Chapter 1: Estimating Risk Market Measure
Using arithmetic
1.1 DATA: returns, we implicitly
assume that the interim payment Dt does not
Our data can come in various forms: earn any return of its own.
Profit/Loss Data Geometric Return Data
The P/L generated by an asset (or portfolio) over
the period t, P / Lt, can be defined as the value
of the asset (or portfolio) at the end of t plus any
interim payments Dt minus the asset value at
The geometric return implicitly assumes that
the end of t - 1:
interim payments are continuously reinvested.
P/Lt = Pt + Dt - Pt-1
Positive values indicate profits and negative
values indicate losses. The geometric return is often more
economically meaningful than the arithmetic
To be strictly correct, we should take account of
return because it ensures that the asset price
the time value of money:
(or portfolio value) can never become negative
Present value (P/Lt) = (Pt + Dt )/(1 + d) - Pt-1 regardless of how negative the returns might be.
where d is the discount rate and we assume for The geometric return is also more convenient.
convenience that Dt is paid at the end of t.
The difference between the two returns is
Forward Value (P / Lt) = Pt + Dt - (1 + d)Pt-i negligible when both returns are small, but the
difference grows as the returns get bigger -
Loss/Profit Data which is to be expected, as the geometric return
When estimating VaR and ES, dealing with data is a log function of the arithmetic return. Since
in loss/profit (L/P) form is sometimes more we would expect returns to be low over short
convenient. L/P data are a simple periods and higher over longer periods, the
transformation of P/L data: difference between the two types of return is
negligible over short periods but potentially
L/Pt = - P/Lt substantial over longer ones.
Positive values indicate losses and negative 1.2 ESTIMATING HISTORICAL SIMULATION VAR
values indicate profits.
The HS approach estimates VaR through ordered
Arithmetic Return Data loss observations.
More generally, if we have n observations, and
our confidence level is a, we would want the
(1 - a). n + 1th highest observation, and we
Arithmetic returns should not be used when we would use the commands 'Large(Loss_data,(1 -
alpha)*n + 1)' using Excel, or 'Loss_data((1 -
are concerned with long horizons.
alpha)*n + 1)' using MATLAB.
In practice, it is often helpful to obtain HS VaR Estimating VaR with Normally Distributed
estimates from a cumulative histogram, or Arithmetic Returns:
empirical cumulative frequency function
cumulative frequency function:
This is a plot of the ordered loss observations
against their empirical cumulative frequency
(e.g., so if there are n observations in total, the Estimating Lognormal VaR
empirical cumulative frequency of the ith such A random variate X is said to be log normally
ordered observation is i/n). distributed if the natural log of X is normally
The empirical frequency function makes it very distributed.
easy to obtain the VaR: we simply move up the
cumulative frequency axis to where the for x > 0. Thus, the lognormal pdf is only defined
cumulative frequency equals our confidence for positive values of x and is skewed to the
level, draw a horizontal line along to the curve,
and then draw a vertical line down to the x-axis,
which gives us our VaR.
1.3 ESTIMATING PARAMETRIC VAR
We can also think of parametric approaches as
fitting curves through the data and then reading
off the VaR from the fitted curve. In making use
of a parametric approach, we therefore need to
take account of both the statistical distribution
and the type of data to which it applies.
Estimating VaR with Normally Distributed
Profits/Losses
right.
Normally distributed geometric returns imply
that the VaR is lognormally distributed.
This illustrates that normal and lognormal Va Rs
are much the same if we are dealing with short
holding periods and realistic return parameters.
1.4 1.4 ESTIMATING COHERENT RISK
where mP /L and sP /L are estimates of the mean MEASURES
and standard deviation of P/L.
Estimating Expected Shortfall
If the data are in P/L form, the VaR is indicated
by the negative of the cut-off point between the The fact that the ES is a probability-weighted
lower 5% and the upper 95% of P/L average of tail losses suggests that we can
observations. It is the contrast for for L/P. estimate ES as an average of 'tail VaRs'. 4 The
easiest way to implement this approach is to
slice the tail into a large number n of slices, each
of which has the same probability mass,
estimate the VaR associated with each slice, and
take the ES as the average of these Va Rs.
we would want a value of n large enough to give
accurate results.
Estimating Coherent Risk Measures
Recall that a coherent risk measure is a
weighted average of the quantiles (denoted by
qp ) of our loss distribution:
So the ES is the f(p) wheighted average of the
VaRs. The estimate does eventually con verge to
We first consider the standard errors of quantile
(or VaR) estimators. ( go back to page eleven)
Standard Errors in Estimators of Coherent Risk
Measures
the true value as n gets large.
the presence of heavy tails might make ES
1.5 1.5 ESTIMATING THE STANDARD ERRORS
estimators in general less accurate than VaR
OF RISK MEASURE ESTIMATORS
estimators. Go back to page 12
should always seek to supplement any risk
estimates we produce with some indicator of
their precision. This is a fundamental principle
of good risk measurement practice. In this part
we focus on the more basic indicator, the
standard error of a risk measure estimator.
Standard Errors of Quantile Estimators