The variance-covariance method for VaR calculation

https://www.simtrade.fr/blog_simtrade/variance-
covariance-method-var-calculation/
Posted on December 17, 2021 by Jayati WALIA

In this article, Jayati WALIA (ESSEC Business School, Grande

Ecole – Master in Management, 2019-2022) presents the
variance-covariance method for VaR calculation.
Introduction
VaR is typically defined as the maximum loss which should not be
exceeded during a specific time period with a given probability
level (or ‘confidence level’). VaR is used extensively to determine
the level of risk exposure of an investment, portfolio or firm and
calculate the extent of potential losses. Thus, VaR attempts to
measure the risk of unexpected changes in prices (or return rates)
within a given period.

The two key elements of VaR are a fixed period of time (say one or
ten days) over which risk is assessed and a confidence level which
is essentially the probability of the occurrence of loss-causing
event (say 95% or 99%). There are various methods used to
compute the VaR. In this post, we discuss in detail the variance-covariance method
for computing value at risk which is a parametric method of VaR calculation.

Assumptions
The variance-covariance method uses the variances and covariances of assets for VaR
calculation and is hence a parametric method as it depends on the parameters of the
probability distribution of price changes or returns.

The variance-covariance method assumes that asset returns are normally distributed
around the mean of the bell-shaped probability distribution. Assets may have
tendency to move up and down together or against each other. This method assumes
that the standard deviation of asset returns and the correlations between asset
returns are constant over time.

VaR for single asset

VaR calculation for a single asset is straightforward. From the distribution of returns
calculated from daily price series, the standard deviation (σ) under a certain time
horizon is estimated. The daily VaR is simply a function of the standard deviation
and the desired confidence level and can be expressed as:
Where the parameter ɑ links the quantile of the normal distribution and the standard
deviation: ɑ = 2.33 for p = 99% and ɑ = 1.645 for p = 90%.
In practice, the variance (and then the standard deviation) is estimated from
historical data.

Where R  is the return on period [t-1, t] and R the average return.
t

Figure 1. Normal distribution for VaR for CAC40

Source: computation by the author (data source: Bloomberg).

You can download below the Excel file for the VaR calculation with the variance-
covariance method. The two parameters of the normal distribution (the mean and
standard deviation) are estimated with historical data from the CAC 40 index.
VaR for a portfolio of assets
Consider a portfolio P with N assets. The first step is to compute the variance-
covariance matrix. The variance of returns for asset X can be expressed as:

To measure how assets vary with each other, we calculate the covariance. The
covariance between returns of two assets X and Y can be expressed as:

Where X  and Y  are returns for asset X and Y on period [t-1, t].
t t

Next, we compute the correlation coefficients as:

We calculation the standard deviation of portfolio P with the following formula:

Where w  corresponds to portfolio weights of asset i.

i

Now we can estimate the VaR of our portfolio as:

Where the parameter ɑ links the quantile of the normal distribution and the standard
deviation: ɑ = 2.33 for p = 99% and ɑ = 1.65 for p = 95%.
Advantages and limitations of the variance-covariance
method
Investors can estimate the probable loss value of their portfolios for different holding
time periods and confidence levels. The variance–covariance approach helps us
measure portfolio risk if returns are assumed to be distributed normally. However,
the assumptions of return normality and constant covariances and correlations
between assets in the portfolio may not hold true in real life.

Useful resources
Jorion P. (2007) Value at Risk, Third Edition, Chapter 10 – VaR Methods, 274-276.