Estimating Market Risk

Measures

Parametric VaR – Refresher

 With Mean:
 VaR1-yr, 5% = [μannual – (z-value × σannual)] × Portfolio Value

 No Mean:
 VaR1-yr,5% = z-value × σ × Portfolio value

© Kaplan, Inc. 1

1
 Parametric VaR – Changing the Time Period

 Annual:
 VaR1-yr, 5% = [μannual – (z-value × σannual)] × Portfolio Value

μ / days σ / √days

 Daily:
 VaR1-day, 5% = [μdaily – (z-value × σdaily)] × Portfolio Value

© Kaplan, Inc. 2

VaR – Historical Simulation

 Assumptions/restrictions:
 Future returns follow historical return process
 Does not require distribution assumption
 Unable to account for shifts in parameter values

 Method:
 Order historical returns (low to high)
 Identify observation corresponding to specified confidence level
© Kaplan, Inc. 3

2
 Identifying the VaR Limit
 In theory, VaR is the quantile that demarcates the tail
from the body of the distribution.
 With 1,000 observations at a 95% confidence level, there
is a certain level of arbitrariness in how the ordered
observations relate to VaR.
 In other words, do we take the VaR to be the 50th observation
(i.e., α × n), the 51st observation [i.e., (α × n) + 1], or some
combination of them (i.e., average of the two observations
closest to α × n)?
© Kaplan, Inc. 4

VaR – Historical Simulation

© Kaplan, Inc. 5

3
 Identifying the VaR Limit (cont.)

 Identifying the ordered observation is an approximation

issue, since taking VaR to be the 51st highest observation
is not unreasonable (and is done in this assigned reading).
 However, on past FRM exams, VaR using the historical
simulation method has been calculated as just (α × n)—in
this case, as the 50th observation.

© Kaplan, Inc. 6

VaR – Parametric Estimation

 VaR given profit/loss statistics (normal distribution):

V a R = − μ P /L + σ P /L × z α
Example: Estimated mean and standard deviation for
profit/loss distribution are $10 million and $12 million,
respectively. Compute VaR at 95% level.
Answer: V a R = – $ 1 0 m illio n + $ 1 2 m illio n × 1 .6 5
= $ 9 .8 m illio n 7
© Kaplan, Inc.

4
 VaR – Parametric Estimation (cont.)

 VaR given return statistics (normal distribution):

VaR = (− μ r + σr × zα ) × Pt−1
Example: Estimated mean and standard deviation of
returns are 10% and 20%, respectively. Given that the
beginning of period value was $100, compute VaR at a
95% level.
Answer: V a R = (− 1 0 % + $ 2 0 % × 1 .6 5 ) × 1 0 0
© Kaplan, Inc. = $ 2 3 .0 0 8

Lognormal VaR
 The lognormal distribution is right skewed with positive
outliers and bounded below by zero.
 Therefore, the lognormal distribution is commonly used
to counter the possibility of negative asset prices.
 If we assume that geometric returns follow a normal
distribution, then the natural logarithm of asset prices
follows a normal distribution and asset prices themselves
follow a lognormal distribution.
© Kaplan, Inc. 9

5
 VaR – Parametric Estimation

 Lognormal VaR given return statistics:

(
V a R = Pt − 1 × 1 − e μ R − σ R × z α )
Example: Estimated mean and standard deviation of
returns are 10% and 20%, respectively. Given that the
beginning of period value was $100, compute lognormal
VaR at a 95% level.
Answer: V a R = 1 0 0 × (1 − e xp [1 0 % − $ 2 0 % × 1 .6 5 ] )
© Kaplan, Inc.
= $ 2 0 .5 5 10

GARP 2016 Market Risk Q2

The annual mean and volatility of a portfolio are 12% and 30%, respectively. The current value
of the portfolio is GBP 2,500,000. How does the 1-year 95% VaR that is calculated using a
normal distribution assumption (normal VaR) compare with the 1-year 95% VaR that is
calculated using the lognormal distribution assumption (lognormal VaR)?

A. Lognormal VaR is greater than normal VaR by GBP 487,050.

B. Lognormal VaR is greater than normal VaR by GBP 154,500.
C. Lognormal VaR is less than normal VaR by GBP 487,050.
D. Lognormal VaR is less than normal VaR by GBP 154,500.

© Kaplan, Inc. 11

6
 GARP 2016 Q2—Answer
Lognormal VaR 1-year 95%
VaRLN = 1 − e (
μ− σ× z value)
 × portfolio value
 
0.12 − ( 0.30×1.645 )
VaRLN = 1 − e  × GBP 2.5m = GBP 779,198
 

Normal VaR 1-year 95% Difference GBP –154,552

VaR = μ − ( z value × σ )  × portfolio value
VaR = 0.12 − (1.645 × 0.30 )  × GBP 2.5m = GBP933,750

Difference:
 Instructor Tip: Lognormal VaR is always lower than Normal VaR
 Correct answer: D
© Kaplan, Inc. 12

GARP 2018 Market Risk Q2

A risk manager is estimating the market risk of a portfolio using both the arithmetic return with
normal distribution assumption and the geometric returns with lognormal distribution
assumptions. The manager gathers the following data on the portfolio:
• Annualized average of arithmetic returns: 15%
• Annualized standard deviation of arithmetic returns: 35%
• Annualized average of geometric returns: 0.3%
• Annualized standard deviation of geometric returns: 44%
• Current portfolio value: EUR 4,800,000
• Trading days in a year: 252

Assuming both daily arithmetic returns and daily geometric returns are serially independent,
which of the following statements is correct?
A. 1-day normal 95%VaR = 4.45% and 1-day lognormal 95%VaR = 3.57%.
B. 1-day normal 95%VaR = 3.57% and 1-day lognormal 95%VaR = 4.45%.
C. 1-day normal 95%VaR = 4.45% and 1-day lognormal 95%VaR = 4.49%.
D. 1-day normal 95%VaR = 3.57% and 1-day lognormal 95%VaR = 3.55%.

© Kaplan, Inc. 13

7
 GARP 2018 Q2—Answer
Lognormal VaR 1-year 95%
VaRLN = 1 − e (
μ− σ× z value)
 × portfolio value
 
0.44 0.003
daily vol = = 2.7717395% daily E / R = = 0.001190476%
252 252

VaRLN = [1 − e0.00001190476−(0.027717395×1.645) ] = 4.45%

Normal VaR 1-year 95%

0.35 0.15
daily vol = = 0.022047928 daily E / R = = 0.0595238%
252 252
VaR = μ − ( z value × σ )  × portfolio value

VaR = 0.000595238 − (1.645 × ) 0.022047928 = 3.57%

 Correct answer: B
© Kaplan, Inc. 14

Portfolio VaR

 Undiversified VaR
 VaRP = VaRA + VaRB

 Diversified VaR
 VaR VaR + VaR + 2 × VaR × VaR × Corr /

© Kaplan, Inc. 15

8
 Net Relationships and Benefit of Diversification

 Net Relationships
 Long – Short, Assets – Liabilities, Fund – Benchmark

 VaR VaR + VaR − 2 × VaR × VaR × Corr /

 Benefit of Diversification
 BoD = Undiversified VaR – Diversified VaR

© Kaplan, Inc. 16

GARP 2018 Market Risk Q9 (also 2017 Q9)

A wealth management firm has a portfolio consisting of USD 48 million invested in U.S. equities
and USD 35 million invested in emerging market equities.The1-day 95% VaR for each
individual position is USD 1.2 million. The correlation between the returns of the U.S. equities
and emerging market equities is 0.36. While rebalancing the portfolio, the manager in charge
decides to sell USD 8 million of the U.S. equities to buy USD 8 million of the emerging market
equities. At the same time, the CRO of the firm advises the portfolio manager to change the
risk measure from1-day 95% VaR to 10-day 99% VaR. Assuming that returns are normally
distributed and that the rebalancing does not affect the volatility of the individual equity
positions, by how much will the portfolio VaR increase due to the combined effect of portfolio
rebalancing and change in risk measure?

A. USD 4.529 million.

B. USD 6.258 million.
C. USD 7.144 million.
D. USD 7.223 million.

© Kaplan, Inc. 17

9
 GARP 2018 Q9—Answer
Before portfolio rebalancing (95% 1-day VaR – U.S. 48m, EM 35m)
2 2
VaRP = VaRUS + VaREM + 2 × VaRUS × VaREM × CorrUS,EM

VaRP = 1.22 + 1.22 + 2 × 1.2 × 1.2 × 0.36 = USD1.979m Difference = $1.979m – $9.20m
= $7.22m

After portfolio rebalancing (reduce VaRUS by 16.67%, increase VaREM by 22.86%)

VaRP,1− day = 12 + 1.474322 + 2 × 1 × 1.47432 × 0.36 = USD 2.058m

Amend to 99% and 10 day:

2.326 2.326
VaR10− day,99% = VaR1− day × 10 × VaR10− day,99% = 2.058 × 10 × = USD9.20m
1.645 1.645

© Kaplan, Inc. 18

GARP 2014 Market Risk Q20

A risk manager is evaluating a pairs trading strategy recently initiated by one of the firm’s traders.
The strategy involves establishing a long position in Stock A and a short position in Stock B. The
following information is also provided:

• 1-day 99% VaR of Stock A is USD 100 million

• 1-day 99% VaR of Stock B is USD 125 million

The estimated correlation between long positions in Stock A and Stock B is 0.8. Assuming that the
returns of Stock A and Stock B are jointly normally distributed, the 1-day 99% VaR of the
combined positions is closest to:

A. USD 0 million.
B. USD 75 million.
C. USD 160 million.
D. USD 225 million.

© Kaplan, Inc. 19

10
 GARP 2014 Q20—Answer
Use the portfolio VaR formula; remember the net relationship of long – short!
VaRP = VaR2A + VaRB2 − 2 × VaR A × VaRB × CorrA,B

VaRP = 1002 + 1252 − 2 × 100 × 125 × 0.8 = USD 75m

Correct answer: B

© Kaplan, Inc. 20

Expected Shortfall
 The ES provides an estimate of the tail loss by averaging the VaRs for
increasing confidence levels in the tail.
 Example: Given the VaRs listed below, compute the ES at the 95%
confidence level.
Confidence Level VaR
96% 1.751
97% 1.881
98% 2.054
99% 2.326
1
Answer: ES =
4
[1 .7 5 1 + 1 .8 8 1 + 2 .0 5 4 + 2 .3 2 6 ] = 2 .0 0 3
© Kaplan, Inc. 21

11
 GARP 2017 Market Risk Q52
A wealth management firm has INR 56 billion in assets. The portfolio manager computes the daily VaR at
various confidence levels as follows:
Confidence Level VaR (INR)
95.0% 226,665,000
95.5% 230,197,500
96.0% 234,000,000
96.5% 244,237,500
97.0% 253,012,500
97.5% 261,787,500
98.0% 272,317,500
98.5% 286,357,500
99.0% 304,785,000
99.5% 333,157,500

What is the closest estimate of the daily ES at the 97.5% confidence level?
A. INR 262 million.
B. INR 264 million.
C. INR 292 million.
D. INR 299 million.

© Kaplan, Inc. 22

GARP 2017 Q52—Answer

Expected shortfall (average loss beyond the 97.5% VaR of $261,787,500)

 98.0% 272,317,500
 98.5% 286,357,500
 99.0% 304,785,000
 99.5% 333,157,500
 Total: 1,186,617,500 / 4
 Expected shortfall (average): $299,154,374

Correct answer: D

© Kaplan, Inc. 23

12
 Coherent Risk Measures

 Definition: A more general risk measure than either VaR or

ES
 Method:
 For (n) quantiles, (n – 1) breakpoints constructed with equal
probability mass
 Coherent measure is weighted average of quantiles
 Weights are user specific based on individual risk aversion

© Kaplan, Inc. 24

Coherent Risk Measures (cont.)

 In order to properly measure risk, one must first clearly

define what is meant by a measure of risk.
 The properties of a coherent risk measure are [ρ(•) = risk
measure for random events]:
 Monotonicity: Y ≥ X → ρ(Y) ≤ ρ(X)
 Subadditivity: ρ(X + Y) ≤ ρ(X) + ρ(Y)
 Positive homogeneity: ρ(hx) = hρ(X) for h > 0
 Translation invariance: ρ(X + n) = ρ(X) – n
© Kaplan, Inc. 25

13
 Coherent Risk Measure: Confidence Interval

 Estimators are only as useful as their precision.

 Estimators that are less precise (i.e., have large standard errors
and wide confidence intervals) will have limited practical value.
 Therefore, it is best practice to also compute the standard error
for all coherent risk measures.
 The process of estimating these standard errors is quite
complex.
© Kaplan, Inc. 26

Coherent Risk Measure: Confidence Interval

(cont.)
 Standard error of quantile (q):
p (1 − p ) / n
s e (q ) =
f (q )
p = p r o b a b ility in th e le ft ta il
f ( q ) = p r o b a b ility m a s s in b in ( w id th o f in te r v a l)
n = s a m p le s iz e
 Confidence interval:
 q + s e ( q ) × z α  > V a R >  q – s e ( q ) × z α 
© Kaplan, Inc. 27

14
 Quantile-Quantile (QQ) Plot

 The QQ plot is a way to visually examine if empirical data fits the

theoretical distribution (e.g., the normal distribution).
 The process graphs the quantiles at regular confidence intervals
for the empirical distribution against the theoretical distribution.
 As an example, if the middles of the QQ plot match up but the
tails do not, then the empirical distribution can be interpreted as
symmetric with tails that differ from the theoretical distribution
(either fatter or thinner).

© Kaplan, Inc. 28

Quantile-Quantile (QQ) Plot (cont.)

© Kaplan, Inc. 29

15
 GARP 2016 Market Risk Q1
An analyst is examining a sample of return data. As a first step, the analyst construct a QQ plot
of the data as shown below:

Based on an examination of the QQ plot, which of the following statements is correct?

A. The returns are normally distributed.

B. The return distribution has thin tails relative to the normal distribution.
C. The return distribution is negatively skewed relative to the normal distribution.
D. The return distribution has fat tails relative to the normal distribution.
© Kaplan, Inc. 30

GARP 2016 Q1—Answer

Correct answer: D

This QQ plot has steeper slopes at the tails of the plot, which indicate fat tails in the
distribution. A normal distribution would result in a linear QQ plot. A distribution with thin tails
would produce a QQ plot with less steep slopes at the tails of the plot than a linear relationship,
while this one is steeper at the tails. It is not a negatively skewed distribution, as the QQ plot is
symmetric.

© Kaplan, Inc. 31

16
 Example: Value at Risk (VaR)
 Gregory Chambers is interested in estimating the daily VaR (with
99% probability) of a bank’s fixed income portfolio, currently
valued at $30 million.
 The portfolio had the following returns over the past 200 days
(ranked from high to low):
1.9%, 1.87%, 1.85%, 1.79%,....
–1.78%, –1.81%, –1.84%, –1.87%, –1.91%
 What will be the VaR estimate using the historical method?

© Kaplan, Inc. 32

Answer

 Since there are 200 trading days, the 99% confidence level
would occur at the second-lowest return (–1.87%):
 VaR = (–0.0187)(30,000,000) = –$561,000

 Therefore, the 1% daily VaR is $561,000.

© Kaplan, Inc. 33

17
 Sample Exam Question
Large Bank uses the traditional historical simulation method for calculating
VaR. Over the past year (i.e., 250 trading days), the seven worst returns
are as follows: –0.5%, –0.4%, –0.6%, –1.6%, –0.8%, –1.0%, and –1.7%.
Based on a $100 million portfolio, what is the 98% VaR?
a. $600,000
b. $800,000
c. $1,200,000
d. $1,700,000

© Kaplan, Inc. 34

Sample Exam Question – Answer

Answer: a
The VaR observation can be identified as α × n:
(1 – 98%) × 250 = 5
The 5th lowest return corresponds to –0.6%.

© Kaplan, Inc. 35

18
 GARP FRM 2013 Practice Exam (Q2)
The annual mean and volatility of a portfolio are 10% and 40%. The
current value of the portfolio is GBP 100,000. How does the 1-year 95%
VaR calculated using a normal distribution (normal VaR) compare with
the 1-year 95% VaR calculated using the lognormal distribution
assumption (lognormal VaR)?
a. Lognormal VaR is greater than normal VaR by GBP 13,040
b. Lognormal VaR is greater than normal VaR by GBP 17,590
c. Lognormal VaR is less than normal VaR by GBP 13,040
d. Lognormal VaR is less than normal VaR by GBP 17,590
© Kaplan, Inc. 36

GARP FRM 2013 Practice Exam (Q2) – Answer

95% normal VaR 1-year:
VaR95% = μ – ( z – value × σ )  × portfolio value

VaR95% = 0.1 – (1.645 × 0.40)  ×100,000 = 55,800

95% lognormal VaR 1-year:

Difference of 13,035
VaR95% = 1– eμ– ( z-value × σ )  ×portfolio value

VaR95% = 1– e (
0.1– 1.645 × 0.40)
 ×100,000 = 42,765
 Note that lognormal VaR will always
be lower than normal VaR.
© Kaplan, Inc. 37

19
 GARP 2013 FRM Practice Exam (Q18)
A portfolio has USD 2 million invested in Stock A and USD 1 million invested in
Stock B. The 95% 1-day VaR for each individual position is USD 40,000. The
correlation between the returns of Stock A and Stock B is 0.5. While
rebalancing, the portfolio manager decides to sell USD 1 million of Stock A to
buy USD 1 million of Stock B. What effect will this have on the 95% 1-day
portfolio VaR?
a. There will be no effect
b. It will increase by USD 20,370
c. It will increase by USD 21,370
d. It will increase by USD 22,370

© Kaplan, Inc. 38

GARP 2013 FRM Practice Exam (Q18) – Answer

VaR initial weightings:

VaRportfolio = 40,0002 + 40,000 2 + 2 × 40,000 × 40,000 × 0.5 = USD 69,282

VaR amended weightings: Difference of $22,370

VaRportfolio = 20,0002 + 80,0002 + 2 × 20,000 × 80,000 × 0.5 = USD91,652

© Kaplan, Inc. 39

20