Risk Modelling -Value at

Risk and Expected Shortfall

Value at Risk (VaR)
• One simple risk definition used widely throughout the financial
industry is known as Value-at-Risk or VaR
• VaR was pioneered by JPMorgan
• The question being asked in VaR is: “What loss level is such that we
are X% confident it will not be exceeded in N business days?"
VaR and Regulatory Capital
• Regulators have traditionally used VaR to calculate the capital they
require banks to keep
• The market-risk capital has been based on a 10-day VaR estimated
where the confidence level is 99%
• Credit risk and operational risk capital are based on a one-year 99.9%
VaR
• VaR is also widely used by pension plans, mutual funds and hedge
funds
• For pension and mutual funds VaR is calculated over longer time
intervals such as months rather than days
• 95% VaR is 25000
VaR
• Value at Risk (VaR)
• Loss corresponding to a very low percentile of the entire return distribution, such as
the fifth or first percentile return.
• Value at risk (VaR) is a measure of the risk of investments. It estimates how
much a set of investments might lose, given normal market conditions, in a
set time period such as a day. VaR is typically used by firms and regulators in
the financial industry to gauge the amount of assets needed to cover
possible losses.
• For a given portfolio, time horizon, and probability p, the p VaR can be
defined informally as the maximum possible loss during the time if we
exclude worse outcomes whose probability is less than p. This assumes
mark-to-market pricing, and no trading in the portfolio
VaR
• For example, if a portfolio of stocks has a one-day 95% VaR of $1
million, that means that there is a 0.05 probability that the portfolio
will fall in value by more than $1 million over a one-day period if there
is no trading. A loss which exceeds the VaR threshold is termed a "VaR
break“.
Methods of Calculating VaR
• Historical Method
• The historical method simply re-organizes actual historical returns,
putting them in order from worst to best. It then assumes that history
will repeat itself, from a risk perspective.
• With 95% confidence, we expect that our worst daily loss will not
exceed (VaR95%).
• If we invest $100, we are 95% confidant that our worst daily loss will
not exceed 4$ (100*VaR95%).
Methods of Calculating VAR
• The Variance-Covariance Method: This method assumes that stock
returns are normally distributed. In other words, it requires that we
estimate only two factors - an expected (or average) return and
a standard deviation - which allow us to plot a normal
distribution curve.
• We will use the norm.inv function in excel to generate the results.
• Suppose that the loss distribution is Gaussian with mean µ and
standard deviation σ. Then is
• The historical 1-year loss distribution of a portfolio of loans in €
million is well approximated by a N(10,5).
What is the 95% VaR? And the 98%?
Reference
• In Excel you can generate a draw from an distribution using the
command: NORM.INV(RAND()).
Advantages of VaR
• It captures an important aspect of risk in a single number
• It is easy to understand
• It asks the simple question: "How bad can things get?"
ES: Expected Shortfall
• Expected Shortfall (ES)
• Expected shortfall (ES) is a risk measure—a concept used in the field of
financial risk measurement to evaluate the market risk or credit risk of a
portfolio. The "expected shortfall at q% level" is the expected return on
the portfolio in the worst q% of cases. ES is an alternative to value at
risk that is more sensitive to the shape of the tail of the loss
distribution.
• Expected shortfall is also called conditional value at risk (CVaR), average
value at risk (AVaR), and expected tail loss (ETL), and conditional tail
expectation (CTE), focuses on the expected loss in the worst-case
scenario (left tail of the distribution)
• Expected shortfall, aka conditional value at risk, answers to the
question “If things go bad, what is the expected loss?”
• It is a measure of risk with many interesting properties
• From a statistical point of view, the expected shortfall is a sort of
mean excess function, i.e. the average value of all the values
exceeding a special threshold, the VaR
Why is ES important?
Problem -A portfolio of loans may lead to the losses in the table;
What is the 95% and 99% Expected Shortfall
VaR from Simulation
• Many times you won't have an analytic expression for the percentiles of a
distribution. This may happen when you wish to use more accurate and
sophisticated asset models (which we will cover later in the course).
• In that case you may need to simulate a large portfolio of assets (which we
will cover in the next slide deck).
• Simulate your portfolio a large number of times N. Rank the simulated
future portfolio profits from best to worst. The Var at the X% confidence is
then the ranked simulation number XN.
• E.g. If you simulated 1,000,000 future profit scenarios and ranked them
from best to worst, the VaR at 95% confidence can be estimated as the
value of the 950,000th best profit.
Modelling Asset Price Dynamics -
Lognormal Model
• One of the most commonly used models for the dynamics of stock prices, forward/futures prices
and forward rates is the lognormal model
• Assume the price of the asset (stock, forward, futures, etc.) at time be
• Assume the average annualized expected return measured with continuous compounding for the
asset between times and is:
• Assume that the average standard deviation of the asset between times and is:
• Then under the lognormal model

where is a draw from the normal distribution with mean zero and variance 1. This formula
is important!
• This formula allows us to simulate possible future outcomes for the asset at future time between
given the information at .
• In Excel you can generate a draw from an distribution using the command: NORM.S.INV(RAND()).
Simulating Multiple-Correlated Random
Variables
• In many cases we have many different assets or risk factors of interest
that we would like to simulate jointly.
• When we wish to simulate multiple asset price movements we need to
have a different random draw for each asset in each time interval.
• However assets are often correlated with one another in which case we
need to know the co-efficient of correlation between each asset and we
then need to know how to draw multiple normally distributed random
numbers with those exact correlations.
• Many software packages have built in functions for generating normally
distributed random variables. However none allow you to generate
correlated random draws.
Simulating Multiple-Correlated Random
Variables
• Suppose that we wish to generate K correlated random variables that
have a correlation matrix given by
• Suppose further that we can generate K uncorrelated, normally
distributed random variables and we put them in a column matrix .
• Then the idea will be to find a linear transformation of these
uncorrelated variables that will result in the correlated random draws
we are looking for. In other words we wish to find a square matrix M
so that if: 𝜙=𝑀 𝜖

then the column matrix will have a correlation structure given by .

• There are many ways of factoring a matrix so that

one of which is called the Cholesky Factorization.

• VB code for generating a Cholesky Factorization M from a correlation matrix will be
given to you so that a user-defined function in Excel can be created that will factorize
correlation matrices as needed (so long as the matrix is in fact a valid correlation matrix).
• Once a factorization can be found the calculation

can be done using the excel command mmult(matrix 1, matrix 2).

Stepwise explanation of construction of
Generating correlated random variables using
a linear transformation.
• Uncorrelated Random Variables (𝜖 ):
• We start with K uncorrelated, normally distributed random variables, denoted
as 𝜖_1, 𝜖_2, …, 𝜖_𝐾.
• These variables have zero correlation with each other, meaning that their
covariance matrix is diagonal (off-diagonal elements are zero).
• Correlation Matrix (Σ):
• We have a desired correlation matrix Σ that specifies the desired correlations
between the 𝜙 variables (𝜙_1, 𝜙_2, …, 𝜙_𝐾).
• Σ is a symmetric positive definite matrix that captures the pairwise
correlations.
• Linear Transformation (Matrix M):
• Our goal is to find a square matrix M such that when we multiply it with the
column matrix 𝝐 (containing 𝜖 variables), we obtain the correlated random
variables 𝝓 (containing 𝜙 variables).
• Mathematically, this can be expressed as:

• Here, 𝝓 is the column matrix of correlated random variables, and 𝝐 is the

column matrix of uncorrelated random variables.
• Correlation Structure:
• If we choose the matrix M correctly, the resulting 𝝓 variables will have the
desired correlation structure given by Σ.
• In other words, the covariance matrix of 𝝓 will match Σ.
Excel Implementations
• You will be provided with a spreadsheet with a user defined VB function
called: "cholesky" which will calculate the Cholesky factorization for any valid
correlation matrix you give it.
• However the formula: assumes that the resulting matrix is a row matrix. This
is a problem for Excel in that Excel only has a limited number of columns and
often you will want to do tens of thousands of simulations for a given
problem. For this reason it is better to generate the uncorrelated random
numbers in which the rows correspond to the number of simulations and the
columns correspond to the number of assets.
• In this case the new formula becomes: the command now becomes:
mmult(epsilon, transpose(BigM)) where BigM is the Cholesky factorization of
the correlation matrix.
In-class Assignment
• Assume that you have invested 5 dollars equally in 5 different assets and
you'd like to know the VaR and CVaR of this portfolio at 95% confidence 1-
year into the future.
• To do this you plan to generate 10,000 different possible future outcomes
simulating all of the 5 assets 1-year forward. You can then rank the resulting
portfolio profits/losses and estimate your risk statistics from these
simulations.
• The expected returns for the next year for each stock (measured with
continuous compounding) are given by: 0.05, 0.07, 0.1, 0.12 and 0.15
respectively.
• The standard deviations are given by: 0.1, 0.15, 0.2,0.25 and 0.35 respectively
In-class Assignment
• Each of the stocks is correlated with the others and the correlation
matrix is given by:
Correlation
Matrix
1 -0.2 -0.13 0.11 0.05
-0.2 1 0.45 -0.39 0.03
-0.13 0.45 1 -0.98 0.02
0.11 -0.39 -0.98 1 0.01
0.05 0.03 0.02 0.01 1
The Following are steps that you must do to complete the assignment.
1. Generate the Cholesky factorization for the correlation matrix.
2. Use this Cholesky matrix to transform the uncorrelated random draws
that are provided into a set of correlated random draws with the required
correlation structure. (mmult(epsilon, transpose(BigM))
3. From these values calculate the simulated asset amounts using the
lognormal formula for each asset.
4. Calculate the 1 year profit/loss for your portfolio in each simulated future.
5. Rank the profits/losses. The VaR at 95% will be the 500th worst outcome
and the CVaR will be the average of the 500 worst outcomes.