Value at Risk
Statistical Background
Dr. Handayani, [Link], MM, MHP, HIA, FLMI, AFSI, AAK, AAIJ, AMRP, FSAI
Value at Risk
When downside risk arises when realizations of 𝑋 are small, we define
the 100𝑝% Value-at-Risk as the solution of 𝑃(𝑋 ≤ 𝑥) = 𝑝. So we have:
100𝑝%ValueatRisk = 𝐹𝑋−1 (𝑝).
Typically, 𝑝 is chosen to be 0.05, 0.025, 0.01, or 0.005 (just like the
significance level for a statistical test). If 𝑝 = 0.01, then only 1 out of
100 times (on average) will the observed value of 𝑋 be less than the
VaR.
Definition of VaR
Definition: The Value at Risk at security level 𝑝 for a random variable 𝑋,
denoted VaR 𝑝 𝑋 , is the 100𝑝𝑡ℎ percentile of 𝑋:
VaR 𝑝 𝑋 = 𝜋𝑝 = 𝐹𝑋−1 𝑝
In practice, 𝑝 is selected to be close to: 95% or 99% or 99.5%.
VaR for Normal and Lognormal Distribution
Let 𝑧𝑝 be the 100𝑝𝑡ℎ percentile of a standard normal distribution.
Then, if 𝑋 is normal, VaR 𝑝 𝑋 = 𝜇 + 𝑧𝑝 𝜎.
VaR reduces to a standard deviation principle.
If 𝑋 is lognormal, then VaR 𝑝 𝑋 = 𝑒 𝜇+𝑧𝑝 𝜎
Example 1
Losses have a lognormal distribution with mean 10 and variance 300.
Calculate the VaR at security levels 95% and 99%.
Solution 1
Let 𝑋 be the loss random variable. We back out parameters μ and σ by
matching moments.
The second moment is 𝑉𝑎𝑟 𝑋 + 𝐸 𝑋 2 = 300 + 102 = 400
The moments of a lognormal are:
𝜇+0.5𝜎 2
𝐸 𝑋 =𝑒
2
𝐸 𝑋2 = 𝑒 2𝜇+2𝜎
𝑉𝑎𝑟 𝑋 = 𝐸 𝑋 2 − 𝐸 𝑋 2
Solution 1
𝜇+0.5𝜎 2
𝑒 = 10 → 𝜇 + 0.5𝜎 2 = ln 10
2𝜇+2𝜎 2
𝑒 = 400 → 2𝜇 + 2𝜎 2 = ln 400
𝜎 2 = ln 400 − 2 ln 10 = 5.9915 − 2 2.3026 = 1.3863
𝜎 = 1.3863 = 1.1774
𝜇 = ln 10 − 0.5𝜎 2 = 2.3026 − 0.5 1.3863 = 1.6094
Solution 1
The 95𝑡ℎ percentile is:
VaR 0.95 = 𝑒 𝜇+1.645𝜎 = 𝑒 1.6094+1.645 1.1774 = 𝟑𝟒. 𝟔𝟖
The 99𝑡ℎ percentile is:
VaR 0.99 = 𝑒 𝜇+2.326𝜎 = 𝑒 1.6094+2.326 1.1774 = 𝟕𝟕. 𝟑𝟔
[Link](95%) = 1.644854
[Link](99%) = 2.326348
Example 2
The returns on an investment follow a normal distribution with mean
7.5% and standard deviation 35%. Find the 10% Value-at-Risk of the
investment return.
Solution 2
Let 𝑅 be the random (one-year) investment return. Then 𝑅 ∼
𝑁(0.075,0.352). We have
𝑟 − 0.075 𝑟 − 0.075
𝐹𝑅 𝑟 = 𝑃 𝑅 ≤ 𝑟 = 𝑃 𝑍 ≤ =𝑁
0.35 0.35
To find the 100𝑝-th percentile of 𝑅, we set 𝐹𝑅 𝑟 = 𝑝:
𝑟 − 0.075
𝑁 =𝑝
0.35
Solution 2
and we have
𝑟 = 0.075 + 0.35𝑁 −1 𝑝
The 10% VaR is:
0.075 + 0.35 −1.28155 = −𝟎. 𝟑𝟕𝟑𝟓𝟒
[Link](10%) = -1.28155
Value at Risk
A very useful rule for computing the percentile of a function of random
variable is the following:
If 𝑌 = 𝑔(𝑋), where 𝑔 is a continuously increasing function, then the
100𝑝-th percentile for 𝑌 is 𝑔(𝑿𝒑 ), where 𝑿𝒑 = 𝑭−𝟏 (𝑝) is the 100𝑝-
th-percentile for 𝑋
Value at Risk
Let us apply the result to redo Example 2. We write
𝑅 = 0.075 + 0.35𝑍
Where 𝑍 ∼ 𝑁 0,1 . Here 𝑔 𝑥 = 0.075 + 0.35𝑥 is strictly increasing.
Since the 10𝑡ℎ percentile of 𝑍 is 𝑁 −1 0,1 = −1.28155, the 10𝑡ℎ
percentile of 𝑅 is 0.075 + 0.35 −1.28155 = −0.37354
If 𝑔 is a continuously decreasing function, then the 100𝑝-th percentile
of 𝑌 is 𝑔(𝑋1−𝑝 ), where 𝑋1−𝑝 = 𝐹 −1 1 − 𝑝
Value at Risk
When the adverse outcomes of 𝑋 correspond to the realization of large
values of 𝑋, 𝑝 is chosen to be close to 1. Typical values are 0.95, 0.975,
0.99, and 0.995. The VaR can be conceived as the amount of capital
required, with a high probability, to protect against insolvency.
Example 3
The operational loss 𝑋 of a line of business can be modelled by a single-
parameter Pareto distribution with the following density function:
𝛼𝜃 𝛼
𝑓 𝑥 = 𝛼+1 , 𝑥≤𝜃
𝑥
Find the 100𝑝% Value-at-Risk of 𝑋.
Solution 3
The cumulative distribution function is
𝑥 𝛼
𝛼𝜃 𝛼 𝜃
F 𝑥 = න 𝛼+1 𝑑𝑦 = 1 − ,𝑥 ≤ 𝜃
𝑦 𝑥
𝜃
𝜃 𝛼
We set F 𝑥 = 𝑝 → 1 − =𝑝
𝑥
𝜃
𝑥= 1/𝛼
1−𝑝
REMINDER
𝑥 𝑥 𝑥
𝛼
𝛼𝜃
න 𝛼+1 𝑑𝑦 = න 𝛼𝜃 𝛼 𝑦 − 𝛼+1 𝑑𝑦 = න 𝛼𝜃 𝛼 𝑦 −𝛼−1 𝑑𝑦
𝑦
𝜃 𝜃 𝜃
𝑦 −𝛼−1+1 𝑥 𝑥
= 𝛼𝜃 𝛼 ቤ = −𝜃 𝑦 ቤ = −𝜃 𝛼 𝑥 −𝛼 − −𝜃 𝛼 𝜃 −𝛼
𝛼 −𝛼
−𝛼 − 1 + 1 𝜃 𝜃
𝛼
𝜃
=1−
𝑥
Value at Risk
If we only have a random sample of risks but not its true distribution, we
can estimate the sample Value-at-Risk by first ranking the data in the
sample and then “picking” the 100𝑝-th percentile.
Value at Risk
For example, if we have a sample with 11 data points, we can estimate
the 50th percentile, the 25th percentile, and the 75th percentile using the
6th, 3rd and the 9th ranked data:
𝑥(1), 𝑥(2), 𝑥(3), 𝑥(4), 𝑥(5), 𝑥(6), 𝑥(7), 𝑥(8), 𝑥(9), 𝑥(10), 𝑥(11),
where 𝑥(1) denotes the smallest realization, and so on. However, it
does not make much sense to estimate the 5% VaR from 11 points.
What we know is that it should be less than the smallest observation
𝑥(1)! To estimate a 100𝑝-th percentile for very small and very large 𝑝, a
large sample is needed
Example 4
The following is the lowest 15 values of a sample of 150 weekly stock
returns:
−0.9704, −0.5105, −0.4713, −0.4669, −0.4340,
−0.4226, −0.4185, −0.4105, −0.3986, −0.3634,
−0.3575, −0.3496, −0.3248, −0.3207, −0.3147.
Estimate the 5% Value-at-Risk of the weekly stock returns.
Solution 4
With a sample size of 150, and with 151 × 0.05 = 7.55, the 5% VaR
should be in between the 7𝑡ℎ and the 8𝑡ℎ ranked data. We can do a
linear interpolation as follows:
(−0.4185) × 0.45 + (−0.4105) × 0.55 = −0.4141,
or to be conservative, we can also take −0.4185 as the 5% VaR.
VaR for Exponential Distribution
For 𝑋 exponential with mean 𝜃, if 𝐹 𝑥 = 𝑝, then:
𝑒 −𝑥Τ𝜃 = 1 − 𝑝
𝑥 = −𝜃 ln 1 − 𝑝
so VaR 𝑝 𝑋 = −𝜃 ln 1 − 𝑝
VaR for Pareto Distribution
For 𝑋 following a two-parameter Pareto distribution with parameters 𝛼
and 𝜃, if 𝐹 𝑥 = 𝑝, then:
𝛼𝜃 𝛼
𝑓 𝑥 = 𝛼+1
, 𝑥≤𝜃
(𝑥 + 𝛼)
𝛼
𝜃
𝐹 𝑥 =1−
𝜃+𝑥
VaR for Pareto Distribution
𝛼
𝜃 𝛼 𝜃 𝜃 1− 1−𝑝
= 1−𝑝 → 𝜃+𝑥 = → 𝑥=
𝜃+𝑥 𝛼
1−𝑝 𝛼
1−𝑝
𝜃 1−𝛼 1−𝑝
VaR 𝑝 𝑋 = 𝛼
1−𝑝
Example 5
Losses follow a Pareto distribution with mean 10 and variance 300.
Calculate the VaR at security levels 95% and 99%.
Solution 5
Let 𝑋 be the loss random variable. We back out parameters α and θ by
matching moments.
The second moment is 𝑉𝑎𝑟 𝑋 + 𝐸 𝑋 2 = 300 + 102 = 400
The moments of a Pareto are:
θ
𝐸 𝑋 = 𝛼>1
α−1
2θ2
𝐸 𝑋2 = 𝛼>2
𝛼−1 𝛼−2
𝑉𝑎𝑟 𝑋 = 𝐸 𝑋 2 − 𝐸 𝑋 2
Solution 5
θ
𝐸 𝑋 = = 10
α−1
2θ2
𝐸 𝑋2 = = 400
𝛼−1 𝛼−2
2
θ θ2
= = 100
α−1 α−1 α−1
Solution 2
We divide the square of the first equation into the second.
2θ2 𝛼−1 𝛼−1 2 𝛼−1
× 2
=4 → =4
𝛼−1 𝛼−2 𝜃 𝛼−2
2𝛼 − 2 = 4𝛼 − 8 → 𝛼 = 3
Plugging this into the equation for 𝐸 𝑋 , we get θ = 20. The 95𝑡ℎ
percentile of the Pareto distribution is 𝑥 such that 𝑆 𝑥 = 0.05.
Solution 5
3 3
𝜃 20
𝑆 𝑥 = = 0.05 → = 0.05
𝜃+𝑥 20 + 𝑥
20 3 20
= 0.05 = 0.368403 → 20 + 𝑥 = = 54.2884
20 + 𝑥 0.368403
VaR 0.95 = 𝑥 = 𝟑𝟒. 𝟐𝟗
Solution 5
Similarly, the 99𝑡ℎ percentile is 𝑥 such that 𝑆 𝑥 = 0.01
3 3
𝜃 20
𝑆 𝑥 = = 0.01 → = 0.01
𝜃+𝑥 20 + 𝑥
20 3
= 0.01 = 0.215443
20 + 𝑥
20
VaR 0.99 =𝑥= − 20 = 𝟕𝟐. 𝟖𝟑
0.215443
Exercise 1
Losses 𝑋 follow a Pareto distribution with parameters 𝛼 = 2 and 𝜃 =
1000. An insurance company pays 𝑌 = 𝑚𝑎𝑥 0, 𝑋 − 2000 for these
losses.
Calculate VaR 0.99 𝑌 .
Solution 1
The 99th percentile is 𝑥 such that 𝑆 𝑥 = 0.01
2 2
𝜃 1000
𝑆 𝑥 = = 0.01 → = 0.01
𝜃+𝑥 1000 + 𝑥
1000
= 0.01 = 0.1
1000 + 𝑥
1000
VaR 0.99 =𝑥= − 1000 = 9000
0.1
Solution 1
𝑌 = 𝑚𝑎𝑥 0, 𝑋 − 2000
𝑌 = 𝑚𝑎𝑥 0,9000 − 2000 → 𝑌 = 𝑚𝑎𝑥 0,7000 = 7000
Exercise 2
Losses 𝑋 follow a paralogistic distribution with 𝛼 = 2 and 𝜃 = 1000.
Calculate VaR 0.99 𝑋 .
Paralogistic distribution:
1/𝛼
−1/𝛼
VaR 𝑝 𝑋 = 𝜃 1−𝑝 −1
𝛼
1
𝑆 𝑥 = 𝛼
1 + 𝑥/𝜃
Solution 2
Method 1
1/𝛼
VaR 𝑝 𝑋 = 𝜃 1−𝑝 −1/𝛼 −1
Substituting 𝑝 = 0.99, 𝛼 = 2, 𝜃 = 1000,
1/2
VaR 0.99 𝑋 = 1000 1 − 0.99 −1/2 −1 = 𝟑𝟎𝟎𝟎
Solution 2
Method 2
We want the 99th percentile, so set 𝐹 𝑥 = 0.99 and solve for 𝑥.
𝑆 𝑥 =1−𝐹 𝑥
2
1
𝐹 𝑥 =1− 2 = 0.99
𝑥
1+
1000
Solution 2
1
2 0.01 = 0.1
𝑥
1+
1000
𝑥 2 𝑥
1+ = 10 → = 9 = 3 → 𝑥 = 𝟑𝟎𝟎𝟎
1000 1000