Expected Loss of credit products and credit portfolios (I)
Theodor Munteanu: Quantitative Risk Analyst
For a portfolio of 𝑛 credits, the aggregated loss is 𝐿 = ∑𝑛𝑖=1 𝐸𝐴𝐷𝑖 ⋅ 𝐿𝐺𝐷𝑖 ⋅ 1(𝜏𝑖 < 𝑇𝑖 ) 𝑤here 𝐸𝐴𝐷𝑖 =
exposure at default at credit 𝑖, 𝐿𝐺𝐷𝑖 = loss given default for credit 𝑖 and 𝐿𝐺𝐷𝑖 = 1 − 𝑅𝑖 , 𝜏𝑖 = time of
default of credit 𝑖 and 𝑇𝑖 = expiry of credit 𝑖
Example 1:
Suppose a retail client has entered a loan with a bank whereby a grant of 3,000,000 euros is given for a
period of 10 years to be returned, all at expiry.
The client has a hazard rate of default 𝜆 = 130 𝑏𝑝𝑠 (1.3%) expressed annually.
In case of default (bankruptcy) it would be recovered 45% of the notional amount. What is the required
capital for this loan?
Solution:
If not stated otherwise if the hazard rate of default is 𝜆 then the time of default is 𝜏 ∼ 𝐸𝑥𝑝(𝜆)
The stochastic loss for 1 single credit can be expressed as follows (where 𝐿𝐺𝐷 = 1 − 𝑅𝑒𝑐𝑜𝑣):
𝐿 = 𝐸𝐴𝐷 ⋅ 𝐿𝐺𝐷 ⋅ 1(𝜏 < 1) = 3,000,000 ⋅ (1 − 45%) ⋅ 1(𝜏 < 1) where 1(𝜏 < 1) ∼ 𝐵𝑒(𝑝 = 1 −
𝑒 −0.013 ) = 𝐵𝑒(1.29%) (𝐸𝐴𝐷 =exposure at default = 3,000,000 𝐸𝑈𝑅) which remains constant in the
first year.
The required capital for such a loan is 𝑅𝐶 = 𝐹𝐿−1 (0.999)
0 1,650,000
Therefore 𝐿 ∼ ( ) ⇒ 𝐹𝐿−1 (0.999) = 1,650,000
98.71% 1.29%
Example 2:
Supose that the recovery rate is uncertain but is distributed 𝑈𝑛𝑖𝑓(0,1) and the hazard rate is the same
as above. The time of default 𝜏 and the recovery rate 𝑅 are independent.
What is the required capital for such a credit?
Solution:
𝑃(𝐿 < 𝑥) = 𝑃(𝐿 = 0) + 𝑃(0 < 𝐿 < 𝑥, 𝜏 < 1) = 𝑃(𝐿 = 0) + 𝑃(0 < 𝐸𝐴𝐷 ⋅ 𝐿𝐺𝐷 < 𝑥, 𝜏 < 1)
𝑥
= 𝑃(𝐿 = 0) + 𝑃 (𝐿𝐺𝐷 < ) ⋅ 𝑃(𝜏 < 1) = 0.999
𝐸𝐴𝐷
−1 0.999−𝑒 −0.013
Therefore 𝐹𝐿−1 (0.999) = 𝐹𝑈(0,1) ( ) ⋅ 𝐸𝐴𝐷 = 2,767,727.51 EUR.
1−𝑒 −0.013
Example 3:
Suppose that the recovery rate is uncertain but is distributed 𝐵𝑒𝑡𝑎(2,2) and the hazard rate is the same
as above. The time of default 𝜏 and the recovery rate 𝑅 are independent.
What is then the required capital for such a credit?
Solution:
𝐿 = 𝐸𝐴𝐷 ⋅ 𝐿𝐺𝐷 ⋅ 1(𝜏 < 1) = 𝐸𝐴𝐷 ⋅ (1 − 𝑅) ⋅ 1(𝜏 < 1)
Because 𝑅 ⊥ 𝜏 ⇒ 1 − 𝑅 = 𝐿𝐺𝐷 ⊥ 𝜏
Let’s practice a bit first.
Question) What is the probability that the loss is 0?
𝑃(𝐿 = 0) = 𝑃(𝜏 > 1) = 98.71%
If we take 𝑥 > 0, 𝑃(𝐿 < 𝑥) = 𝑃(𝐿 = 0) + 𝑃(0 < 𝐿 < 𝑥) = 98.71% + 𝑃(0 < 𝐿 < 𝑥, 𝜏 < 1) =
𝑥
𝑃(𝐿 = 0) + 𝑃(𝐿𝐺𝐷 ⋅ 𝐸𝐴𝐷 < 𝑥)𝑃(𝜏 < 1) = 98.71% + 1.29% ⋅ 𝑃 (𝐿𝐺𝐷 < 𝐸𝐴𝐷) = 99.9%
𝑥
Therefore 𝑃 (𝐿𝐺𝐷 < ) = 99.9% − 98.71% = 1.19%
𝐸𝐴𝐷
The cumulative distribution function for the 𝐵𝑒𝑡𝑎(2,2) distribution:
𝑥
𝐹𝐵𝑒𝑡𝑎(2,2) (𝑥) = ∫0 6𝑢(1 − 𝑢)𝑑𝑢 = 3𝑥 2 − 2𝑥 3
𝑥
Therefore 𝑃 (𝐿𝐺𝐷 < 𝐸𝐴𝐷) = 𝑃(𝐿𝐺𝐷 < 𝑦) = 3𝑦 2 − 2𝑦 3 = 1.19% ⇒ 𝑦 ∗ = 8.18%
𝑥
So 𝐸𝐴𝐷 = 8.18% ⇒ 𝑥 = 8.18% ⋅ 3,000,000 = 245,400 euros.
Example 4:
Suppose a bank’s retail portfolio has 3 retail loans with face values 𝑁1 = 1,000,000 𝐸𝑈𝑅𝑂, 𝑁2 =
2,000,000 𝐸𝑈𝑅, 𝑁3 = 3,000,000 𝐸𝑈𝑅 all expiring in 5 years.
The recovery rates are 40%, 50% and 60% and the default times are 𝜏1 ∼ 𝐸𝑥𝑝(𝜆1 ), 𝜏2 ∼ 𝐸𝑥𝑝(𝜆2 ), 𝜏3 ∼
𝐸𝑥𝑝(𝜆3 ), 𝜆1 = 3%, 𝜆2 = 4%, 𝜆3 = 5%
Solution:
The total loss can be expressed as 𝐿 = 𝑁1 (1 − 𝑅1 )1(𝜏1 < 1) + 𝑁2 (1 − 𝑅2 )1(𝜏2 < 1) +
𝑁3 (1 − 𝑅3 )1(𝜏3 < 1)
ln(1−𝑈1 ) ln(1−𝑈2 ) ln(1−𝑈3 )
I use Monte Carlo simulations of (𝜏1 , 𝜏2 , 𝜏3 ) = − ( 𝜆1
, 𝜆 , 𝜆 ) , (𝑈1𝑖 , 𝑈2𝑖 , 𝑈3𝑖 )𝑖=1,𝑛
2 3 𝑠
The 99.9% quantile is 𝐹𝐿−1 (0.999) = 2,200,000 EUR.
Example 5:
a. Suppose a bank gives a grant to a client worth 1,250,000 euros that is payable in 5 years while a
real estate property is used as guarantee. The value of the real estate property is 450,000.
The hazard rate is 3% per first year.
b. Suppose that the real estate property is a geometric Brownian motion with drift estimated at
4.5% and volatility of returns is 20%.
Solution:
450,000
a. 𝑳 = (1 − 𝑅) ⋅ 𝐸𝐴𝐷 ⋅ 1(𝜏 < 1), 1 − 𝑅 = 1 − = 0.64
1,250,000
0 800,000
Or rather 𝐿 = 800,000 ⋅ 1(𝜏 < 1) ⇒ 𝐿 ∼ ( )
exp (−0.03) 1 − exp(−0.03) = 2.95%
The 99.9% quantile is 800,000
b. This time the recovery rate is stochastic depending on the property value 𝑉𝑡 whose dynamic is
𝑑𝑉𝑡 = 𝜇𝑉𝑡 𝑑𝑡 + 𝜎𝑉𝑡 𝑑𝐵𝑡 , 𝑉0 = 450,000$
𝐿𝜏 = (1 − 𝑅𝜏 ) ⋅ 𝐸𝐴𝐷 ⋅ 1(𝜏 < 1)
𝑉 𝑒 𝜇⋅𝜏
For simplicity I will consider 𝑉𝜏 = 𝐸[𝑉𝜏 ] = 𝑉0 𝑒 𝜇𝜏 ⇒ 𝐿𝜏 = (1 − 1,250,000
0
) ⋅ 𝐸𝐴𝐷 ⋅ 1(𝜏 < 1)
I simulate 𝜏 ∼ 𝐸𝑥𝑝(𝜆)and 𝐹𝐿−1
𝜏
(0.999) = 799,283.30
Example 6:
Suppose an SME client has entered a loan for 10 years. The grant is for 850,000 euros, and there is a
guarantee of 250,000 euros from unused land. The hazard rate is 4.5% for the first year, 3% for the 2nd
year and 1.5% for the remaining period.
What is the expected loss and required capital? What is the unexpected loss?
Solution:
For an SME, according to the F-IRB approach one uses a 2.5 year horizon.
𝐿 = 𝐸𝐴𝐷 ⋅ 𝐿𝐺𝐷 ⋅ 1(𝜏 < 2.5) ⇒ 𝐸[𝐿] = 𝐸𝐴𝐷 ⋅ 𝐸[𝐿𝐺𝐷] ⋅ 𝑃(𝜏 < 2.5)
850,000 − 250,000
𝐸𝐴𝐷 = 850,000; 𝐸[𝐿𝐺𝐷] = = 70.58%
850,000
The survival probability for 2.5 years is 𝑆(2.5) = exp(−𝜆1 ⋅ 1 − 𝜆2 ⋅ 1 − 𝜆3 ⋅ (2.5 − 2)) =
1
exp (−0.045 − 0.03 − 0.015 ⋅ 2) = exp(−0.0825) ⇒ 𝑃(𝜏 < 2.5) = 1 − 𝑆(2.5) = 7.91%
So 𝐸[𝐿] = 600,000 ⋅ 7.91% = 47,460.0 EUR.
0 600,000
So: 𝐿 ∼ ( ) therefore 𝐹𝐿−1 (0.999) = 600,000 which is also the required capital.
92.09% 7.91%
𝑈𝐿𝛼 = 𝑈𝐿99.9% = 600,000 − 47,460.0 = 552,540 EUR.
