MARKET RISK Nkomo D.J.
(FRM)
MARKET RISK
It is defined as the potential loss resulting from declining prices in
the financial market
It is the risk that declining prices or volatility of prices in the
financial markets will result in a loss.
It includes stochastic market risk factors: Interest Rate, FX,
Commodity and Equity
Two drivers of the market risk exposures:
Investment Position
Market Volatility (σ)
TYPES OF MARKET RISK
They are two types of market risk – absolute and relative risks
Absolute risk – focuses on the volatility of total returns while relative risk
is referred to as the tracking error because it is usually measured relative
to a benchmark index or portfolio.
Directional risks are linear risk exposures in economic or financial
variables e.g. interest rates or stock indices. Non-directional risks are
risks that have non-linear exposures or neutral exposures to changes in
economic or financial variables.
Basis risk is the risk that the price of a hedging instrument and the price
of the instrument being hedged are not perfectly correlated.
The risk of loss from changes in actual or implied volatility of market
prices is known as volatility risk.
MARKET RISK MANAGEMENT
FLOWCHART
MARKET RISK MANAGEMENT
TOOLBOX
Market Risk Factors Identification
Sensitivity Analysis
VaR for Market Risk
Stress Testing and Scenario Analysis
Economic Capital Adequacy Level
COHERENT RISK MEASURES
MEASURES OF MARKET RISK
Standard Deviation
Not stable because it is dependent on assumptions on the loss
distribution.
Not coherent because it violates the monotonicity assumption.
Simple, but not very meaningful in the risk decomposition process.
VaR (the most commonly used measure)
Not stable because it is dependent on assumptions on the loss
distribution.
Not coherent because it violates the subadditivity assumption.
(could cause problems in internal capital allocation and limit setting
for sub-portfolios)
MEASURES OF MARKET RISK
Expected Shortfall
May or may not be stable, depending on the loss distribution
Not easy to interpret, and the link to the bank’s desired target
rating is not clear
Spectral and distorted risk measures
Not intuitive not easily understood (and rarely used in p practice).
May or may not be stable depending on the loss distribution.
MARKET RISK - VAR
Pioneered by major US banks in the ’90s as derivatives developed
Defined as the predicted worst-case loss at a specific confidence
level over a certain period of time
Adopted by all major financial institutions – cornerstone of day-to-
day market risk measurement
Usage of VaR as the risk measurement:
Across different markets and products
Across different time periods
VAR CAVEATS
VaR is a useful summary measure of risk, subject to some caveats:
VaR does not describe the worst loss – We would expect VaR figure to be
exceeded with a frequency of p e.g. 5 days out of a 100 for a 95%
confidence level.
VaR does not describe the losses in the left tail – VaR does not say
anything about the distribution of losses in its left tail. It just indicates
the probability of such value occurring. For the same VaR number we can
have very different distribution shapes.
VaR is measured with some error – The VaR number itself is subject to
normal sampling variation. Different statistical methodologies or
simplifications can also lead to different VaR numbers.
VAR PARAMETERS
Confidence Level
Horizon
Remember VaR measures the potential loss in value of an asset
over a defined period for a given confidence interval. Any
meaningful description of Value at Risk should comprise the above
including the VaR amount itself.
STEPS IN COMPUTING VAR
Establish a return distribution. In this step, based on some
combination of real conviction about the actual distribution and mere
convenience, we stipulate a distributional hypothesis, that is, we state
what family of distributions the returns are drawn from. We can then
estimate the parameters of the distribution. Or we may instead let the
data speak more directly for themselves and rely on the quantiles of
the returns, without imposing a distributional hypothesis. Either way, a
judgement call is required. Nothing in theory or the data will tell the risk
manager unambiguously which approach is better.
Choose the horizon and confidence level. This choice will be based
both on the type of market participant using the VaR, and on limitations
of data and models.
STEPS IN COMPUTING VAR
Mapping. This step relates positions to risk factors. It is as
important as the distributional hypothesis to the accuracy of the
results. To actually compute a VaR, we need return data. For any
portfolio, we have to decide what data best represent its future
returns. Sometimes this is obvious, for example, a single position in
a foreign currency. But most often it is not obvious, for reasons
ranging from paucity of data to the complexity of the portfolio, and
choices must be made.
Compute the VaR. The last step, of course, is to put data and
models together and compute the results. Many implementation
details that have a material impact on the VaR estimates are
decided in this stage.
APPLICATION OF VAR –
BASEL RULES
The Basel market risk charge requires VaR to be computed with the
following parameters
A horizon of 10 trading days, or two calendar weeks
A 99% confidence interval
An observation period based on at least a year of historical data
and updated at least once per quarter.
VAR METHODS
I. Historical Simulation (Simulation Approach)
Uses historical rates and revalues positions
Used for linear & non-linear instruments
II. Parametric (Delta Normal Analytic Approach)
Constructs Variance Covariance Matrix
Used for traditional assets and linear derivatives
III. Monte Carlo Simulation (Simulation Approach)
Simulates random scenarios, revalues positions
Used for linear & non-linear instruments
LINEAR AND FULL
VALUATION METHODS
ESTIMATING RETURNS
HISTORICAL SIMULATION
APPROACH
By far the most simplest and most straightforward VaR method
Steps – order return observations from largest to the smallest
The observation that follows the threshold loss level denotes the
VaR limit.
It is essentially the observation that separates the tail from the
body of the distribution.
Observations that denotes that VaR for n observations at the 1 – α
confidence level would be ((1-α)*n) + 1
HISTORICAL SIMULATION
APPROACH
Example:
Assume you have 1000 monthly returns for a security. You decide
that you want to compute the monthly VaR for this security at a
confidence level of 95%. At a 95% confidence level the lower tail
displays the lowest 5% of the underlying distribution’s returns. See
next slide
For this distribution, the value associated with a 95% confidence
level is a return of -15.5%. If you have $1,000,000 invested in this
security, the one-month VaR is $155,000 (-15.5%*1,000,000)
EXAMPLE – HISTORICAL
SIMULATION
EXAMPLE – HISTORICAL
SIMULATION
BOOTSTRAP HISTORICAL
SIMULATION
HISTORICAL SIMULATION
APPROACHES
ADVANTAGES – HISTORICAL
SIMULATION
DISADVANTAGES –
HISTORICAL SIMULATION
DELTA NORMAL METHOD
The parametric approach (delta normal approach) explicitly
assumes a distribution for the underlying observations.
Two cases to be analysed: - (i) Normal Distribution VaR (ii)
Lognormal Distribution VaR
NORMAL VAR
VaR for a given confidence level denotes the point that separates
the tail losses from the remaining distribution. The VaR cut-off will
be in the left tail of the returns distribution.
VaR (α%) = - μP/L + δP/L*zα
NORMAL VAR - EXAMPLE
VAR – ARITHMETIC RETURNS
LOGNORMAL VAR
EXAMPLE –LOGNORMAL VAR
MERITS AND DEMERITS –
DELTA NORMAL METHOD
MONTE CARLO SIMULATION
METHOD
MONTE CARLO SIMULATION
METHOD
MONTE CARLO SIMULATION
METHOD
MONTE CARLO SIMULATION
METHOD
MONTE CARLO SIMULATION
METHOD
MONTE CARLO SIMULATION
METHOD
MONTE CARLO SIMULATION
METHOD
COMPUTING VAR USING
MONTE CARLO METHOD
COMPUTING VAR USING
MONTE CARLO METHOD
COMPUTING VAR USING
MONTE CARLO METHOD
VAR CONVERSIONS
VAR CONVERSIONS
VAR CONVERSIONS
EXPECTED SHORTFALL
EXPECTED SHORTFALL
EXPECTED SHORTFALL