Académique Documents
Professionnel Documents
Culture Documents
© EduPristine – www.edupristine.com
© EduPristine For VaR-I (2016)
VaR Methods
Risk can be broadly defined as the degree of uncertainty about future net returns
• Credit risk relates to the potential loss due to the inability of a counterpart to meet its obligation
• Operational risk takes into account the errors that can be made in instructing payments or settling
transactions
• Liquidity risk is caused by an unexpected large and stressful negative cash flow over a short period
• Market risk estimates the uncertainty of future earnings, due to the changes in market conditions
Broadly the standard deviation of the variable measures the degree of risk inherent in the variable
Say the standard deviation of returns from the assets owned by you is 50% and the standard
deviation of returns from assets I own is 0%. We can say that risk of my assets is zero
Say the 95% daily VAR of your assets is $120, then it means that out of those 100 days there would
be 95 days when your daily loss would be less than $120. This implies that during 5 days you may
lose more than $120 daily.
0.15
0.1
0.05 Z values
0
-4 -3 -2 -1 0 1 2 3 4
Mean = 0
The colored area of the normal curve constitutes 5% of the total area under the curve.
There is 5% probability that the losses will lie in the colored area i.e. more than the VAR number.
0.45
0.4
0.35
0.3
0.25
0.2
VAR X % (in %) Z X % *
0.15
0.1
0.05
0
-4 -2 0 2 4
Mean = 0
ZX% : the normal distribution value for the given probability (x%) (normal distribution has mean as
0 and standard deviation as 1)
σ : standard deviation (volatility) of the asset (or portfolio)
VAR in absolute terms is given as the product of VAR in % and Asset Value:
VAR VARX % (in %) * AssetValue
This can also be written as:
VAR Z X % * * AssetValue
This comes from the known fact that the n-period volatility equals 1-period volatility multiplied by
the square root of number of periods(n).
As the volatility of the portfolio can be calculated from the following expression:
The above written expression can also be extended to the calculation of VAR:
VaR portfolio (in %) wa2 (%VAR a ) 2 w 2b (%VAR b ) 2 2wa w b * (%VAR a ) * (%VAR b ) * ab
VaR
Linear: When the value of the delta is constant for any change in the underlying
• Primarily in the case of forwards and futures we have linear assets
• The method to calculate VAR for linear assets is called Delta Normal method
• Delta Normal method assumes that the variables are normally distributed
Non Linear: When the value of the delta keeps on changing with the change in the underlying
asset
• Options are non-linear assets, where delta-normal method cannot be used as they assume the linear payoff
of the assets
• To calculate the VAR for non-linear assets, full revaluation of the portfolio needs to be done
• Monte Carlo methods or Historical Simulation are commonly used to fully revaluate the portfolio
linear
Payoff
0
k Share/asset price
Long position
Delta of Derivative: Change in price of Derivative to change in underlying asset
For example:
• The permitted lot size of S&P CNX Nifty futures contracts is 200 and multiples thereof.
• So VAR of Nifty Futures contract is 200 * VAR of Nifty Index
If the daily VaR at 5%of Nikkei is USD 0.8 mn and you have 100 lots of Nikkei contract.
Calculate annual VaR at 95% confidence for your portfolio assuming 250 days?
Here, delta = 100, because for every 1 unit change in the Index Nikkei, the futures price will change
by 100 units because the lot size is 100
The fat-tailed unconditional distribution can be broken down into two conditional distributions,
either with similar means and different variances or similar variances and different means.
Many a times when we observe marked differences between the estimated and actual volatilities,
it’s a result of regime switching which means that the average volatility in the market has now
changed too much when compared to the previous estimate
1 n
(R i E ( R))2
2
n i 1 Actual volatility
30bp/day
18bp/day
Low Vol Estimate
Unconditional
1 n
volatility (R i E ( R))2
2
n i 1
10bp/day
x
n
2
i x
Variance i 1
n 1
• Mean of returns (x-bar) is usually zero, especially if returns are over short-time period (say, daily returns). In
that case, variance estimate for next day is nothing but simple average (equally weighted average) of
previous ‘n’ days’ squared returns.
n
ix 2
Variance i 1
n 1
What if the volatility is dependent on the values of volatility observed in the recent past?
What if they also depend on the latest returns?
m i 1 2
2
n (1 ) u n i m n2 m
i 1
• Variance estimate for next day (n) is given by (1-λ) weight to recent squared return and λ weight to the
previous variance estimate.
• Risk-metrics (by JP Morgan) assumes a Lambda of 0.94.
In this equation, variance for time ‘t’ was also an estimate. So we can substitute for it as follows:
Example 1: On Tuesday, return on a stock was 4%. Volatility (Std. deviation) estimate for Tuesday
was 1%. Find volatility estimate for Wednesday using λ of 0.94
Example 2: Continuing the previous example, volatility estimate for Wednesday was 1.378%.
Assume that actual return on Wednesday was 0%. What is the variance estimate for Thursday?
Using a daily RiskMetrics EWMA model with a decay factor λ = 0.95 to develop a forecast of the
conditional variance, which weight will be applied to the return that is 4 days old?
A. 0.000
B. 0.043
C. 0.048
D. 0.950
B.
• A. Incorrect. The wrong factor has been squared. The EWMA RiskMetrics model is defined as:
ht = λ*ht-1 + (1- λ)* r2t-1. For t = 4, and processing r0 through the equation three times produces a factor of (1-
0.95)3*0.95 = 0.00012 for r0 when t = 4.
• B. Correct. The EWMA RiskMetrics model is defined as:
ht = λ*ht-1 + (1- λ)* r2t-1. For t = 4, and processing r0 through the equation three times produces a factor of (1-
0.95)*0.95^3 = 0.043 for r0 when t = 4.
• C. Incorrect. The 0.95 has not been squared. The EWMA RiskMetrics model is defined as:
ht = λ*ht-1 + (1- λ)* r2t-1. For t = 4, and processing r0 through the equation three times produces a factor of (1-
0.95)*0.95 = 0.048 for r0 when t = 4.
• D. Incorrect. The weight is not simply λ. The EWMA RiskMetrics model is defined as:
ht = λ*ht-1 + (1- λ)* r2t-1. For t = 4, and processing r0 through the equation three times produces a factor of 0.95 =
0.950 for r0 when t = 4.
2
t 1 VL t
2
t
2
2
t 1 t
2
t
2
VL
1
Solution: In the GARCH model, 12% is the weight given to latest squared return (reactive factor).
85% is the weight given to latest variance estimate (persistence factor). Therefore,
1-0.12-0.85 = 3% is weight given to long-term average Volatility.
• Therefore, 3%*VL = 0.000005 i.e. VL = 0.017%
• Also, variance estimate for t+1 = .000005 + 0.12*(-1%)^2 + 0.85*(1.88%)^2 = 0.0317%
• Volatility (Std. Dev.) estimate for t+1 = sqrt (0.0317%) = 1.782%
• For a stable GARCH model, alpha + Beta <=1. If alpha + Beta>1, then weight given to long-term volatility is
negative and the model becomes ‘mean-fleeing’
B.
• A. Incorrect. The model that will take the shortest time to revert to its mean is the model with the lowest
persistence defined by α1 + β. In this case the persistence factor is the second largest:
α1 + β = 0.03 + 0.96 = 0.99.
• B. Correct. The model that will take the shortest time to revert to its mean is the model with the lowest
persistence defined by α1 + β. In this case the persistence factor is the second lowest:
α1 + β = 0.02 + 0.95 = 0.97.
• C. Incorrect. The model that will take the shortest time to revert to its mean is the model with the lowest
persistence defined by α1 + β. In this case the persistence factor is the largest:
α1 + β = 0.01 + 0.97 = 0.98.
• D. Incorrect. The model that will take the shortest time to revert to its mean is the model with the lowest
persistence defined by α1 + β. In this case the persistence factor is the lowest:
α1 + β = 0.01 + 0.98 = 0.99.
n2 0.000002 013
. un21 0.86 n21
Suppose that the current estimate of the volatility is 1.6% per day and the most recent percentage
change in the market variable is 1%. What is the new variance rate?
Ans -2733814418
HS Ans -2.35
Third Approach
• Aggregate the simulated returns and then apply the parametric approach to aggregated portfolio
Truly Predictive in nature and is not a historical simulation but it assumes that Black Scholes
(or any other model) is correct model for calculating market Value of Option
Full Valuation method is the process of measurement of risk of a portfolio by fully re-pricing it
under a set of scenarios over a time period. It can be used to cover a large range of values of the
portfolio returns in order to provide more accurate results. It generally provides more accurate
results compared to delta normal approach but is a complicated process.
Two popular methods under full revaluation approach have been explained in the subsequent
slides.
Payoff
Gains !
Price
Risk !
Disadvantages:
• The calculation of Monte Carlo VaR can take 1,000 times longer than Parametric VaR because the potential
price of the portfolio has to be calculated thousands of times.
• Unlike Historical VaR, it typically requires the assumption that the risk factors have a Normal distribution.
The area under the normal curve for confidence value is:
0.005 Mean = 0
+- stdev
H = 100
Z < -2.33 ( 99% VaR ) 1 out of 100
Distribution of WCS mean = -2.51 and it’s 1st and 5th percentile are -3.1 and -3.9 resp.
What does this mean to a financial risk manager?
Limitations of VaR
• Tells that n number of times in 100 days, the loss is not going to exceed N$
• But it cannot predict the loss when it exceeds!
• Does not focus on large losses (Tails of distribution)
Stress Testing:
• Supplement to VaR
• "VaR should always be supplemented with stress testing" has been one of the recommendations
of the supervisor
• Testing how well a portfolio performs under some of the most extreme market moves seen in the last 10 to
20 years
Advantages:
• Can take a large number of risk factors into consideration
• Can specifically focus on the tails (extreme losses)
Disadvantages:
• Highly subjective and can become overcautious
• Requires complete top management support
Capital Allocation
Exposure
Ride out Turmoil
The one significant shortcoming of VaR that stress testing does address is sudden changes in
historical correlations
If two currencies have been pegged to one another, they will exhibit a high historical correlation. A
VaR analysis based on that historical correlation will not address the risk that one of the currencies
may be devalued relative to the other. If this is a scenario that concerns management, a simple
stress test will offer more insights than would, say, a VaR analysis performed with a modified
correlation assumption
Goals
• Identify scenarios that would not occur under standard VAR models
• Simulating shocks that have never occurred
• Simulating shocks that reflect permanent structural breaks or temporally changed statistical patterns
Factor push method shifts each risk factor in the direction that would have an adverse impact on
the portfolio.
Conditional Scenario Method incorporates the correlation between various key risk factors .
Stress tests can be improved by adopting one or more of the following methods:
Buy protection through insurance or other derivative products
Change portfolio composition to decrease the exposure or diversify
Change business strategy to suit the changing business environment
Develop back up plans for unforeseen events
Secure alternative funding in stressed environment
Implied Volatility 2
n
(R
i 1
i E ( R )) 2 2n 2n 1 (1 ) u n21 90% 1.28
Regime Switching 95% 1.65
The worst case scenario
GARCH (1,1): 97.5% 1.96
Stress Testing
VAR Z X % * * AssetValue
t21 VL t2 t2 99% 2.32
VaR (n days) (in %) VaR (daily VaR) (in %) * n VARLinear Derivative *VARUnderlying Risk Factor
VaR portfolio (in %) wa2 (%VAR a ) 2 w 2b (%VAR b ) 2 2wa w b * (%VAR a ) * (%VAR b ) * ab
support@edupristine.com
www.edupristine.com
© EduPristine – www.edupristine.com
© EduPristine For VaR-I (2016)