1
λ
e +1
λ
e 1
= λ ÷ λ
e +1 e +1
eλ e λ +1
= λ ×
e +1 1
λ
=e
Therefore, it is proven,
y λ
=e
1− y
ln ( y
1− y )=ln ( e λ )

ln ( y
1− y )= λ ln ( e )

ln ( y
1− y )=λ×1

ln ( y
1− y )=λ

ln ( y
1− y )= β 0 + β 1 x ....... [ this is the logit regression ]

The final ‘logit regression’ will have its values within 0-100%.

LDA (Linear Discriminant Analysis)

Dividing the dataset into two groups.
Altman had information about 86-87 companies, some of which have defaulted. He came up with
22-23 variables and their coefficient so that he can divide the companies in question into two
groups, one group made up of companies which had defaulted, and the other which have not. At
first, this might seem very easy, but given the technology level of that time, it was very difficult.
Altman started with 22 variables, and he gradually reduced them to 5 variables (through trial and
error). Altman Z score is not a predictive model, but rather it allows division of the dataset into two
groups based on the probability of defaulting.
Z= ( 1.2× X 1 ) + ( 1.4 × X 2 ) + ( 3.3× X 3 ) + ( 0.6 × X 4 ) +( 1 × X 5 )
Where ,
Working Capital
X 1=
Total Assets
Retained Earnings
X 2=
Total Assets
EBIT
X 3=
Total Assets
Market Value of Equity +Book Value of Liabilities
X 4=
Total Liability
Sales
X 5=
Total Assets
LDM model is essentially a category or dimension reduction model which allows for efficient
categorization. There were two categories in Altman Z score, Default or Non-Default. The model
ensures linearity.
Decision Criteria

Range Decision
Z <1.80 High Default
1.81 < Z < 3.1 Neutral
Z > 3.2 Low Default

Math
A potential borrower has the following features

X1 1.2
X2 1
X3 -1
X4 1.5
X5 2

The Z value will come as = 2.4

Limitations of Altman Z score model
a) There are many phases between actual default and non default. Altman did not make room
for such phases. Default and non-default both are extreme categories.
b) Z scoring model is a dynamic model. If we work with the same variables on a different
dataset, the coefficient will change. Furthermore, for a different dataset, the variables
themselves may change. Again, the number of variables in a new dataset may be different as
well. This eventually means that the dimensions of the model may change in many ways.
 This is not a universal model. Altman himself says this is a model for US based
manufacturing firms. Altman himself came up with a different model for non-manufacturing
companies like financial services.
c) The numbers used in Altman’s model are fully based on annual reports (except for market
value of equity). Non numeric factors may influence the outcome as well. One important
factor is the value of Security. If the value of the security is very high, the borrower will be
discouraged from defaulting. There are many business dynasties for whom the reputation is
important. They are less likely to default as well.

RAROC model (Risk Adjusted Return on capital Model)

Math
Consider a $1 million loan. The duration of the loan is 5.4 years The borrower belongs to a AA
rating category. The current market interest rate on AA category bond is 6%. By observing the last
year’s information, we have concluded that the maximum possible yield change AA category is
1.5%.

Loan fees to earned 0.15%

Spread on the loan 1%
Bank’s cost of capital 12%

Should the bank provide the loan?

The model considers two sources of risk for the loan. One is interest rate risk (as illustrated by its
adoption of duration) and the other is credit risk (illustrated by its focus on credit rating).
Market yield is going to change based on the market’s expectation on whether the possibility of
default for a specific category increases or decreases. If the market perceives a category as more
risky, then the yield on that category will change. According to our assumption, the average change
will be zero.
We are not concerned about the minimum, rather the maximum as that will be the source of a risk.

Spread Loan Fee

The difference between the quoted interest rate Managing the lending process and other clerical
for lending and cost of capital for receiving the tasks.
fund

Δy
Loan Risk=duration × loan amount ×
1+ R
0.015
=5.4 × 1000000 ×
1+0.06
=76415.09
Net income of loan

Spread [ Loan Amount × Spread ] 10,000 [ Note : 1,000,000 ×1 % ]

Loan Fee [ Loan Amount × fees ] 5,000 [ Note : 1,000,000 ×0.5 % ]
Total (Net Income of the Loan) 15,000

Net income of the loan

RAROC=
Loan risk
15,000
= ×100
76415.6
=19.63 %
The loan should be disbursed if the RAROC is higher than the cost of capital [19.63% > 12%]
Limitations of RAROC model
a) The model has not made room for partial recovery.
Alternative version of RAROC model
in real life, collateral value may sometimes increase, and even may create situations where the value
of collateral exceeds the value of the loan. In this alternative version of RAROC model, partial
recovery issues are addressed.
Math:
Suppose you are evaluating a corporate borrower. The firm belongs to the BB category.

Annual net income on $1 loan 0.003

Default rate at the most extreme group (99%) 4%
Loan loss in case of default 80%
The cost of borrowing (bank) 12%

Requirement: should the bank distribute the loan?

Note
Let’s say there are 10,000 BB bonds being traded in the market. If we group bonds by taking 1%
of them in each group, each group will contain 100 bonds. At the 99% group, the bonds will be
9900 to 10,000. It has been seen the in the last year from these 100 bonds 4 bonds has defaulted.

0.003
RAROC= =9.38 %
0.04 × 0.80
Here, 9.38% < 12% (Cost of Capital), therefore the loan should not be disbursed.

Term structure based CRM

Math 1:
You are evaluating a BB rated borrower. There are a number of treasury bonds and BB rated
corporate bonds traded in the market.

Year Average yield of Average yield of BB Credit risk premium

treasury securities rated bonds
(Average yield of BB
rated bonds – Average
yield of treasury
securities)
1 8% 12% 4%
2 10% 15% 5%

Requirement:
a) Calculate the marginal probability of default and year 1
b) Calculate the marginal probability of default at year 2
c) Calculate the cumulative probability of default at year 1.
Probability of payment at year 1 = P 1
 1+i
P 1=
1+r
1.08
P 1=
1.12
P 1=0.9642

Therefore, the probability of default (marginal) at year 1 = 1 – 0.9642 = 3.58%

At year 2,

( 1.10 )2
Forward rate ( treasury bond ) = −1
1.08
=12.04 %
( 1.15 )2
forward rate for BB rated corporate bond = −1
1.12
=18.08 %
1.1204
P 1= =0.95
1.1808
Probability of Default at year 2=1 − 0.95=0.05
Cumulative:
Cumulative probability of default =1 − ( P 1 × P2 )=8.40 %

Why it is good?
a) Based on market expectations, not the risk manager’s personal views.
b) Uses Public information which is easily verifiable.
c) Is a forward looking method
Problems
a) The volume of trade in the bond market is very marginal. Therefore, the prices determined
by such small volume of trade is not reliable. Most of the bonds in question are not very
liquid. This is because American banks can only invest in investment grade bonds.
Math 2:

1 Year T security yield (i) 8%

1 year BB rated bond yield (c) 12%
Recovery if default (y) 90%

Requirement: Calculate the implicit credit risk premium.

1+i
Credit risk premium ,ϕ= − ( 1+i ) =0.39 %
y +P − P× y
Here, P is the probability of recovery calculated in Math example 1.
Let’s compare with the original situation,
without recovery, the credit risk premium is 4%. with recovery the credit risk premium is 0.39%.
This proves that if I provide the bank with a good security (collateral), then I will have to bear much
less cost. In this particular situation, with good collateral and recovery, my borrowing rate would
have been 8.00% + 0.39% = 8.39%.

Option based credit risk framework

Credit risk can be thought of as a output of option valuation.
Asset (A) can be financed in two ways, either by equity (E) or bonds (B). Equity holders are held as
residual claimers. Claim of the bondholders are secured by law. The value of equity is stochastic
value. There is a particular moment when A = B, and beyond that point the residual value of the
asset equals to the value of the equity.

Equity holder payoff closely resembles the call option buyer’s payoff.
bond holder’s payoff closely resembles the payoff of a put option seller.
Math Example
A borrower approached a bank for a loan of $100,000. He wants to purchase a real estate with a
leverage ratio of 0.9 = d.
The maturity of the loan is 1 year = m
The risk free rate is 5%
Historical volatility = 12% (also can be thought of as sigma) = σ
Requirement:
a) Calculate the credit risk premium under Martle’s option pricing framework
b) Calculate the quoted interest rate.

Credit risk premium=

−1
m [ 1
]
ln N ( h2 ) + N ( h1 )
d

h1 =
[1
] [ 1
− × σ 2 ×m − ln ( d ) − × ( 0.122 ) × 1− ln ( 0.9 )
2
=
2 ]
=− 0.938
σ × √m 0.12 × √ ( 1 )

h 2=
[1
2
2

=
] [ 1
− × σ ×m+ln ( d ) − × ( 0.12 ) × 1+ln ( 0.9 )
2
2
] =0.818
σ × √m 0.12 × √ ( 1 )
use Black Scholes Table find the probability
N ( h1 ) =N ( − 0.938 ) =0.1741 [ the value will be found through the Black Scholes Table ]
N ( h2 ) =N ( 0.818 ) =0.7933

Credit risk premium=

−1
m [ 1
]
ln N ( h2 ) + N ( h1 )
d
Credit risk premium=
−1
1 [
ln 0.7933+
1
0.9 ]
× 0.1741 =0.0133=1.33 %

Quoted interest rate = risk free rate + credit risk premium = 5% + 1.33% = 6.33%.
Observations
a) The higher the historical volatility, the higher the credit risk premium
b) The higher the leverage ratio, the higher the credit risk premium.
Limitations of the model
a) In this example real estate is discussed, which has a secondary market. When an asset is
taken which doesn’t have a secondary market, then the observed historical volatility data
will be absent.
b) Leverage ratio calculation requires data about market value. Therefore for an asset without
secondary market, the model will not be appropriate.

Qualitative issues considered in CRM

Two broad classifications for the factors:
1) Borrower specific factors
a) Reputation
b) Leverage
c) Volatility of earnings
d) Collateral
2) Economy specific factors
a) Business cycle
b) Reference interest rate
Borrower specific factors

Reputation Being long term client or customer entails some implicit agreements on
top of explicit contract as well. For example, despite having some
deficiency in paperwork, loan gets sanctioned. The Client always pays
back in the right time.
If all banks start disbursing loans based on reputation, that will prevent the
banks from acquiring new clients as the disburseable fund for new clients
become very low
Leverage After the threshold level, the marginal benefit from tax savings will be
less than the marginal cost of increased distress cost. For this reason,
many large companies make use of much less debt, example Microsoft’s
 D/E ratio is less than 1%
Volatility of earnings Start ups generally have higher volatility of earnings. Bankers do not like
that. Companies with higher volatility of earnings is less likely to receive
a loan.

Economy specific factors

Business Cycles Companies belonging to cyclical industries are less likely to receive loan
compared to companies that belong to defensive (?) industries (that do not
experience business cycle).
Reference interest rate Call money market is the primary defense. If the bank fails to secure
enough money from call money market, it goes to the Central Bank. The
short term rate at which Central bank lends money is called the reference
rate. If the reference rate goes up, all the interest rate goes up. Therefore, it
will be harder for newer companies for securing loans. Smaller companies
don’t find enough funds, but larger companies enough funds (as they can
use their reputation to issue unsecured commercial paper). In this
situation, smaller companies are starved of funds and default in greater
numbers.
 Chapter 11: Loan concentration risk
Migration analysis
Another name is Loan transition analysis.
Example
The following data represents the situation of a imaginary textile industry.
A is the best one, followed by B followed by C.
In 2022:

Risk Category A B C D [default]

A 90% 5% 5% 0%
[90% of securities [5% of securities [5% of securities [0% of securities
with A rating carrying A rating carrying A rating carrying A rating
retained their A became B rated became C rated defaulted in one
rating after one after one year] after one year] year]
year]
B 5% 85% 5% 5%
C 0% 10% 80% 10%

Requirement:
a) What is the probability that a ‘B’ rated bond will default will default within 2 years.
This will be done through making a decision making tree:
P ( A ∩ D ) =5 % × 0 %=0 %
P ( B∩ D ) =85 % ×5 %=4.25 %
P ( C ∩ D ) =5 % × 10 %=0.5 %
P ( D ) =5 %
P ( Default within 2 years ) =0 %+4.25 %+0.5 %+5 %=9.75 %
Loan concentration limit
Maximum loss that can be absorbed in a sector = 15%
Textile sector loan loss rate = 40%
1
Limit=0.15 × =37.5 %
0.40
No matter how well textile sector is doing, the maximum amount of disbursement for the textile
sector should never exceed 37.5%.

Modern Portfolio Theory

Markowitz’s approach gave us a working approach to viewing and managing portfolios. It can be
used for a wide variety of assets, including loan-type-assets.
Math

Weight Return Risk

Loan A 60% 6% 30%
Loan B 40% 4% 5%

Assume the correlation between A and B is = -0.2

a) Find the portfolio return
b) Find the portfolio risk
Use simple standard portfolio formulas.
In this case, the Portfolio return = 5.2%
Portfolio risk = 5.95%
Problems with MPT approach for loans
a) American financial institutions can be classified into 3 segments: Credit Union (only for the
credit union members, therefore no room for diversification), Savings Institutions (Generally
for mortgage and housing, which leaves little room for diversification), Banks (American
Banking industry is highly regulated. Even the permission to open branches in different
states took much time to actually take place. There was a time when foreign banks couldn’t
do banking operations in the US). Therefore, it is clear that there isn’t much room for
financial institutions to reduce their risk by using MPT for their loans.
b) The focus of Brac is this: a village woman is capable of producing nice shital paati.
However, she requires a loan to start production, something that no one will give her. Brac
filled the gap and provided her with microcredit. However, they saw that it is not the only
solution, as the produced shital paatis will need a market to be sold to. Brac had to fill that
gap as well. As the process continued, when Brac eventually received banking license, they
were highly skilled and experienced in SME banking and they were fully focused on SME
industry by choice. Therefore, although there are options for diversification, Brac chose not
to due to their ability and experience in a specific sector.
c) There are some practical and conceptual problem as well. When finding the portfolio return
and risk, individual return and risk were given. This is dependent on having a secondary
market for the asset in question. In real life situation, there is no secondary market for loans
and loans are seldom bought and sold (unless there is a merger and acquisition). Due to this,
it would not be possible for finding the return of the loans.

KMV model
KMV was an organization before it was acquired by Moody’s. They developed a possible solution
for the problem left unsolved by MPT model.
Math
 Weight Spread Fees Expected Loan Loss
probability of
default
Loan A 50% 3% 1% 25% 20%
Loan B 50% 2% 1.5% 40% 30%

Requirement:
a) Find the portfolio return
b) Find the portfolio risk
Return of a loan=( Spread+ Fees ) − ( Expected probability of default × Loan Loss )
Risk of a loan=√ Expected probability of default × ( 1 − probability of default ) × Loan Loss
First: find the return of loan A and B
Return of Loan A=( 3 %+1 % ) − ( 25 % ×20 % ) =3.95 %
Return of Loan B= ( 2 %+1.5% ) − ( 40 % ×30 % ) =3.38 %
Second: find the risk of loan A and B

Risk of Loan A= √ 0.25 × ( 1− 0.25 ) × 0.20=√ ( 0.25 × 0.75 ) × 0.20=8.667 %

Risk of Loan B= √ ( 0.40 × 0.60 ) × 0.30=14.7 %
Now just calculate the portfolio return and risk as per has been done in MPT model (normal
portfolio formulas).
Here, Portfolio return = 9.67% and Portfolio risk = 6.40%.

How probability of default was calculated

Equity is basically a call of option on firm’s assets.
Ē=f ( A , σ A , B̄ , r¯f , t̄ )
σ¯e = g ( σ A )

Here, there are 2 equations and 2 unknown parameters. They are A ,σ A .

Let’s take a asset where,
Value of asset = 100 m
Possible fluctuations of the price of the asset (risk) = 12 m
Outstanding debt = 80 m
Asset value may fall by around 20 m before the firm becomes technically bankrupt.

Note
In a normal distribution
within ±1 sigma = 65% of values
within ±2 sigma = 95% of values
within ±3 sigma = 99% of values

If we multiply 12 m by 1.67, the number will become 20 m. They used this number to find the
probability of default.

How was Loan loss was calculated

In America, rating is needed before issuing bonds. This type of historical information is generally
found:

Default Loan Loss

AAA *** **
AA ** **

Based on this data, they run a time series regression. The result of the time series regression is
included as the loan loss.

Is KMV a better model?

Definitely. It is basically an extension of the MPT model.