LECTURE NOTES

Mathematical Modeling and its Application in Finance

This file is copyright protected. You can make hardcopies for your
yo ur own use
during the semester and in the context of the course. If you would
wou ld like to use
the material for other purposes or make softcopies of this Powerpoint
presentation you are hereby granted permission to do so, provided
provide d that each
and every slide you copy caries the copyright information on the lower left
corner.

© Stavros A. Zenios, 5/18/01

1
 Lecture 3:
RiskMetrics and Value-at-Risk

Stavros A. Zenios
Operations and Information Management Department
The Wharton School
University of Pennsylvania

© Stavros A. Zenios, 5/18/01

2
 OUTLINE

n Overview of RiskMetrics
n Value-at-Risk (VaR) methodologies
n Monte Carlo simulations
• Reading:
Bulkpack, items no. 7 and 8.
• Further reading:
Introduction to RiskMetrics, Morgan Guarantee Trust Company, Fourth
edition.

RiskMetrics-Technical Document, Morgan Guarantee Trust Company,

Fourth edition.
– General descriptionin of the technical document, pp
pp.. I-
I-vi,
– Part I: Risk Measurement Framework
– Chapter 6. Market risk methodology
– Appendix E: Routines to simulate correlated normal random variables,
variables,
© Stavros A. Zenios, 5/18/01
3
 I. Overview of RiskMetrics™

n Tools for assessing risk exposure of a position (in global market

risks) using VaR (Value-at-Risk)
n Set of risk measurement methodologies
n Data sets of volatility and correlation data used in computing
market risk
• Daily re-estimated
• 1-day horizon, 1-month horizon, according to BIS
requirements (three data sets)
• Each data set contains information on 400 instruments: FX,
money markets, bonds, equity indices in 23 countries,
commodities

© Stavros A. Zenios, 5/18/01

4
 n Value-at-Risk (VaR): There is a
probability of x% that the portfolio will
suffer a loss greater than VaR during
the planning period.
• Typically x is taken to be 1% or 5%
• Planning period is 1 -day (for active
trading), or 1 -month (for portfolio
management), or the 10-day holding period
specified in the BIS directives.
n Uses historical return data to calculate
volatilities and correlations

© Stavros A. Zenios, 5/18/01

5
 I. Overview of RiskMetrics™:
Practical use of VaR information

n Management information on the risk exposure of trading and

investment operations of the institution.
n Setting limits on positions and resource allocation.
n Performance assessment: efficiency with which market risk
(measured by VaR) is translated to high revenues and low
realized volatility of revenues.
n Regulatory reporting: capital adequacy requirements based on
market risk exposure.
• European Union directive (effective January 1996) requires banks
and investment firms to set aside capital to cover market risks
• Bank of International Settlement:
– allows internal models for measuring market risk but imposes stringent
stringent
requirements on some of the risk factors.
– 10
10--day risk horizon
– multiplier of VAR estimate to determine capital adequacy margins

© Stavros A. Zenios, 5/18/01

6
 I. Overview of RiskMetrics:
Why is accurate measurement of VaR important?

n Underestimating VaR: Regulatory

pressures, cost of Federal or Central
Bank guarantees etc.
n Overestimating VaR: Paying too much
for insurance.
• Example: State of California drivers’ insurance requirements:
– Post $30,000 bond with the State, e.g. earning 5% if not used.
– Buy an insurance policy. E.g., Pay $2,500 per year with an actuarial
actuarial value of $2,000, and
therefore costs $500 per year if not used.

© Stavros A. Zenios, 5/18/01

7
 ii. Value-at-Risk Methodologies

n Calculate the VaR due to holdings in a financial instrument in

RiskMetrics dataset .

n Map cashflows of target position to RiskMetrics instruments and

calculate VaR of target position:
• Linear relations?
• Nonlinear relations?
n Delta and Delta-Gamma Approximations or Monte Carlo simulations

© Stavros A. Zenios, 5/18/01

8
 Example 1: Single position VaR calculation

USD -based corporation holding DEM 140 million FX position. What is your
USD- y our VaR over a 1-
1-
day horizon at the 5% probability level, I.e., what is the least amount you could loose in
the next day with probability less 5%. (Once every 20 days you may m ay loose this money!)

Assume FX rate is today 1.40 DEM/USD and daily standard deviation

deviatio n of this rate has,
historically, been σ = 0.565%
Assuming normality of standardized returns rt / σ t then the DEM/USD exchange
then
rate is not expected to drop by more than

1.65σ = 0.932% , 95% of the time.

Hence, FX risk of our position is $100 million x 0.932% = $932,000.

$932,000.

RiskMetrics provide users with the VaR statistic 1.65σ .

© Stavros A. Zenios, 5/18/01

9
 Example 2: Two position VaR calculation

USD -based corporation holding DEM 140 million position in 10-

USD- 10-year German
governemnt bonds. What is your VaR over a 1- 1 -day horizon at the 5% probability level,
I.e., what is the least amount you could loose in the next day with
w ith probability less 5%.
(Once every 20 days you may loose this money!)

Risk exposure: Still USD 100, but exposed to interest rate risk on the bund and FX risk.

σ FX = 0.565% σ10−b = 0.605%

FX risk= $100 x 1.65 x 0.565% = $932,000
Interest rate risk=$100 x 1.65 x 0.605% = $999,000.

Correlation of 10-
10-year German govt bond with DEM/USD exchage rates:

ρFX ,10−b = −0.27

Hence,
VaR = (.999 2
+ .9322 + 2 ρ FX ,10 −b × 0.999 × 0.932)
= $1.168 million.
© Stavros A. Zenios, 5/18/01
10
 Example 3: VaR calculation on a portfolio of cashflows

Consider an instrument that gives rise to three USD 100 cashflows each occuring at the end of
1, 4 and 7 months.

RiskMetrics provides volatility and correlation estimates of risk-

risk -free interest rates at maturities

1m 3m 6m 1yr 2yr 3yr 5yr 7yr 9yr 10yr 15yr

15yr 15yr 20yr 30yr

RiskMetrices vertices (or RiskMetrics cashflows)

cashflows )

Portfolio 100 100 100

1m 4m 7m

Cashflow mapping 100 60 40+70 30

1m 3m 6m 1yr

RiskMetrics cashflow 100 60 110 30

rp = .33r1m + .20r3 m + .37r6 m + .10r1 yr

© Stavros A. Zenios, 5/18/01
11
 General VaR calculation on a portfolio of cashflows

Portfolio return is a weighted linear combination of RiskMetrics returns:

M
rp = ∑ wj rj
j =1

The VaR of the portfolio is given by VaR = VQV T where

V = (w1 ⋅1.65σ1, w2 ⋅1.65σ 2 ,.....,wM ⋅1.65σ M )

and Q is the correlation matrix

1 ρ 3m,1m ρ 6 m,1m ρ1 yr ,1m 

 
 ρ1m, 3m 1 ρ 6 m,3m ρ1yr , 3m 
Q=
ρ ρ 3m, 6 m 1 ρ1 yr , 6m 
 1m, 6 m 
 ρ1m,1 yr ρ 3m,1 yr ρ 6 m,1 yr 1 

© Stavros A. Zenios, 5/18/01

12
 VaR calculations for nonlinear positions

In Example 1 our position (in FX) varies linearly with the DEM/USD
exchange rate. In Example 2 our position varies as a linear comb ination of
changes in the 10-year bund and the DEM/USD exchange rate.

What to do if position varies varies nonlinearly with changes in the underlying

rate.
Position Full valuation
value Delta+Gamma

Delta

DEM/USD
1.40

© Stavros A. Zenios, 5/18/01

13
 Delta and Gamma approximations

∂P ( r ) 1 ∂ P( r )
2
P( r1 , r ) = r1 + P( r0 ) + ( r − r0 ) + ( r − r ) 2

∂r 2 ∂r 2 0

1
∆P = ∆r1 + δ∆r + γ∆r 2

2
For a linear approximation (delta approximation):

VaR ( P) = 1.65 (σ + δ σ + 2δσ 1σ r ρ1, r )

2
1
2 2
r

All information is given by RiskMetrics and the delta of the security.

© Stavros A. Zenios, 5/18/01

14
 For a quadratic approximation (delta-gamma): RiskMetrics sigma’s are not
quite applicable, since the quadratic term introduces skewness in the
distribution. (Negative returns are squared.)

© Stavros A. Zenios, 5/18/01

15
 Monte Carlo simulations

n Scenario generation: Using the RiskMetrics volatility and

correlation estimates generate a large number of scenarios for
the underlying assets in our portfolio. (First identify the relevant
RiskMetrics instruments and do the cashflow mapping if
needed).
n Portfolio valuation: for each scenario calculate the portfolio
value.
n Calculate VaR: Report the distribution of the portfolio returns, or
calculate the value that can be lost with a certain probability
(VaR).

© Stavros A. Zenios, 5/18/01

16
 Mapping cashflows onto RiskMetrics vertices

Split the portfolio cashflows among the nearest RiskMetrics vertices in such
a proportion that:

Market value is preserved: total market value of the RiskMetrics cashflows

equals the market value of the original cashflow
Market risk is preserved: total market risk of the RiskMetrics cashflows
equals the market risk of the original cashflow
Sign is preserved: RiskMetrics cashflows have same sign as original cashflow

© Stavros A. Zenios, 5/18/01

17
 Monte Carlo simulation of correlated normal
random variables

Assume a covariance matrix Q of correlated normal random variables.

1. Let Q = P T P

2. Compute a vector of standard normal random variables (e), I.e., the

covariance matrix of e is the identity I.

3. Compute y = PT e

The random vector y has a multivariate normal distribution

with covariance matrix Q.

© Stavros A. Zenios, 5/18/01

18