Chapter 13

Historical Simulation and

Extreme Value Theory

©2015 by Scott Warlow 1

 Historical Simulation
• Collect data on the daily movements in all market
variables.
• The first simulation trial assumes that the
percentage changes in all market variables are as
on the first day
• The second simulation trial assumes that the
percentage changes in all market variables are as
on the second day
• and so on

©2015 by Scott Warlow 2

 Historical Simulation (cont’d)
• Suppose we use n days of historical data with
today being day n
• Let vi be the value of a variable on day i (i = 1,
…, n)
• There are (n  1) simulation trials (i = 2, …, n)
• The ith trial assumes that the value of the market
variable tomorrow, i.e., on day (n + 1), is
 vi 
vn  
 vi 1 
©2015 by Scott Warlow 3
Example: Portfolio on Sept 25, 2008
(Table 13.1, page 278)

Index Amount Invested ($000s)

DJIA 4,000
FTSE 100 3,000
CAC 40 1,000
Nikkei 225 2,000
Total 10,000

©2015 by Scott Warlow 4

U.S. Dollar Equivalent of Stock Indices
(Table 13.2, page 279)

Day Date DJIA FTSE CAC 40 Nikkei

0 Aug 7, 2006 11,219.38 11,131.84 6,373.89 131.77
1 Aug 8, 2006 11,173.59 11,096.28 6,378.16 134.38
2 Aug 9, 2006 11,076.18 11,185.35 6,474.04 135.94
3 Aug 10, 2006 11,124.37 11,016.71 6,357.49 135.44
…. …… …… ….. …… ……
499 Sep 24, 2008 10,825.17 9,438.58 6,033.93 114.26
500 Sep 25, 2008 11,022.06 9,599.90 6,200.40 112.82

©2015 by Scott Warlow 5

 Scenarios (Table 13.3, page 279)
11,173.59
 11,022.06 
11,219.38

Scenario DJIA FTSE CAC Nikkei Portfolio Loss

Number Value
1 10,977.08 9,569.23 6,204.55 115.05 10,014.334 -14.334
2 10,925.97 9,676.96 6,293.60 114.13 10,027.481 -27,481
3 11,070.01 9,455.16 6,088.77 112.40 9,946.736 53,264
…. …… ….. ….. ….. ….. …..
499 10,831.43 9,383.49 6,051.94 113.85 9,857.465 142.535
500 11,222.53 9,763.97 6,371.45 111.40 10,126.439 -126.439

©2015 by Scott Warlow 6

 Losses (Table 13.4, page 281)

Scenario Number Loss ($000s)

494 477.841
339 345.435
349 282.204
329 277.041
487 253.385
227 217.974
131 205.256

One-day 99% VaR=$253,385

One-day 99% Expected Shortfall is
(477,841+345,453+282,204+277,041+253,385)/5=$327,181

©2015 by Scott Warlow 7

 Stressed VaR and Stressed ES
• Instead of basing calculations on the
movements in market variables over the last
n days, we can base calculations on
movements during a period in the past that
would have been particularly bad for the
current portfolio
• This produces measures known as “stressed
VaR” and “stressed Expected Shortfall”
©2015 by Scott Warlow 8
 Accuracy (page 282)
Suppose that x is the qth quantile of the loss
distribution when it is estimated from n observations.
The standard error of x is

1 q (1  q )
f ( x) n

where f(x) is an estimate of the probability density of

the loss at the qth quantile calculated by assuming a
probability distribution for the loss

©2015 by Scott Warlow 9

 Example 13.1 (page 283)
• We estimate the 0.01-quantile from 500 observations as
$25 million
• We estimate f(x) by approximating the actual empirical
distribution with a normal distribution mean zero and
standard deviation $10 million
• The 0.01 quantile of the approximating distribution is
NORMINV(0.01,0,10) = 23.26 and the value of f(x) is
NORMDIST(23.26,0,10,FALSE)=0.0027
• The estimate of the standard error is therefore
1 0.01 0.99
  1.67
0.0027 500

©2015 by Scott Warlow 10

 Extension 1

• Let weights assigned to observations

decline exponentially as we go back in time
• Rank observations from worst to best
• Starting at worst observation sum weights
until the required quantile is reached

©2015 by Scott Warlow 11

 Application to 4-Index Portfolio
= 0.995 (Table 13.5, page 285)
Scenario Loss ($000s) Weight Cumulative
Number Weight
494 477.841 0.00528 0.00528
339 345.435 0.00243 0.00771
349 282.204 0.00255 0.01027
329 277.041 0.00231 0.01258
487 253.385 0.00510 0.01768
227 217.974 0.00139 0.01906
131 205.256 0.00086 0.01992

One-day 99% VaR=$282,204

One day 99% ES is
[0.00528×477,841+0.00243×345,435+(0.01−0.00528 − 0.00243)×282,204]/0.01
= $400,483 ©2015 by Scott Warlow 12
 Extension 2
• Use a volatility updating scheme and adjust the
percentage change observed on day i for a market
variable for the differences between volatility on
day i and current volatility
• Value of market variable under ith scenario
becomes
vi 1  (vi  vi 1 )  n 1 /  i 
vn
vi 1

©2015 by Scott Warlow 13

Volatilities (% per Day) Estimated for Next
Day in 4-Index Example (Table 13.6, page 286)

Day Date DJIA FTSE CAC 40 Nikkei

0 Aug 7, 2006 1.11 1.42 1.40 1.38
1 Aug 8, 2006 1.08 1.38 1.36 1.43
2 Aug 9, 2006 1.07 1.35 1.36 1.41
3 Aug 10, 2006 1.04 1.36 1.39 1.37
…. …… …… …… …… ……
499 Sep 24, 2008 2.21 3.28 3.11 1.61
500 Sep 25, 2008 2.19 3.21 3.09 1.59

©2015 by Scott Warlow 14

 Volatility Adjusted Losses
(Table 13.7, page 287)
Scenario Number Loss ($000s)
131 1,082.969
494 715.512
227 687.720
98 661.221
329 602.968
339 546.540
74 492.764

One-day 99% VaR = $602,968

One-day 99% ES =
(1,082,969+715,512+687,720+661,221+602,968)/5 = $750,078
©2015 by Scott Warlow 15
 Extension 3
• Suppose there are 500 daily changes
• Calculate a 95% confidence interval for
VaR by sampling 500,000 times with
replacement from daily changes to obtain
1000 sets of changes over 500 days
• Calculate VaR for each set and determine a
confidence interval for VaR
• This is known as the bootstrap method

©2015 by Scott Warlow 16

 Computational Issues
• To avoid revaluing a complete portfolio 500 times
a delta/gamma approximation is sometimes used
• When a derivative depend on only one underlying
variable, S
1
P  S   (S ) 2
2
• With many underlying variables,
1 n
P  i 1  i Si  i , j 1  ij Si S j
n

2
©2015 by Scott Warlow 17
 Extreme Value Theory (page 289)
• Extreme value theory can be used to investigate the
properties of the right tail of the empirical distribution
of a variable x. (If we interested in the left tail we
consider the variable –x.)
• We first choose a level u somewhat in the right tail of
the distribution
• We then use Gnedenko’s result which shows that for a
wide class of distributions as u increases the
probability distribution that v lies between u and u+y
conditional that it is greater than u tends to a
generalized Pareto distribution

©2015 by Scott Warlow 18

 Generalized Pareto Distribution

• This has two parameters the shape

parameter) and the scale parameter)
• The cumulative distribution is
1 / 
  
1  1  y
  

©2015 by Scott Warlow 19

Maximum Likelihood Estimator
(Equation 13.7, page 291)

• The observations, xi, are sorted in

descending order. Suppose that there are
nu observations greater than u
• We choose  and  to maximize

nu  1   (v  u )  1/ 1 
 ln  
  
1  i
  

i 1 

©2015 by Scott Warlow 20

Using Maximum Likelihood for 4-Index
Example, u=160 (Table 13.10, page 293)
 1  (v  u )  1/  1 
Scenario Loss ($000s) Rank ln  1  i
 
     

494 477.841 1 -8.97

339 345.435 2 -7.47
349 282.204 3 -6.51
329 277.041 4 -6.42
487 253.385 5 -5.99

304 160.778 22 -3.71

Total -108.37

©2015 by Scott Warlow 21

 Tail Probabilities
• Our estimator for the cumulative
probability that the variable  is greater
than x is
1/ 
nu  x u
1  
n   


• Setting u    , we see that this corresponds

to the power law
Prob   x   K x 
©2015 by Scott Warlow 22
 Tail Probabilities (cont’d)
where
1/ 
nu    1
K   
n   

• Extreme Value Theory therefore explains

why the power law holds so widely

©2015 by Scott Warlow 23

Estimating VaR Using Extreme
Value Theory (page 292)
The estimate of VaR when the confidence level
is q is obtained by solving

1 / 
nu  VaR  u 
q  1 1   
n   
It is

  n 


VaR  u   (1  q )  1
   nu  

©2015 by Scott Warlow 24
Estimating VaR Using Extreme
Value Theory (cont’d) (page 292)

Also
VaR    u
ES 
1 

©2015 by Scott Warlow 25